Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
This action is in response to the amendment and remarks filed April 21st, 2026. In the amendment, claims 1-2, 5-6, 9, 14-15, and 17 were amended and claims 18-20 were added. As such, claims 1-20 are pending.
Response to Arguments
Applicant’s arguments, see Pages 10-11, filed April 21st, 2026, with respect to the claim objections and the 112 (f) invocations, and 112 (a) rejections have been fully considered and are persuasive. The objected claims, the 112 (f) invocations, and 112 (a) rejections of claims 15-16 have been withdrawn.
Applicant’s arguments, see Page 10, filed March 31st 2026, with respect to the 35 U.S.C 112 (b)
rejections have been fully considered and are persuasive. Amendments to the claims obviate the
rejections of record. The rejections of claims have been withdrawn. See updated rejections below.
Applicant’s arguments with respect to the rejections of claims 1-17 under 35 U.S.C § 101 and 103 are
not persuasive for the following reasons:
35 U.S.C. 101:
Applicant argues that under Enfish, LLC v. Microsoft Corp. (Fed. Cir. 2016) the claims are directed to an improvement in machine learning technology, specifically the training of the model itself, and are not directed to an abstract idea (Pages 12-13 of Remarks). The Examiner respectfully disagrees. Applicant's reliance on Enfish is not persuasive because the Federal Circuit found those claims eligible because their focus was a specific improvement to computer functionality itself which was the self-referential database table. Here, the claims use a generic neural network and generic computing components as a tool to carry out the abstract idea of evaluating trajectory data and computing parameter differences, which is different than Enfish. The recited steps do not improve how the computer itself operates in the manner of a new data structure, chip architecture, memory system, or the model itself but instead apply known model training to a particular type of data. Enfish is therefore distinguishable and does not render the claims eligible.
Applicant argues that amended claim 1 no longer recites the verbs "acquiring," "obtaining," and "adjusting" relied upon in the rejection making the Step 2A Prong One rejection moot (Page 13 of Remarks). The Examiner respectfully disagrees. The mental process grouping is evaluated based on the substance of the recited steps and not the particular verb chosen. Amended claim 1 still recites limitations such as "acquiring at least one trajectory point... comprising... a place passed by the each user, a start time of reaching the place and a duration of staying in the place" and "running a search operation on the generated trajectory representation... by using the start time and the duration," which involve observation, evaluation, and judgment that can be performed in the human mind or with pen and paper. Relabeling "acquiring" as "extracting" or "adjusting" as "self-adjusting" does not remove the mental process or mathematical concept from the claim. The Step 2A Prong One analysis therefore remains applicable as set forth in the rejection.
Applicant argues that the "data triplet" of place, start time, and duration compresses raw trajectory data into a compact, semantically meaningful format such that "the modeling efficiency can be improved and the semantic representation can be enhanced" (Paragraphs 18 and 31 of instant specification) stating that it is an improvement to computer technology (Page 13 of Remarks). The Examiner respectfully disagrees. Paragraphs 18 and 31 of the instant specification merely provides a technical explanation of how the improvement is achieved. These paragraphs only state that efficiency and semantic representation "can be improved" without explaining how the triplet structure produces that result at a technical level. Moreover, selecting and organizing data into a place, start time, and duration triplet is itself part of the abstract idea and not a technical improvement to the computer. The claim therefore does not reflect an improvement to technology under MPEP 2106.04(d).
Applicant argues that the search operation and difference-based parameter adjustment are a specific self-supervised machine learning mechanism in which the model learns relationships, and thus cannot be a mental step (Page 13 of Remarks). The Examiner respectfully disagrees. Characterizing a step as "machine learning" does not remove it from the abstract idea groupings because "running a search operation... by using the start time and the duration" to "output... a position" is a mental process, and the recited "difference between the place passed by the each user and the position" is a mathematical concept. The recited "self-adjusting a network parameter of the pre-trained model" based on that difference amounts to mere instructions to apply the abstract idea on a computer under MPEP 2106.05(f) and imposes no meaningful limit on the exception. The claim recites a result to be achieved rather than any specific, non-generic technique for performing the search or the adjustment. These limitations therefore remain directed to the abstract idea and do not provide an inventive concept.
Applicant argues that, even if the claims recite an abstract idea, the additional elements integrate it into a practical application because the structured triplet enables more efficient training and the trained model can be directly used for downstream analytics such as next location prediction, anomaly detection, and path classification (Page 14 of Remarks). The Examiner respectfully disagrees. The additional elements of "inputting the data triplet... into a neural network," using a "pre-trained model," and outputting a result are recited at a high level of generality and amount to mere data gathering, applying the abstract ideas with a computer, and necessary data output under MPEP 2106.05(f) and (g). The claim does not recite how the downstream prediction, detection, or classification is technically accomplished, and therefore does not go beyond generally linking the abstract idea to a technological environment. Therefore, the practical application argument is not persuasive.
Applicant argues that the combination of limitations is unconventional because the cited prior art does not teach the data triplet with duration, the search operation using time and duration, and the difference-based adjustment, and that this unconventional combination renders the claims allowable under Step 2B (Page 14 of Remarks). The Examiner respectfully disagrees. Whether a claim is novel or nonobvious under 35 U.S.C. 102 or 103 is a separate inquiry from subject matter eligibility, and a claim may recite a novel abstract idea yet remain ineligible under 35 U.S.C. 101. The additional elements under Step 2B are the generic neural network, processor, and memory functions which are well-understood, routine, and conventional per MPEP 2106.05(d) and are not significantly more than the abstract idea.
Applicant argues that claim 9 is eligible because it recites a Transformer neural network with a self-attention mechanism, a search operation implemented by a fully connected layer, and an output of a predicted next location, path category, anomaly indicator, or schedule (Page 14 of Remarks). The Examiner respectfully disagrees. The recited “Transformer neural network having a self-attention mechanism” and “fully connected layer” name generic, well-known components used to apply the abstract idea, which amounts to mere instructions to implement the idea on a computer under MPEP 2106.05(f), and the specification does not explain how these components improve the functioning of the computer itself as required by MPEP 2106.05(a). The recited output of a predicted next location, category, anomaly indicator, or schedule is the result of the classification and amounts to necessary data output under MPEP 2106.05(g)(3). Claim 9 therefore remains directed to an abstract idea without significantly more.
Applicant argues that new claims 19 and 20 recite a multi-level model with a particular technical improvement, citing specification paragraph 51 that the structure reduces computation and space by about 100 to 200 times for long sequences, qualifying as an improvement under Step 2B (Pages 13-14 of Remarks). The Examiner respectfully disagrees. The recited “dividing the inputted sequence into segments… at intervals of days” and “constructing the representations into a sequence” are abstract steps of partitioning and organizing data, while “inputting the segments into the pre-trained model” and “outputting the representation” are mere data gathering and necessary data output under MPEP 2106.05(g). Although paragraph 51 recites a quantitative reduction, claims 19 and 20 do not recite the specific technical mechanism by which it is achieved, and the claim does not reflect the disclosed improvement. Claims 19 and 20 therefore fail to integrate the exception into a practical application and do not amount to significantly more.
35 U.S.C. 103:
Applicant argues that a statement in Zhou that duration "can be encoded in a similar way" is a mere suggestion or possibility and not an enabling disclosure, and that Zhou nowhere defines a trajectory point as including a duration of stay (Pages 15 to 16 of Remarks). The Examiner respectfully disagrees. The Examiner notes that this argument is moot in view of the new grounds of rejection. As set forth in the new grounds, Li expressly discloses that the attribute information of the communication connection includes the start time of the communication connection and the duration of the communication connection, together with the identification of the AP, which together correspond to the claimed place, start time, and duration triplet.
Applicant argues that Huberman is non analogous prior art directed to a different technical field (autonomous vehicle navigation), and that one of ordinary skill would not look to a real time autonomous vehicle control system to modify Zhou's offline tour recommendation system. Applicant further argues that Huberman fails to teach using a start time and a duration as a query to retrieve a position from a trajectory representation (Pages 16-17 of Remarks). The Examiner respectfully disagrees, and notes that this argument is moot. The new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 9-16 and 19-20 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claim 9 recites the limitation "to obtain the path representation model, wherein the search task is performed by a fully connected layer" in third to last line of the claim. There is insufficient antecedent basis for this limitation in the claim.
Claims 19 and 20 recite the limitation "dividing the inputted sequence into segments according to the periodicity… inputting the constructed sequence into a sequence model, and outputting the representation of the whole sequence" throughout the claims. There is insufficient antecedent basis for this limitation in the claims.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an
abstract idea without significantly more.
Claim 1
Step 1: The claim recites a method; therefore, it is directed to the statutory category of a process.
Step2A Prong 1: The claim recites, inter alia:
acquiring at least one trajectory point of at least one user, wherein each trajectory point of each user comprises a data triplet comprising a place passed by the each user, a start time of reaching the place and a duration of staying in the place: This limitation encompasses a mental process because it involves the evaluation/judgement/opinion to capture a user’s trajectory point by observing their location, start time, and duration which can be performed mentally or by pen and paper.
outputting, for each user, a position of each trajectory point from the generated trajectory representation of the user comprising running a search operation on the generated trajectory representation, the search operation performed by using the start time and the duration of each trajectory point of the each user: This limitation encompasses a mental process because it involves the evaluation/judgement/opinion to search the generated trajectory representation using the start time and duration of each trajectory point, and produce a position for each corresponding trajectory point which can be performed mentally or by pen and paper.
and generating a trajectory representation of each user: This limitation is a mental process because it involves generating a trajectory representation, which can be performed by pen and paper.
Step2A Prong 2: This judicial exception is not integrated into a practical application because the
additional elements are as follows:
inputting the data triplet of the at least one trajectory point of the at least one user into a neural network comprising a pre-trained model: Data Gather- Mere data gathering recited at a high level of generality, and thus is an insignificant extra-solution activity (MPEP 2106.05(g)).
training a path representation model, comprising…: Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
wherein the data triplet is extracted from a log of a wireless access point which is configured to record access by a mobile communication device of a user: The limitation merely describes the type of data being processed and thus amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)).
and in a machine learning process, self-adjusting a network parameter of the pre-trained model according to a difference between the place passed by the each user and the position of the each trajectory point obtained from running the search operation, to obtain the path representation model: Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
Step 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because the additional elements are as follows:
inputting the data triplet of the at least one trajectory point of the at least one user into a neural network comprising a pre-trained model: The additional element of “inputting” does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of receiving steps amounts to no more than mere data gathering. This element amounts to receiving data over a network and are well-understood, routine, conventional activity. See MPEP 2106.05(d), subsection II (i). This cannot provide an inventive concept.
training a path representation model, comprising…: Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
wherein the data triplet is extracted from a log of a wireless access point which is configured to record access by a mobile communication device of a user: The limitation merely describes the type of data being processed and thus amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself which cannot provide inventive concept (MPEP 2106.05(h)).
and in a machine learning process, self-adjusting a network parameter of the pre-trained model according to a difference between the place passed by the each user and the position of the each trajectory point obtained from running the search operation, to obtain the path representation model: Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea and cannot provide inventive concept (MPEP 2106.05(f)).
The elements in combination as an ordered whole still do not amount to significantly more than the judicial exception (i.e., the abstract ideas of obtaining trajectory data and mathematical concepts for calculating parameter differences). The claim merely describes a process of applying known data processing techniques (inputting trajectory points and searching by the start time and duration) and standard computing functions (training a model by self-adjusting parameters based on the calculated difference).
Therefore, the claim as a whole remains focused on the abstract idea and fails Step 2B of the
eligibility analysis.
Claim 2
Step 1: A process, as above.
Step2A Prong 1: The claim recites, inter alia:
[A]cquiring a sample set, a sample in the sample set comprising a sample trajectory and a tag: This limitation is a mental process because it involves acquiring data samples containing a trajectory and a tag.
Step2A Prong 2: This judicial exception is not integrated into a practical application because the
additional elements are as follows:
and using respectively the sample trajectory and the tag in the sample set as an input and an expected output of the path representation model…: Data Gather- Mere data gathering recited at a high level of generality, and thus is an insignificant extra-solution activity (MPEP 2106.05(g)).
to perform self-supervised training on the path representation model: Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
Step 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because the additional elements are as follows:
and using respectively the sample trajectory and the tag in the sample set as an input and an expected output of the path representation model…: The additional element of “using” does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of receiving steps amounts to no more than mere data gathering. This element amounts to receiving data over a network and are well-understood, routine, conventional activity. See MPEP 2106.05(d), subsection II (i). This cannot provide an inventive concept.
to perform self-supervised training on the path representation model: Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea and cannot provide inventive concept (MPEP 2106.05(f)).
Even when considered in combination, these additional elements represent mere instructions to apply
an exception and therefore do not provide an inventive concept. The claim is ineligible.
Claim 3
Step 1: A process, as above.
Step2A Prong 1: The claim recites, inter alia:
dividing, for a target sample trajectory with a total duration exceeding a predetermined value in the sample set, the target sample trajectory into at least one segment according to a predetermined time interval: This limitation is viewed as a mental process as it involves partitioning a dataset based on determining whether certain values exceed different thresholds.
and constructing, for each target sample trajectory, the representation of each segment into a sequence of representations of the target sample trajectory: This limitation is seen as a mental process as it involves organizing the representation of each segment into a sequence of representations.
Step2A Prong 2: This judicial exception is not integrated into a practical application because the
additional elements are as follows:
inputting, for each target sample trajectory, at least one segment of the target sample trajectory into…: Data Gather- Mere data gathering recited at a high level of generality, and thus is an insignificant extra-solution activity (MPEP 2106.05(g)).
…the path representation model to obtain a representation of each segment of the target sample trajectory: Insignificant extra-solution as the limitation amounts to necessary data outputting (MPEP 2106.05(g)(3)).
…and inputting the sequence and a time identifier corresponding to each segment into a sequence model…: Mere data gathering recited at a high level of generality, and thus is an insignificant extra-solution activity (MPEP 2106.05(g)).
…to output a sequence representation of the target sample trajectory: Insignificant extra-solution as the limitation amounts to necessary data outputting (MPEP 2106.05(g)(3)).
Step 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because the additional elements are as follows:
inputting, for each target sample trajectory, at least one segment of the target sample trajectory into…: The additional element of “inputting” does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of receiving steps amounts to no more than mere data gathering. This element amounts to receiving data over a network and are well-understood, routine, conventional activity. See MPEP 2106.05(d), subsection II (i). This cannot provide an inventive concept.
…the path representation model to obtain a representation of each segment of the target sample trajectory: Insignificant extra-solution as the limitation amounts to necessary data outputting (MPEP 2106.05(g)(3)). This falls under Well-Understood, Routine, Conventional activity -see MPEP 2106.05(d)(II)(vi).
…and inputting the sequence and a time identifier corresponding to each segment into a sequence model…: The additional element of “inputting” does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of receiving steps amounts to no more than mere data gathering. This element amounts to receiving data over a network and are well-understood, routine, conventional activity. See MPEP 2106.05(d), subsection II (i). This cannot provide an inventive concept.
…to output a sequence representation of the target sample trajectory: Insignificant extra-solution as the limitation amounts to necessary data outputting (MPEP 2106.05(g)(3)). This falls under Well-Understood, Routine, Conventional activity -see MPEP 2106.05(d)(II)(vi).
Even when considered in combination, these additional elements represent mere instructions to apply an exception and therefore do not provide an inventive concept. The claim is ineligible.
Claim 4
Step 1: A process, as above.
Step2A Prong 1: The claim recites, inter alia:
and adjusting a network parameter…: This limitation is seen as a mental process since it deals with adjusting a parameter and not changing the model itself.
…of the sequence model according to a difference between the prediction result of the each target sample trajectory and a tag corresponding to the each target sample trajectory: This limitation is viewed as a mathematical concept as it deals with calculating the loss value which calculates the difference between two numbers.
Step2A Prong 2: This judicial exception is not integrated into a practical application because the
additional elements are as follows:
outputting the sequence representation of each target sample trajectory by a prediction model, to obtain a prediction result of the each target sample trajectory: Insignificant extra-solution as the limitation amounts to necessary data outputting (MPEP 2106.05(g)(3)).
Step 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because the additional elements are as follows:
outputting the sequence representation of each target sample trajectory by a prediction model, to obtain a prediction result of the each target sample trajectory: Insignificant extra-solution as the limitation amounts to necessary data outputting (MPEP 2106.05(g)(3)). This falls under Well-Understood, Routine, Conventional activity -see MPEP 2106.05(d)(II)(vi).
Even when considered in combination, these additional elements represent mere instructions to apply an exception and therefore do not provide an inventive concept. The claim is ineligible.
Claim 5
Step 1: A process, as above.
Step2A Prong 1: The claim recites, inter alia:
the tag comprises at least one of: a path category tag, an abnormal event tag, or a schedule tag: This limitation is viewed as a mental process because it involves the mental classification of data into specific categories (e.g., path, event, schedule).
Step 2A Prong Two and Step 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception under step 2B. Thus, the judicial exception is not integrated into a practical application (see MPEP 2106.04(d) I.), failing step 2A prong 2. The claim is ineligible.
Even when considered in combination, these additional elements represent mere instructions to apply an exception and therefore do not provide an inventive concept. The claim is ineligible.
Claim 6
Step 1: A process, as above.
Step2A Prong 1: The claim recites, inter alia:
masking, according to a masking rule, places passed by the user in a part of the at least one trajectory point of the at least one user, to obtain at least one masked trajectory point: This limitation is viewed as a mental process of masking data and obtaining that masked data.
and adjusting a network parameter…: This limitation is seen as a mental process since it deals with adjusting a parameter and not changing the model itself.
…of the pre-trained model according to a difference between the mask position and a position of the place masked according to the masking rule…: This limitation is viewed as a mathematical concept as it deals with calculating the difference between two numbers.
Step2A Prong 2: This judicial exception is not integrated into a practical application because the
additional elements are as follows:
inputting the at least one masked trajectory point into the…: Data Gather- Mere data gathering recited at a high level of generality, and thus is an insignificant extra-solution activity (MPEP 2106.05(g)).
…pre-trained model to obtain a mask position: Insignificant extra-solution as the limitation amounts to necessary data outputting (MPEP 2106.05(g)(3)).
…to obtain the path representation model: Insignificant extra-solution as the limitation amounts to necessary data outputting (MPEP 2106.05(g)(3)).
Step 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because the additional elements are as follows:
[I]nputting the at least one masked trajectory point into the…: The additional element of “inputting” does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of receiving steps amounts to no more than mere data gathering. This element amounts to receiving data over a network and are well-understood, routine, conventional activity. See MPEP 2106.05(d), subsection II (i). This cannot provide an inventive concept.
…pre-trained model to obtain a mask position: Insignificant extra-solution as the limitation amounts to necessary data outputting (MPEP 2106.05(g)(3)). This falls under Well-Understood, Routine, Conventional activity -see MPEP 2106.05(d)(II)(vi).
…to obtain the path representation model: Insignificant extra-solution as the limitation amounts to necessary data outputting (MPEP 2106.05(g)(3)). This falls under Well-Understood, Routine, Conventional activity -see MPEP 2106.05(d)(II)(vi).
Even when considered in combination, these additional elements represent mere instructions to apply an exception and therefore do not provide an inventive concept. The claim is ineligible.
Claim 7
Step 1: A process, as above.
Step2A Prong 1: The claim recites, inter alia:
acquiring to-be-analyzed user trajectory information: This limitation encompasses a mental process dealing with acquiring trajectory data.
Step2A Prong 2: This judicial exception is not integrated into a practical application because the
additional elements are as follows:
inputting the user trajectory information into the… and inputting the path representation into a…: Data Gather- Mere data gathering recited at a high level of generality, and thus is an insignificant extra-solution activity (MPEP 2106.05(g)).
…path representation model to output a path representation… prediction model to output a prediction result: Insignificant extra-solution as the limitation amounts to necessary data outputting (MPEP 2106.05(g)(3)).
Step 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because the additional elements are as follows:
inputting the user trajectory information into the… and inputting the path representation into a…: The additional element of “inputting” does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of receiving steps amounts to no more than mere data gathering. This element amounts to receiving data over a network and are well-understood, routine, conventional activity. See MPEP 2106.05(d), subsection II (i). This cannot provide an inventive concept.
…path representation model to output a path representation… prediction model to output a prediction result: Insignificant extra-solution as the limitation amounts to necessary data outputting (MPEP 2106.05(g)(3)). This falls under Well-Understood, Routine, Conventional activity -see MPEP 2106.05(d)(II)(vi).
Even when considered in combination, these additional elements represent mere instructions to apply an exception and therefore do not provide an inventive concept. The claim is ineligible.
Claim 8
Step 1: A process, as above.
Step2A Prong 1: The claim recites, inter alia:
the prediction result comprises at least one of: a path category, an abnormal event, a next position, or a schedule: This limitation is viewed as a mental process because it involves the mental classification of the result into specific categories (e.g., path, abnormal event, next position, schedule).
Step 2A Prong Two and Step 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception under step 2B. Thus, the judicial exception is not integrated into a practical application (see MPEP 2106.04(d) I.), failing step 2A prong 2. The claim is ineligible.
Even when considered in combination, these additional elements represent mere instructions to apply an exception and therefore do not provide an inventive concept. The claim is ineligible.
Claim 18
Step 1: A process, as above.
Step2A Prong 1: This claim does not recite an additional abstract idea, but the claim depends on
claim 3 which recites an abstract idea.
Step2A Prong 2: This judicial exception is not integrated into a practical application because the
additional elements are as follows:
the pre-trained model comprises a Transformer neural network comprising a self-attention mechanism: Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
Step 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because the additional elements are as follows:
the pre-trained model comprises a Transformer neural network comprising a self-attention mechanism: Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea and cannot provide inventive concept (MPEP 2106.05(f)).
Therefore, the claim as a whole remains focused on the abstract idea and fails Step 2B of the eligibility analysis.
Claim 19
Step 1: A process, as above.
Step2A Prong 1: The claim recites, inter alia:
dividing the inputted sequence into segments according to the periodicity, comprising dividing into segments at intervals of days: This limitation is a mental process as it involves the evaluation/judgement/opinion of dividing a target sample trajectory into segment(s) based on a duration exceeding a threshold, which can be performed mentally and/or by pen and paper.
constructing the representations into a sequence: This limitation is a mental process as it involves the evaluation/judgement/opinion of creating a representation of each segment into a sequence of representations, which can be performed mentally and/or by pen and paper.
Step2A Prong 2: This judicial exception is not integrated into a practical application because the
additional elements are as follows:
inputting the segments into the pre-trained model to obtain the representation of each segment… inputting the constructed sequence into a sequence model: Data Gather- Mere data gathering recited at a high level of generality, and thus is an insignificant extra-solution activity (MPEP 2106.05(g)).
and outputting the representation of the whole sequence: Insignificant extra-solution as the limitation amounts to necessary data outputting (MPEP 2106.05(g)(3)).
Step 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because the additional elements are as follows:
inputting the segments into the pre-trained model to obtain the representation of each segment… inputting the constructed sequence into a sequence model: The additional element of “inputting” does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of receiving steps amounts to no more than mere data gathering. This element amounts to receiving data over a network and are well-understood, routine, conventional activity. See MPEP 2106.05(d), subsection II (i). This cannot provide an inventive concept.
and outputting the representation of the whole sequence: Insignificant extra-solution as the limitation amounts to necessary data outputting (MPEP 2106.05(g)(3)). This falls under Well-Understood, Routine, Conventional activity -see MPEP 2106.05(d)(II)(vi).
Therefore, the claim as a whole remains focused on the abstract idea and fails Step 2B of the eligibility analysis.
Claim 20 recites similar limitations to claim 19. Therefore, claim 20 is rejected using the same rationale as claim 19.
Claim 9
Step 1: The claim recites an apparatus; therefore, it is directed to the statutory category of machine.
Step2A Prong 1: The claim recites, inter alia:
acquiring at least one trajectory point of at least one user, wherein each trajectory point of each user comprises a data triplet comprising a place passed by the each user, a start time of reaching the place and a duration of staying in the place: This limitation encompasses a mental process because it involves the evaluation/judgement/opinion to capture a user’s trajectory point by observing their location, start time, and duration which can be performed mentally or by pen and paper.
outputting, for each user, a position of each trajectory point from the generated trajectory representation of the user comprising running a search operation on the generated trajectory representation, the search operation performed by inputting the start time and the duration of each trajectory point of the each user: This limitation encompasses a mental process because it involves the evaluation/judgement/opinion to search the generated trajectory representation using the start time and duration of each trajectory point, and produce a position for each corresponding trajectory point which can be performed mentally or by pen and paper.
and generating a trajectory representation of each user: This limitation is a mental process because it involves generating a trajectory representation, which can be performed by pen and paper.
Step2A Prong 2: This judicial exception is not integrated into a practical application because the
additional elements are as follows:
An apparatus for training a path representation model, comprising: at least one processor; and a storage device, wherein the storage device stores instructions executable by the at least one processor, and the instructions when executed by the at least one processor cause the at least one processor to perform operations comprising: Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
wherein the data triplet is extracted from a log of a wireless access point which is configured to record access by a mobile communication device of a user: The limitation merely describes the type of data being processed and thus amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)).
inputting the data triplet of the at least one trajectory point of the at least one user into a neural network comprising a pre-trained model: Data Gather- Mere data gathering recited at a high level of generality, and thus is an insignificant extra-solution activity (MPEP 2106.05(g)).
wherein the pre-trained model comprises a Transformer neural network having a self-attention mechanism: Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
and in a machine learning process, self-adjusting a network parameter of the pre-trained model according to a difference between the place passed by the each user and the position of the each trajectory point obtained from running the search operation, to obtain the path representation model: Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
wherein the search task is performed by a fully connected layer: Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
and outputting, by the path representation model output at least one of: a predicted next location of the user, a category of a path taken by the user, an anomaly indicator associated with a path taken by the user, or a schedule for the user: Insignificant extra-solution as the limitation amounts to necessary data outputting (MPEP 2106.05(g)(3)).
Step 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because the additional elements are as follows:
An apparatus for training a path representation model, comprising: at least one processor; and a storage device, wherein the storage device stores instructions executable by the at least one processor, and the instructions when executed by the at least one processor cause the at least one processor to perform operations comprising: Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea and cannot provide inventive concept (MPEP 2106.05(f)).
wherein the data triplet is extracted from a log of a wireless access point which is configured to record access by a mobile communication device of a user: The limitation merely describes the type of data being processed and thus amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself which cannot provide inventive concept (MPEP 2106.05(h)).
inputting the data triplet of the at least one trajectory point of the at least one user into a neural network comprising a pre-trained model: The additional element of “inputting” does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of receiving steps amounts to no more than mere data gathering. This element amounts to receiving data over a network and are well-understood, routine, conventional activity. See MPEP 2106.05(d), subsection II (i). This cannot provide an inventive concept.
wherein the pre-trained model comprises a Transformer neural network having a self-attention mechanism: Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea and cannot provide inventive concept (MPEP 2106.05(f)).
and in a machine learning process, self-adjusting a network parameter of the pre-trained model according to a difference between the place passed by the each user and the position of the each trajectory point obtained from running the search operation, to obtain the path representation model: Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea and cannot provide inventive concept (MPEP 2106.05(f)).
wherein the search task is performed by a fully connected layer: Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
and outputting, by the path representation model output at least one of: a predicted next location of the user, a category of a path taken by the user, an anomaly indicator associated with a path taken by the user, or a schedule for the user: Insignificant extra-solution as the limitation amounts to necessary data outputting (MPEP 2106.05(g)(3)). This falls under Well-Understood, Routine, Conventional activity -see MPEP 2106.05(d)(II)(vi).
The elements in combination as an ordered whole still do not amount to significantly more than the judicial exception (i.e., the abstract ideas of obtaining trajectory data and mathematical concepts for calculating parameter differences). The claim merely describes a process of applying known data processing techniques (inputting trajectory points and searching by the start time and duration) and standard computing functions (training a model by adjusting parameters) using a generic computer.
Therefore, the claim as a whole remains focused on the abstract idea and fails Step 2B of the eligibility analysis.
Claim 10
Step 1: A machine, as above.
Step2A Prong 1: The claim recites, inter alia:
[A]cquiring a sample set, a sample in the sample set comprising a sample trajectory and a tag: This limitation is viewed as a mental process of acquiring data samples containing a trajectory and a tag.
Step2A Prong 2: This judicial exception is not integrated into a practical application because the
additional elements are as follows:
[A]nd using respectively the sample trajectory and the tag in the sample set as an input and an expected output of the path representation model…: Data Gather- Mere data gathering recited at a high level of generality, and thus is an insignificant extra-solution activity (MPEP 2106.05(g)).
to perform supervised training on the path representation model: Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
Step 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because the additional elements are as follows:
[A]nd using respectively the sample trajectory and the tag in the sample set as an input and an expected output of the path representation model…: The additional element of “receiving” does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of receiving steps amounts to no more than mere data gathering. This element amounts to receiving data over a network and are well-understood, routine, conventional activity. See MPEP 2106.05(d), subsection II (i). This cannot provide an inventive concept.
to perform supervised training on the path representation model: Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea and cannot provide inventive concept (MPEP 2106.05(f)).
Even when considered in combination, these additional elements represent mere instructions to apply
an exception and therefore do not provide an inventive concept. The claim is ineligible.
Claim 11 recites similar limitations to claim 3. Therefore, claim 11 is rejected using the same rationale as claim 3.
Claim 12 recites similar limitations to claim 4. Therefore, claim 12 is rejected using the same rationale as claim 4.
Claim 13 recites similar limitations to claim 5. Therefore, claim 13 is rejected using the same rationale as claim 5.
Claim 14 recites similar limitations to claim 6. Therefore, claim 14 is rejected using the same rationale as claim 6.
Claim 15 recites similar limitations to claim 7. Therefore, claim 15 is rejected using the same rationale as claim 7.
Claim 16 recites similar limitations to claim 8. Therefore, claim 16 is rejected using the same rationale as claim 8.
Claim 17
Step 1: The claim recites a non-transitory computer medium; therefore, it is directed to an article of manufacture.
Step2A Prong 1: The claim recites, inter alia:
acquiring at least one trajectory point of at least one user, wherein each trajectory point of each user comprises a data triplet comprising a place passed by the each user, a start time reaching the place and a duration of staying in the place: This limitation encompasses a mental process because it involves the evaluation/judgement/opinion to capture a user’s trajectory point by observing their location, start time, and duration which can be performed mentally or by pen and paper.
outputting, for each user, a position of each trajectory point from the generated trajectory representation of the user comprising running a search operation on the generated trajectory representation, the search operation performed by inputting the start time and the duration of each trajectory point of the each user: This limitation encompasses a mental process because it involves the evaluation/judgement/opinion to search the generated trajectory representation using the start time and duration of each trajectory point, and produce a position for each corresponding trajectory point which can be performed mentally or by pen and paper.
and generating a trajectory representation of each user: This limitation is a mental process because it involves generating a trajectory representation, which can be performed by pen and paper.
the operations further comprising: acquiring a sample set, a sample in the sample set comprising a sample trajectory and a tag: This limitation is a mental process as it involves the evaluation/judgement/opinion of acquiring a sample set comprising a tag and a sample trajectory, which can be performed mentally.
dividing, for a target sample trajectory with a total duration exceeding a predetermined value in the sample set, the target sample trajectory into at least one segment according to a predetermined time interval: This limitation is a mental process as it involves the evaluation/judgement/opinion of dividing a target sample trajectory into segment(s) based on a duration exceeding a threshold, which can be performed mentally and/or by pen and paper.
…to obtain a representation of each segment of the target sample trajectory: This limitation is a mental process as it involves the evaluation/judgement/opinion of obtaining a representation of a target sample trajectory, which can be performed in the human mind.
and constructing, for each target sample trajectory, the representation of each segment into a sequence of representations of the target sample trajectory: This limitation is a mental process as it involves the evaluation/judgement/opinion of creating a representation of each segment into a sequence of representations, which can be performed mentally and/or by pen and paper.
Step2A Prong 2: This judicial exception is not integrated into a practical application because the
additional elements are as follows:
A non-transitory computer readable storage medium, storing a computer instruction, wherein the computer instruction when executed by a computer causes the computer to perform operations comprising: Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
[I]nputting the at least one trajectory point of the at least one user into a pre-trained model to obtain a trajectory representation of each user: Data Gather- Mere data gathering recited at a high level of generality, and thus is an insignificant extra-solution activity (MPEP 2106.05(g)).
wherein the data triplet is extracted from a log of a wireless access point which is configured to record access by a mobile communication device of a user: The limitation merely describes the type of data being processed and thus amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself (MPEP 2106.05(h)).
inputting the data triplet of the at least one trajectory point of the at least one user into a neural network comprising a pre-trained model: Data Gather- Mere data gathering recited at a high level of generality, and thus is an insignificant extra-solution activity (MPEP 2106.05(g)).
wherein the pre-trained model comprises a Transformer neural network comprising a self-attention mechanism: Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
and in a machine learning process, self-adjusting a network parameter of the pre-trained model according to a difference between the place passed by the each user and the position of the each trajectory point obtained from running the search operation, to obtain the path representation model: Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
to perform self-supervised training on the path representation model: Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f)).
inputting, for each target sample trajectory, at least one segment of the target sample trajectory into the path representation model…: Data Gather- Mere data gathering recited at a high level of generality, and thus is an insignificant extra-solution activity (MPEP 2106.05(g)).
to output a sequence representation of the target sample trajectory: Insignificant extra-solution as the limitation amounts to necessary data outputting (MPEP 2106.05(g)(3)).
and using respectively the sample trajectory and the tag in the sample set as an input and an expected output of the path representation model: Data Gather- Mere data gathering recited at a high level of generality, and thus is an insignificant extra-solution activity (MPEP 2106.05(g)).
Step 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because the additional elements are as follows:
A non-transitory computer readable storage medium, storing a computer instruction, wherein the computer instruction when executed by a computer causes the computer to perform operations comprising: Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea and cannot provide inventive concept (MPEP 2106.05(f)).
[I]nputting the at least one trajectory point of the at least one user into a pre-trained model to obtain a trajectory representation of each user: The additional element of “inputting” does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of receiving steps amounts to no more than mere data gathering. This element amounts to receiving data over a network and are well-understood, routine, conventional activity. See MPEP 2106.05(d), subsection II (i). This cannot provide an inventive concept.
wherein the data triplet is extracted from a log of a wireless access point which is configured to record access by a mobile communication device of a user: The limitation merely describes the type of data being processed and thus amounts to merely indicating a field of use or technological environment in which to apply a judicial exception. This does not amount to significantly more than the exception itself which cannot provide inventive concept (MPEP 2106.05(h)).
inputting the data triplet of the at least one trajectory point of the at least one user into a neural network comprising a pre-trained model: The additional element of “inputting” does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of receiving steps amounts to no more than mere data gathering. This element amounts to receiving data over a network and are well-understood, routine, conventional activity. See MPEP 2106.05(d), subsection II (i). This cannot provide an inventive concept.
wherein the pre-trained model comprises a Transformer neural network comprising a self-attention mechanism: Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea and cannot provide inventive concept (MPEP 2106.05(f)).
and in a machine learning process, self-adjusting a network parameter of the pre-trained model according to a difference between the place passed by the each user and the position of the each trajectory point obtained from running the search operation, to obtain the path representation model: Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea and cannot provide inventive concept (MPEP 2106.05(f)).
to perform self-supervised training on the path representation model: Performing self-supervised training on a model is a conventional machine-learning operation, as seen in Yang (US 20220351009 A1, filed April 15, 2021). Yang treats self-supervised training as a known, pre-existing technique rather than any new or improved method. Yang explicitly mentions in paragraph 60 that the “training process is performed with the conventional method for self-supervised learning” which confirms that this is a conventional training step and therefore does not add significantly more than the judicial exception and cannot supply an inventive concept (MPEP 2106.05(f).
inputting, for each target sample trajectory, at least one segment of the target sample trajectory into the path representation model…: The additional element of “inputting” does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of receiving steps amounts to no more than mere data gathering. This element amounts to receiving data over a network and are well-understood, routine, conventional activity. See MPEP 2106.05(d), subsection II (i). This cannot provide an inventive concept.
to output a sequence representation of the target sample trajectory: Insignificant extra-solution as the limitation amounts to necessary data outputting (MPEP 2106.05(g)(3)). This falls under Well-Understood, Routine, Conventional activity -see MPEP 2106.05(d)(II)(vi).
and using respectively the sample trajectory and the tag in the sample set as an input and an expected output of the path representation model: The additional element of “receiving” does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of receiving steps amounts to no more than mere data gathering. This element amounts to receiving data over a network and are well-understood, routine, conventional activity. See MPEP 2106.05(d), subsection II (i). This cannot provide an inventive concept.
The elements in combination as an ordered whole still do not amount to significantly more than the judicial exception (i.e., the abstract ideas of obtaining trajectory data and mathematical concepts for calculating parameter differences). The claim merely describes a process of applying known data processing techniques (inputting trajectory points and searching by the start time and duration) and standard computing functions (training a model by adjusting parameters) using a generic computer.
Therefore, the claim as a whole remains focused on the abstract idea and fails Step 2B of the eligibility analysis.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 1-4, 6-12, and 14-18 are rejected under 35 U.S.C. 103 as being unpatentable over Zhou ("Contrastive Trajectory Learning for Tour Recommendation," 2021) in view of Li (CN 113498070 A) in further view of Trivedi ("WiFiMod: Transformer-based Indoor Human Mobility Modeling using Passive Sensing," 2021).
Regarding claim 1,
Zhou teaches [a] method for training a path representation model, comprising: acquiring at least one trajectory point of at least one user (Page 4 Introduction of Zhou, "…we propose a novel tour recommendation method Contrastive Trajectory Learning for Tour Recommendation (CTLTR)… leveraging RNN as the basic building block… We propose a new auxiliary training objective to enhance the recommendation accuracy and enrich trip representations…", Page 8 Definition 3.1, "INPUT: A user-provided query consisting of the desired start point ls and start time ts… and the end point le at time te"
Zhou's CTLTR is a deep-neural-network framework that learns trajectory (path) representations from acquired user trajectories defined as sequences of visited POIs, which corresponds to a method for training a path representation model by acquiring trajectory points.).
wherein each trajectory point of each user comprises… a place passed by the each user, a start time of reaching the place… (Page 7 Definition 3.1, “INPUT: A user-provided query consisting of the desired start point ls and start time ts, the length of the trip N (i.e., the number of POIs to visit), and the end point le at time te.”, Page 10 Section 4.1.2 Equation (8), "we encode the spatiotemporal context of each location in a trajectory… other contexts and constraints, e.g., duration time and queuing time, can be encoded in a similar way"
Zhou encodes each POI visit with its location (place) and visiting time (start time) through the process of acquiring the start and end times for staying in that location.).
inputting … of the at least one trajectory point of the at least one user into a neural network comprising a pre-trained model, and generating a trajectory representation of each user (Page 10 Section 4.1.2 Equation (8), "we encode the spatiotemporal context of each location in a trajectory… other contexts and constraints, e.g., duration time and queuing time, can be encoded in a similar way", Page 10 Section 4.1, "The Base model serves as a basic supervised framework to encode the trajectories into latent representations", Page 12 Section 4.2, "Once the CTLTR model is pre-trained, it can be used as fine-tuning on the tour recommendation problem"
Zhou inputs trajectory points into a pre-trained hierarchical RNN Base model that outputs latent trajectory representations, which corresponds to generating a trajectory representation via a neural network comprising a pre-trained model.).
outputting, for each user, a position of each trajectory point from the generated trajectory representation of the user comprising running a search operation on the generated trajectory representation, the search operation performed by using the start time and the duration of each trajectory point of the each user (Page 12 Section 4.1.2, "…we concatenate the historical information hQC t−1 and the length embedding rt… to form the current query", Page 11 Section 4.1.3, "a POI recommender based on LSTM is used to generate the next recommended POI given… all past POIs as input", Page 7 Definition 3.1, “INPUT: A user-provided query consisting of the desired start point ls and start time ts, the length of the trip N (i.e., the number of POIs to visit), and the end point le at time te. OUTPUT: The tour recommender system returns a tour route T =(l1 = ls,l2,l3,...,lN = le).”
Zhou constructs a query from a time/length embedding and uses it to recover the POI (position) at a given step from the learned trajectory representation, which maps to running a search operation on the generated trajectory representation using the start time and the duration of each trajectory point.).
and in a machine learning process, self-adjusting a network parameter of the pre-trained model according to a difference between the place passed by the each user and the position of the each trajectory point obtained from running the search operation, to obtain the path representation model (Page 14 Section 4.3.3, “We use the cross-entropy loss function to optimize the model. Specifically, for a certain trajectory T , the loss is calculated by the following:
PNG
media_image1.png
68
499
media_image1.png
Greyscale
where N is the length of the trajectory, li is the ith ground truth POI, and ˆ li is the predicted POI.”
Zhou adjusts network parameters by minimizing the difference between the ground-truth place and the predicted position, thereby producing the trained path representation model.).
Zhou does not teach wherein the data triplet is extracted from a log of a wireless access point which is configured to record access by a mobile communication device of a user.
Li, in the same field of endeavor, teaches wherein the data triplet is extracted from a log of a wireless access point which is configured to record access by a mobile communication device of a user (Last Paragraph of Page 2 in Li, "the AP prediction model is a model obtained based on historical roaming path information training of a plurality of mobile devices", Paragraph 5 Page 32, "The historical roaming path information of any one of the plurality of mobile devices includes an identification of an AP for reflecting a historical roaming path… in sequence"
Li sources each trajectory from access-point roaming records of mobile devices, where the AP identifier is the place and the connection's start time and duration complete the triplet which corresponds to extracting the data triplet from a wireless-access-point log.).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date to combine the teaching of Zhou's CTLTR framework for learning trajectory representations with Li's extraction of historical roaming-path data from wireless access point logs in order to source the trajectory from access records of mobile communication devices (Page 7 Paragraph 5 of Li).
Zhou in view of Li does not teach wherein each trajectory point of each user comprises a data triplet comprising a place passed by the each user, a start time of reaching the place and a duration of staying in the place.
Trivedi, in the same field of endeavor, teaches wherein each trajectory point of each user comprises a data triplet comprising a place passed by the each user, a start time of reaching the place and a duration of staying in the place (Page 3 Section 2 of Trivedi, "A trajectory is essentially a temporally ordered sequence of locations visited, duration of stay at each location, with transitions between two successive locations where the transit is the path used to move from the previous location to the next one. Figure 1 shows the trajectory of users P1 and P2 as a sequence of locations each visited for a specific time duration at a certain time of the day... In this case, a trajectory comprises visit to buildings, time spend inside a building, visit time of buildings, and transitions between buildings;", Page 7 Section 3.3, "all APs on the campus and indexing them by timestamp gives us a sequence of APs visited and duration of visit by each user device."
Trivedi teaches that each access point has a fixed physical location (buildings) associated with each user device. The data triplet comprises the place passed by each user (buildings), the start time of reaching the place (visit time of buildings), and the duration of staying in the place (time spent inside a building).)
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date to combine Zhou’s CTLTR framework for learning trajectory representations and Li’s extraction of historical roaming-path data from wireless access point logs with Trivedi’s teaching of a data triplet in order to characterize each visited place with its arrival time and duration when learning the trajectory representations (Page 3 Section 2 of Trivedi).
Regarding claim 2,
Zhou teaches perform self-supervised training on the path representation model (Page 4 Introduction of Zhou, “We present a mutual information-based self-supervised pre-training diagram to better capture tourists’ transition patterns and interest preferences.”)
Zhou does not teach acquiring a sample set, a sample in the sample set comprising a sample trajectory and a tag; and using respectively the sample trajectory and the tag in the sample set as an input and an expected output of the path representation model…
Li, in the same field of endeavor, teaches acquiring a sample set, a sample in the sample set comprising a sample trajectory and a tag (Paragraph 5 Page 32, "The historical roaming path information… includes an identification of an AP… in sequence", Last Paragraph of Page 3, "the expected output (i.e., tag) comprises the actual roaming AP of the second AP in the history roaming path information"
Li forms training samples from sequences of AP identifiers (the sample trajectory) each paired with the next roaming AP as the tag, which maps to a sample set of sample-trajectory/tag pairs. Li's history roaming path information sequence corresponds to the claimed sample trajectory, and Li's expected output corresponds to the claimed tag.).
Li teaches using respectively the sample trajectory and the tag in the sample set as an input and an expected output of the path representation model… (Last Paragraph of Page 3, "the continuous non-end feature data in the history roaming path information is the input data; correspondingly, the expected output (i.e., tag)…", Paragraph 2 of Page 19, "the supervised learning algorithm… sample data… is composed of input data and expected output… called tag"
Li uses the sample trajectory as the input and the tag as the expected output to train the model.)
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date to combine the teaching of Zhou's CTLTR framework and its self-supervised training signals with Li's training of a model using sample trajectories paired with a tag as input and expected output in order to generate sequence embeddings and improve predictive accuracy (Page 7, Paragraph 5 of Li).
Regarding claim 3,
Zhou teaches dividing, for a target sample trajectory…, the target sample trajectory into at least one segment (Page 14 Section 4.3.2, "we propose to model the segment-trajectory correlation… define the pretext task as a subsequence Cloze problem… We mask the subsequence [mask1,mask2,…] in the original trajectory T"
Zhou divides each trajectory into segments (subsequences), which corresponds to dividing the target sample trajectory into at least one segment.).
inputting, for each target sample trajectory, at least one segment of the target sample trajectory into the path representation model to obtain a representation of each segment of the target sample trajectory (Page 14 Section Modeling Segment-trajectory Correlation 4.3.2, “Then, we predict the masked segment based on the surrounding context T s = {l1,t1 ,..., [mask1,mask2,...],...,lN,tN}. The model is also optimized by a similarity loss function based on mutual information maximization
PNG
media_image2.png
84
527
media_image2.png
Greyscale
…Similarly to Equation (16), the mutual information between the context and the trajectory segment can be computed as
PNG
media_image3.png
54
465
media_image3.png
Greyscale
”
Zhou explicitly divides each target sample trajectory into at least one segment by masking a subsequence (the segment) and separating it from the remaining unmasked context, creating distinct trajectory segments used during learning. The path representation model then inputs these trajectory segments to obtain segment-level representations and optimizes them via mutual information maximization (MIM) which corresponds to the instant’s claim limitation of inputting at least one segment of a target sample trajectory into the model to obtain a representation.).
and constructing, for each target sample trajectory, the representation of each segment into a sequence of representations of the target sample trajectory, and inputting the sequence and a time identifier corresponding to each segment into a sequence model, to output a sequence representation of the target sample trajectory (Page 12 Section 4.1.2, " The POI generating process is executed in a recursive manner. Specifically, for a running round at time t ∈ [2, N − 1], we concatenate the historical information hQC t−1 and the length embedding rt…", Page 11 Section 4.1.3, " In CTLTR, we select the long short-term memory (LSTM) as the basic recurrent unit to model the temporal dependencies among POI trajectories."
Zhou constructs segment representations into a sequence and inputs them, with a time/length identifier rt, into an LSTM sequence model that outputs the trajectory's sequence representation.).
Zhou does not teach a target sample trajectory with a total duration exceeding a predetermined value… a target sample trajectory with a total duration exceeding a predetermined value in the sample set… according to a predetermined time interval.
Li, in the same field of endeavor, teaches a target sample trajectory with a total duration exceeding a predetermined value in the sample set … according to a predetermined time interval (Paragraphs 2 and 4 of 17, " filter condition comprises… the duration of the target communication connection is greater than the specified time threshold.", Paragraph 5 Page 32, "The historical roaming path information… includes an identification of an AP… in sequence", Last Paragraph of Page 3, "the expected output (i.e., tag) comprises the actual roaming AP of the second AP in the history roaming path information"
Li filters samples by a duration threshold which corresponds to selecting target sample trajectories whose total duration exceeds a predetermined value.)
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date to combine the teaching of Zhou's segment-based trajectory representation with Li's filtering of a sample set to include only connections whose duration exceeds a predetermined value in order to use the filtered samples for subsequent processing (Page 15, Paragraph 3 of Li).
Regarding claim 4,
Zhou teaches outputting the sequence representation of each target sample trajectory by a prediction model, to obtain a prediction result of the each target sample trajectory (Page 11 Section 4.1.3, "According to the current prediction along with previous predicted POIs, our model adjusts the query and carries out the next round of POI planning", Page 8 Section 3.1, "OUTPUT: The tour recommender system returns a tour route"
Zhou's prediction layer outputs a tour recommendation (prediction result) from the trajectory representation.).
Zhou does not teach and adjusting a network parameter of the sequence model according to a difference between the prediction result of the each target sample trajectory and a tag corresponding to the each target sample trajectory
Li, in the same field of endeavor, teaches and adjusting a network parameter of the sequence model according to a difference between the prediction result of the each target sample trajectory and a tag corresponding to the each target sample trajectory (Paragraph 3 Page 19, "based on the output data and the expected output, updating the parameter set of the initial AP prediction model by means of reverse transmission", Page 3 Paragraph 6, "repeatedly executing… until the loss value corresponding to the loss function is converged"
Li updates the model's parameters via backpropagation based on the difference between the predicted AP (prediction result) and the actual next AP (tag), which maps to adjusting the sequence-model parameter according to the prediction-result/tag difference.).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date to combine the teaching of Zhou's prediction-model output with Li's parameter updating based on the difference between the predicted next AP and the actual next AP tag in order to improve prediction accuracy for each trajectory (Page 7, Paragraph 6 of Li).
Regarding claim 6,
Zhou teaches masking, according to a masking rule, places passed by the user in a part of the at least one trajectory point of the at least one user, to obtain at least one masked trajectory point (Page 14 Section 4.3.2, " we propose to model the segment-trajectory correlation in a similar way, i.e., define the pretext task as a subsequence Cloze problem… We mask the subsequence [mask1,mask2,…] in the original trajectory T."
Zhou masks POIs (places) in part of the trajectory, which maps to obtaining masked trajectory points per a masking rule.).
Zhou teaches inputting the at least one masked trajectory point into the pre-trained model to obtain a mask position (Page 14 Section 4.3.2, "we predict the masked segment based on the surrounding context Ts"; Page 11 Section 4.1.3, "a POI recommender based on LSTM is used to generate the next recommended POI given… all past POIs as input"
Zhou inputs the masked context into the model to predict the masked segment (mask position).).
and adjusting a network parameter of the pre-trained model according to a difference between the mask position and a position of the place masked according to the masking rule, to obtain the path representation model (Page 12 Section 4.2, "We used self-supervised signals to minimize the pre-training losses via mutual information maximization (MIM)", Page 8 Section 3.1, "where y is the true label, y^ is the predicted label"
Zhou adjusts parameters by minimizing, via MIM, the difference between the predicted masked segment (mask position) and the ground-truth masked place, which maps to the claimed parameter adjustment.).
Regarding claim 7,
Zhou teaches acquiring to-be-analyzed user trajectory information; inputting the user trajectory information into the path representation model, to output a path representation; and inputting the path representation into a prediction model to output a prediction result (Page 10 Figure 1 and Caption,
PNG
media_image4.png
441
710
media_image4.png
Greyscale
"a hierarchical Base model… and a fine-tuned prediction layer for tour recommendation", Page 10 Section 4.1, "encode the trajectories into latent representations…"
Zhou acquires trajectory data, encodes it into a path representation via the Base model, and feeds that representation into a fine-tuned prediction layer to output a recommended tour (prediction result).).
Regarding claim 8,
Zhou teaches wherein the prediction result comprises at least one of: a path category, an abnormal event, a next position, or a schedule (Page 6 Section 2.3, "ST-RNN [43] models spatio-temporal data using RNN for next location prediction"
Zhou's prediction result includes next-location prediction, which satisfies the list by teaching a next position.).
Regarding claim 9,
Zhou teaches …for training a path representation model (Page 4 Introduction of Zhou, "…we propose a novel tour recommendation method Contrastive Trajectory Learning for Tour Recommendation (CTLTR)… leveraging RNN as the basic building block… We propose a new auxiliary training objective to enhance the recommendation accuracy and enrich trip representations…", Page 8 Definition 3.1, "INPUT: A user-provided query consisting of the desired start point ls and start time ts… and the end point le at time te"
Zhou's CTLTR is a deep-neural-network framework that learns trajectory (path) representations from acquired user trajectories defined as sequences of visited POIs, which corresponds to a method for training a path representation model by acquiring trajectory points).
acquiring at least one trajectory point of at least one user (Page 8 Definition 3.1 of Zhou, "INPUT: A user-provided query consisting of the desired start point ls and start time ts… and the end point le at time te"
Zhou acquires user trajectories defined as sequences of visited POIs, which corresponds to acquiring trajectory points.).
wherein each trajectory point of each user comprises… a place passed by the each user, a start time of reaching the place… (Page 7 Definition 3.1, “INPUT: A user-provided query consisting of the desired start point ls and start time ts, the length of the trip N (i.e., the number of POIs to visit), and the end point le at time te.”, Page 10 Section 4.1.2 Equation (8), "we encode the spatiotemporal context of each location in a trajectory… other contexts and constraints, e.g., duration time and queuing time, can be encoded in a similar way"
Zhou encodes each POI visit with its location (place) and visiting time (start time) through the process of acquiring the start and end times for staying in that location.)
inputting … the at least one trajectory point of the at least one user into a neural network comprising a pre-trained model … and generating a trajectory representation of each user (Page 10 Section 4.1.2 Equation (8), "we encode the spatiotemporal context of each location in a trajectory… other contexts and constraints, e.g., duration time and queuing time, can be encoded in a similar way", Page 10 Section 4.1, "The Base model serves as a basic supervised framework to encode the trajectories into latent representations", Page 12 Section 4.2, "Once the CTLTR model is pre-trained, it can be used as fine-tuning on the tour recommendation problem"
Zhou inputs trajectory points into a pre-trained hierarchical RNN Base model that outputs latent trajectory representations, which corresponds to generating a trajectory representation via a neural network comprising a pre-trained model.)
outputting, for each user, a position of each trajectory point from the generated trajectory representation of the user comprising running a search operation on the generated trajectory representation, the search operation performed by inputting to the start time and the duration of each trajectory point of the each user (Page 12 Section 4.1.2, "…we concatenate the historical information hQC t−1 and the length embedding rt… to form the current query", Page 11 Section 4.1.3, "a POI recommender based on LSTM is used to generate the next recommended POI given… all past POIs as input", Page 7 Definition 3.1, “INPUT: A user-provided query consisting of the desired start point ls and start time ts, the length of the trip N (i.e., the number of POIs to visit), and the end point le at time te. OUTPUT: The tour recommender system returns a tour route T =(l1 = ls,l2,l3,...,lN = le).”
Zhou constructs a query from a time/length embedding and uses it to recover the POI (position) at a given step from the learned trajectory representation, which maps to running a search operation on the generated trajectory representation using the start time and the duration of each trajectory point.)
and in a machine learning process, self-adjusting a network parameter of the pre-trained model according to a difference between the place passed by the each user and the position of the each trajectory point obtained from running the search operation, to obtain the path representation model (Page 14 Section 4.3.3, “We use the cross-entropy loss function to optimize the model. Specifically, for a certain trajectory T , the loss is calculated by the following:
PNG
media_image1.png
68
499
media_image1.png
Greyscale
where N is the length of the trajectory, li is the ith ground truth POI, and ˆ li is the predicted POI.”
Zhou adjusts network parameters by minimizing the difference between the ground-truth place and the predicted position, thereby producing the trained path representation model.)
and outputting, by the path representation model, at least one of: a predicted next location of the user, a category of a path taken by the user, an anomaly indicator associated with a path taken by the user, or a schedule for the user (Page 6 Section 2.3 of Zhou, "ST-RNN [43] models spatio-temporal data using RNN for next location prediction"
Zhou's output includes a predicted next location, which satisfies the list by teaching a predicted next location of the user.)
Zhou does not teach an apparatus… comprising: at least one processor; and a storage device, wherein the storage device stores instructions executable by the at least one processor, and the instructions when executed by the at least one processor cause the at least one processor to perform operations comprising… wherein the data triplet is extracted from a log of a wireless access point which is configured to record access by a mobile communication device of a user… wherein the pre-trained model comprises a Transformer neural network having a self-attention mechanism… wherein the search task is performed by a fully connected layer
Li, in the same field of endeavor, teaches an apparatus… comprising at least one processor; and a storage device, wherein the storage device stores instructions executable by the at least one processor, and the instructions when executed by the at least one processor cause the at least one processor to perform operations comprising (Page 6 Paragraph 9, “a third aspect, there is provided an AP prediction device, comprising: a processor and a memory; the memory is used for storing the computer program; the computer program comprises a program instruction; the processor is used for invoking the computer program, realizing the AP prediction method according to any one of the first aspect.”
Li discloses a processor-and-memory apparatus storing instructions executable by the processor to predict the AP.).
wherein the data triplet is extracted from a log of a wireless access point which is configured to record access by a mobile communication device of a user (Last Paragraph of Page 2 in Li, "the AP prediction model is a model obtained based on historical roaming path information training of a plurality of mobile devices", Paragraph 5 Page 32, "The historical roaming path information of any one of the plurality of mobile devices includes an identification of an AP for reflecting a historical roaming path… in sequence"
Li sources each trajectory from access-point roaming records of mobile devices, where the AP identifier is the place and the connection's start time and duration complete the triplet which corresponds to extracting the data triplet from a wireless-access-point log.)
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date to combine the teaching of Zhou's CTLTR framework for learning trajectory representations with Li's extraction of historical roaming-path data from wireless access point logs and Li's processor-and-memory apparatus storing executable instructions in order to source the trajectory from access records of mobile communication devices and to implement the training on computing hardware (Page 7 Paragraph 5 and Page 6 Paragraph 9).
Zhou in view of Li does not teach wherein each trajectory point of each user comprises a data triplet comprising a place passed by the each user, a start time of reaching the place and a duration of staying in the place…. wherein the pre-trained model comprises a Transformer neural network having a self-attention mechanism… the … task is performed by a fully connected layer.
Trivedi, in the same field of endeavor, teaches wherein the pre-trained model comprises a Transformer neural network having a self-attention mechanism (Page 6 Section 3.4.2 of Trivedi, " The Transformer neural network architecture… follows an encoder-decoder structure… which is comprised of L layers of the same form. Each layer j passes its inputs through two sub-layers, multi-head self-attention and a feed-forward layer, with residual connections… Attention(Q,K,V) = softmax(QK^T/sqrt(d_k)V)”.
Trivedi models user mobility with a Transformer employing multi-head self-attention, which corresponds to the pre-trained model comprising a Transformer neural network having a self-attention mechanism.)
wherein the … task is performed by a fully connected layer (Page 6 Section 3.4.3 of Trivedi, "We train this model using a self-supervised autoregressive training objective… we pass the outputs obtained from the Transformer encoder through an additional linear layer of dimension d ×V, which is shared across all modalities."
Trivedi teaches passing the Transformer output through an additional linear layer of dimension d × V to produce the output distribution. The linear layer corresponds to the fully connected layer since a linear layer connects every input node to every output node through its weight matrix and is the same component referred to as a dense or fully connected layer.).
wherein each trajectory point of each user comprises a data triplet comprising a place passed by the each user, a start time of reaching the place and a duration of staying in the place (Page 3 Section 2 of Trivedi, "A trajectory is essentially a temporally ordered sequence of locations visited, duration of stay at each location, with transitions between two successive locations where the transit is the path used to move from the previous location to the next one. Figure 1 shows the trajectory of users P1 and P2 as a sequence of locations each visited for a specific time duration at a certain time of the day... In this case, a trajectory comprises visit to buildings, time spend inside a building, visit time of buildings, and transitions between buildings;", Page 7 Section 3.3, "all APs on the campus and indexing them by timestamp gives us a sequence of APs visited and duration of visit by each user device."
Trivedi teaches that each access point has a fixed physical location (buildings) associated with each user device. The data triplet comprises the place passed by each user (buildings), the start time of reaching the place (visit time of buildings), and the duration of staying in the place (time spent inside a building).)
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date to further combine Zhou in view of Li’s teaching with Trivedi’s Transformer-with-self-attention architecture and its fully connected/linear output layer in order to capture long-term representations and to map that representation to an output position (Section 3.4.3 of Trivedi).
Regarding claim 10,
Zhou does not teach acquiring a sample set, a sample in the sample set comprising a sample trajectory and a tag; and using respectively the sample trajectory and the tag in the sample set as an input and an expected output of the path representation model to perform supervised training on the path representation model.
Li, in the same field of endeavor, teaches acquiring a sample set, a sample in the sample set comprising a sample trajectory and a tag (Paragraph 5 Page 32, "The historical roaming path information… includes an identification of an AP… in sequence", Last Paragraph of Page 3, "the expected output (i.e., tag) comprises the actual roaming AP of the second AP in the history roaming path information"
Li forms training samples from sequences of AP identifiers (the sample trajectory) each paired with the next roaming AP as the tag, which maps to a sample set of sample-trajectory/tag pairs. Li's history roaming path information sequence corresponds to the claimed sample trajectory, and Li's expected output corresponds to the claimed tag.).
Li teaches using respectively the sample trajectory and the tag in the sample set as an input and an expected output of the path representation model… (Last Paragraph of Page 3, "the continuous non-end feature data in the history roaming path information is the input data; correspondingly, the expected output (i.e., tag)…", Paragraph 2 of Page 19, "the supervised learning algorithm… sample data… is composed of input data and expected output… called tag"
Li uses the sample trajectory as the input and the tag as the expected output to train the model.)
to perform supervised training on the path representation model (Paragraph 2 of Page 19, “the machine learning algorithm can be divided into supervised learning algorithm… the supervised learning algorithm …sample data… is composed of input data and expected output… called tag”, Paragraph 7 of Page 4,“The AP prediction model is obtained based on historical roaming path information training of the plurality of mobile devices.”).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date to combine the teaching of Zhou's CTLTR framework and its self-supervised training signals with Li's training of a model using sample trajectories paired with a tag as input and expected output in order to generate sequence embeddings and improve predictive accuracy (Page 7, Paragraph 5 of Li).
Claim 11 recites similar limitations to claim 3. Therefore, claim 11 is rejected using the same rationale as claim 3.
Claim 12 recites similar limitations to claim 4. Therefore, claim 12 is rejected using the same rationale as claim 4.
Claim 14 recites similar limitations to claim 6. Therefore, claim 14 is rejected using the same rationale as claim 6.
Regarding claim 15,
Zhou teaches acquiring to-be-analyzed user trajectory information; inputting the user trajectory information into the path representation model, to output a path representation; and inputting the path representation into a prediction model comprising a classifier or a fully connected layer to output a prediction result (Page 10 Figure 1 and Caption,
PNG
media_image4.png
441
710
media_image4.png
Greyscale
"a hierarchical Base model… and a fine-tuned prediction layer for tour recommendation", Page 10 Section 4.1, "encode the trajectories into latent representations…", Page 9 Section 4.1, “The Base model serves as a basic supervised framework to encode the trajectories into latent representations containing semantic relationships and sequential visiting patterns between POIs.”
Zhou acquires trajectory data, encodes it into a path representation via the Base model (classifier), and feeds that representation into a fine-tuned prediction layer to output a recommended tour (prediction result).).
Claim 16 recites similar limitations to claim 8. Therefore, claim 16 is rejected using the same rationale as claim 8.
Regarding claim 17,
Zhou teaches acquiring at least one trajectory point of at least one user (Page 4 Introduction of Zhou, "…we propose a novel tour recommendation method Contrastive Trajectory Learning for Tour Recommendation (CTLTR)… leveraging RNN as the basic building block…", Page 8 Definition 3.1, "INPUT: A user-provided query consisting of the desired start point ls and start time ts… and the end point le at time te"
Zhou's CTLTR is a deep-neural-network framework that learns trajectory (path) representations from acquired user trajectories defined as sequences of visited POIs, which corresponds to acquiring trajectory points.).
inputting … the at least one trajectory point of the at least one user into a neural network comprising a pre-trained model … and generating a trajectory representation of each user (Page 10 Section 4.1.2 Equation (8), "we encode the spatiotemporal context of each location in a trajectory… other contexts and constraints, e.g., duration time and queuing time, can be encoded in a similar way", Page 10 Section 4.1, "The Base model serves as a basic supervised framework to encode the trajectories into latent representations", Page 12 Section 4.2, "Once the CTLTR model is pre-trained, it can be used as fine-tuning on the tour recommendation problem"
Zhou inputs trajectory points into a pre-trained hierarchical RNN Base model that outputs latent trajectory representations, which corresponds to generating a trajectory representation via a neural network comprising a pre-trained model.).
Zhou teaches outputting, for each user, a position of each trajectory point from the generated trajectory representation of the user comprising running a search operation on the generated trajectory representation, the search operation performed by using the start time and the duration of each trajectory point of the each user (Page 12 Section 4.1.2, "…we concatenate the historical information hQC t−1 and the length embedding rt… to form the current query", Page 11 Section 4.1.3, "a POI recommender based on LSTM is used to generate the next recommended POI given… all past POIs as input", Page 7 Definition 3.1, “INPUT: A user-provided query consisting of the desired start point ls and start time ts, the length of the trip N (i.e., the number of POIs to visit), and the end point le at time te. OUTPUT: The tour recommender system returns a tour route T =(l1 = ls,l2,l3,...,lN = le).”
Zhou constructs a query from a time/length embedding and uses it to recover the POI (position) at a given step from the learned trajectory representation, which maps to running a search operation on the generated trajectory representation using the start time and the duration of each trajectory point.).
and in a machine learning process, self-adjusting a network parameter of the pre-trained model according to a difference between the place passed by the each user and the position of the each trajectory point obtained from running the search operation, to obtain the path representation model (Page 14 Section 4.3.3, “We use the cross-entropy loss function to optimize the model. Specifically, for a certain trajectory T , the loss is calculated by the following:
PNG
media_image1.png
68
499
media_image1.png
Greyscale
where N is the length of the trajectory, li is the ith ground truth POI, and ˆ li is the predicted POI.”
Zhou adjusts network parameters by minimizing the difference between the ground-truth place and the predicted position, thereby producing the trained path representation model.).
to perform self-supervised training on the path representation model (Page 4 Introduction of Zhou, "We present a mutual information-based self-supervised pre-training diagram to better capture tourists' transition patterns and interest preferences."
Zhou performs self-supervised training in which the supervisory signal is from the trajectory data itself.).
Zhou teaches dividing … the target sample trajectory into at least one segment (Page 14 Section 4.3.2, "we propose to model the segment-trajectory correlation… define the pretext task as a subsequence Cloze problem… We mask the subsequence [mask1,mask2,…] in the original trajectory T"
Zhou divides each trajectory into segments (subsequences), which corresponds to dividing the target sample trajectory into at least one segment.).
Zhou teaches inputting, for each target sample trajectory, at least one segment of the target sample trajectory into the path representation model to obtain a representation of each segment of the target sample trajectory (Page 14 Section Modeling Segment-trajectory Correlation 4.3.2, “Then, we predict the masked segment based on the surrounding context T s = {l1,t1 ,..., [mask1,mask2,...],...,lN,tN}. The model is also optimized by a similarity loss function based on mutual information maximization
PNG
media_image2.png
84
527
media_image2.png
Greyscale
…Similarly to Equation (16), the mutual information between the context and the trajectory segment can be computed as
PNG
media_image3.png
54
465
media_image3.png
Greyscale
”
Zhou explicitly divides each target sample trajectory into at least one segment by masking a subsequence (the segment) and separating it from the remaining unmasked context, creating distinct trajectory segments used during learning. The path representation model then inputs these trajectory segments to obtain segment-level representations and optimizes them via mutual information maximization (MIM) which corresponds to the instant claim limitation of inputting at least one segment of a target sample trajectory into the model to obtain a representation.).
Zhou teaches constructing, for each target sample trajectory, the representation of each segment into a sequence of representations of the target sample trajectory, and inputting the sequence and a time identifier corresponding to each segment into a sequence model, to output a sequence representation of the target sample trajectory (Page 12 Section 4.1.2, " The POI generating process is executed in a recursive manner. Specifically, for a running round at time t ∈ [2, N − 1], we concatenate the historical information hQC t−1 and the length embedding rt…", Page 11 Section 4.1.3, " In CTLTR, we select the long short-term memory (LSTM) as the basic recurrent unit to model the temporal dependencies among POI trajectories."
Zhou constructs segment representations into a sequence and inputs them, with a time/length identifier rt, into an LSTM sequence model that outputs the trajectory's sequence representation.).
Zhou does not teach a non-transitory computer readable storage medium, storing a computer instruction, wherein the computer instruction when executed by a computer causes the computer to perform operations comprising … wherein the data triplet is extracted from a log of a wireless access point which is configured to record access by a mobile communication device of a user… wherein the pre-trained model comprises a Transformer neural network comprising a self-attention mechanism… acquiring a sample set, a sample in the sample set comprising a sample trajectory and a tag; and using respectively the sample trajectory and the tag in the sample set as an input and an expected output of the path representation model, to perform self-supervised training on the path representation model.
Li, in the same field of endeavor teaches a non-transitory computer readable storage medium, storing a computer instruction, wherein the computer instruction when executed by a computer causes the computer to perform operations comprising (Page 6 Paragraph 9 of Li, "an AP prediction device, comprising: a processor and a memory; the memory is used for storing the computer program; the computer program comprises a program instruction; the processor is used for invoking the computer program, realizing the AP prediction… a fifth aspect, providing a chip, the chip comprises a programmable logic circuit and/or program instruction, when the chip running, realizing the AP prediction..."
Li discloses a non-transitory storage medium storing computer instruction executed by a processor to carry out the method.).
wherein the data triplet is extracted from a log of a wireless access point which is configured to record access by a mobile communication device of a user (Last Paragraph of Page 2 of Li, "the AP prediction model is a model obtained based on historical roaming path information training of a plurality of mobile devices", Paragraph 5 Page 32, "The historical roaming path information of any one of the plurality of mobile devices includes an identification of an AP for reflecting a historical roaming path… in sequence"
Li sources each trajectory from access-point roaming records of mobile devices, where the AP identifier is the place and the connection's start time and duration complete the triplet, which corresponds to extracting the data triplet from a wireless-access-point log.).
acquiring a sample set, a sample in the sample set comprising a sample trajectory and a tag, and using respectively the sample trajectory and the tag in the sample set as an input and an expected output of the path representation model (Paragraph 5 Page 32, "The historical roaming path information… includes an identification of an AP… in sequence", Last Paragraph of Page 3, "the continuous non-end feature data in the history roaming path information is the input data; correspondingly, the expected output (i.e., tag)…", Paragraph 2 of Page 19, "the supervised learning algorithm… sample data… is composed of input data and expected output… called tag", Paragraph 2 of Page 19, "the supervised learning algorithm… sample data… is composed of input data and expected output… called tag"
Li forms training samples from sequences of AP identifiers (the sample trajectory) each paired with a tag, and uses the sample trajectory as the input and the tag as the expected output to train the model.).
Li further teaches …a target sample trajectory with a total duration exceeding a predetermined value in the sample set … according to a predetermined time interval (Paragraphs 2 and 4 of Page 17 in Li, "filter condition comprises… the duration of the target communication connection is greater than the specified time threshold.", Paragraph 5 Page 32, "The historical roaming path information… includes an identification of an AP… in sequence", Last Paragraph of Page 3, "the expected output (i.e., tag) comprises the actual roaming AP of the second AP in the history roaming path information"
Li filters samples by a duration threshold, which corresponds to selecting a target sample trajectory whose total duration exceeds a predetermined value and applying a predetermined time interval.).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date to combine the teaching of Zhou's CTLTR framework for learning trajectory representations with Li's non-transitory storage medium storing executable instructions and extraction of data triplets from wireless access point logs in order to source the trajectory from access records of mobile communication devices and implement the operations on a computer-readable medium to train the model.
Zhou in view of Li does not teach wherein each trajectory point of each user comprises a data triplet comprising a place passed by the each user, a start time of reaching the place and a duration of staying in the place…. wherein the pre-trained model comprises a Transformer neural network comprising a self-attention mechanism.
Trivedi, in the same field of endeavor, teaches wherein the pre-trained model comprises a Transformer neural network comprising a self-attention mechanism (Page 6 Section 3.4.2 of Trivedi, "The Transformer neural network architecture… follows an encoder-decoder structure… which is comprised of L layers of the same form. Each layer j passes its inputs through two sub-layers, multi-head self-attention and a feed-forward layer, with residual connections… Attention(Q,K,V) = softmax(QK^T/sqrt(d_k)V)."
Trivedi models user mobility with a Transformer employing multi-head self-attention, which corresponds to the pre-trained model comprising a Transformer neural network comprising a self-attention mechanism.).
wherein each trajectory point of each user comprises a data triplet comprising a place passed by the each user, a start time of reaching the place and a duration of staying in the place (Page 3 Section 2 of Trivedi, "A trajectory is essentially a temporally ordered sequence of locations visited, duration of stay at each location, with transitions between two successive locations where the transit is the path used to move from the previous location to the next one. Figure 1 shows the trajectory of users P1 and P2 as a sequence of locations each visited for a specific time duration at a certain time of the day... In this case, a trajectory comprises visit to buildings, time spend inside a building, visit time of buildings, and transitions between buildings;", Page 7 Section 3.3, "all APs on the campus and indexing them by timestamp gives us a sequence of APs visited and duration of visit by each user device."
Trivedi teaches that each access point has a fixed physical location (buildings) associated with each user device. The data triplet comprises the place passed by each user (buildings), the start time of reaching the place (visit time of buildings), and the duration of staying in the place (time spent inside a building).)
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date to further combine Zhou in view of Li’s teachings with Trivedi's Transformer-with-self-attention architecture in order to capture long-term mobility dependencies in the learned representation (Section 3.4.2 of Trivedi).
Regarding claim 18,
Zhou in view of Li does not teach the pre-trained model comprises a Transformer neural network comprising a self-attention mechanism.
Trivedi, in the same field of endeavor, teaches the pre-trained model comprises a Transformer neural network comprising a self-attention mechanism (Page 6 Section 3.4.2 of Trivedi, "The Transformer neural network architecture… follows an encoder-decoder structure… which is comprised of L layers of the same form. Each layer j passes its inputs through two sub-layers, multi-head self-attention and a feed-forward layer, with residual connections… Attention(Q,K,V) = softmax(QK^T/sqrt(d_k)V)."
Trivedi models user mobility with a Transformer employing multi-head self-attention, which corresponds to the pre-trained model comprising a Transformer neural network comprising a self-attention mechanism.).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date to further combine Zhou in view of Li’s teachings with Trivedi's Transformer-with-self-attention architecture in order to capture long-term mobility dependencies in the learned representation (Section 3.4.2 of Trivedi).
Claims 5 and 13 are rejected under 35 U.S.C. 103 as being unpatentable over Zhou ("Contrastive Trajectory Learning for Tour Recommendation," 2021) in view of Li (CN 113498070 A), in view of Trivedi ("WiFiMod: Transformer-based Indoor Human Mobility Modeling using Passive Sensing," 2021), and in further view of Jia (CN 113268291 A).
Regarding claim 5,
Zhou in view of Li in view of Trivedi does not teach wherein the tag comprises at least one of: a path category tag, an abnormal event tag, or a schedule tag.
Jia, in the same field of endeavor teaches wherein the tag comprises at least one of: a path category tag, an abnormal event tag, or a schedule tag (Claim 15 of Jia, “…obtaining a plurality of schedule events corresponding to the plurality of schedule category tags in the selected time range.”)
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date to combine the teaching of Zhou in view of Li in view of Trivedi’s teaching with Jia’s schedule tag in order to provide a labeled schedule tag as the output for classification (Page 5 Paragraphs 5 and 6 of Jia).
Claim 13 recited similar limitations to claim 5. Therefore, claim 13 is rejected using the same rationale.
Claims 19–20 are rejected under 35 U.S.C. § 103 as being unpatentable over Zhou ("Contrastive Trajectory Learning for Tour Recommendation," 2021) in view of Li (CN 113498070 A), in view of Trivedi ("WiFiMod: Transformer-based Indoor Human Mobility Modeling using Passive Sensing," 2021), and in further view of Gui et al. ("A hierarchical temporal attention-based LSTM encoder-decoder model for individual mobility prediction," 2020).
Regarding claim 19,
Zhou in view of Li in view of Trivedi does not dividing the inputted sequence into segments according to the periodicity, comprising dividing into segments at intervals of days; inputting the segments into the pre-trained model to obtain the representation of each segment; constructing the representations into a sequence, inputting the constructed sequence into a sequence model, and outputting the representation of the whole sequence.
Gui, in the same field of endeavor, teaches dividing the inputted sequence into segments according to the periodicity, comprising dividing into segments at intervals of days (Page 3 Section 3.1 of Gui, "To capture such daily and weekly periodicity as well as avoid data sparseness problem, we organize location sequences by day. Locations where an individual orderly visited in each day is represented as a location sequence, and we input location sequences of last week to predict the location sequences of next week."
Gui divides the input location sequence into per-day segments to capture daily and weekly periodicity, where each day's visited locations form one segment. The organizing of location sequences by day corresponds to the dividing of the inputted sequence into segments at intervals of days according to the periodicity.).
inputting the segments into the pre-trained model to obtain the representation of each segment (Page 5 Section 3.3.2 of Gui, "Given the location sequence of the dth day (x1d, x2d . . . xLd ), the local temporal attention obtains the weighted summation vector of this day… Htd represents the regularity vector of the dth day."
Gui inputs each day's location sequence (the segment) into its temporal attention model and produces a per-day regularity vector Htd. The per-day regularity vector Htd corresponds to the representation of each segment.).
constructing the representations into a sequence (Page 5 Section 3.3.2 of Gui, "Local temporal attention obtains the regularity vector sequence {Ht1 , Ht2 . . . Ht7 } of a week.", Page 4 Section 3.1, “We use day-ID, an enumeration value ranging from one to seven, to denote the day of a week.”
Gui assembles the per-day regularity vectors into a week-level sequence of seven daily vectors. The regularity vector sequence {Ht1 … Ht7} corresponds to the constructing of the representations into a sequence, and Gui's day-ID embedding corresponds to the time identifier associated with each segment.).
inputting the constructed sequence into a sequence model, and outputting the representation of the whole sequence (Page 4 Section 3.2 of Gui, "After T time steps, the encoder summarizes the whole input sequence into the final vectors cT and hT", Page 5 Section 3.3.2, "H′t is the week-level vector that summarizes all correlated information of last week for target location prediction."
Gui feeds the week-level sequence of daily vectors into its LSTM encoder-decoder and produces a week-level vector that summarizes the whole sequence. The LSTM encoder-decoder corresponds to the claimed sequence model, and the week-level summarizing vector corresponds to the claimed representation of the whole sequence.).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date to combine the teaching of Zhou in view of Li in view of Trivedi’s teaching with Gui's division of the location sequence into day-interval segments according to periodicity in order to capture daily and weekly travel regularities for improved long-term prediction (Page 5 and Introduction of Gui).
Regarding claim 20,
Zhou in view of Li in view of Trivedi does not dividing the inputted sequence into segments according to the periodicity, comprising dividing into segments at intervals of days; inputting the segments into the pre-trained model to obtain the representation of each segment; constructing the representations into a sequence, inputting the constructed sequence into a sequence model, and outputting the representation of the whole sequence.
Gui, in the same field of endeavor, teaches dividing the inputted sequence into segments according to the periodicity, comprising dividing into segments at intervals of days (Page 3 Section 3.1 of Gui, "To capture such daily and weekly periodicity as well as avoid data sparseness problem, we organize location sequences by day. Locations where an individual orderly visited in each day is represented as a location sequence, and we input location sequences of last week to predict the location sequences of next week."
Gui divides the input location sequence into per-day segments to capture daily and weekly periodicity, where each day's visited locations form one segment. The organizing of location sequences by day corresponds to the dividing of the inputted sequence into segments at intervals of days according to the periodicity.).
inputting the segments into the pre-trained model to obtain the representation of each segment (Page 5 Section 3.3.2 of Gui, "Given the location sequence of the dth day (x1d, x2d . . . xLd ), the local temporal attention obtains the weighted summation vector of this day… Htd represents the regularity vector of the dth day."
Gui inputs each day's location sequence (the segment) into its temporal attention model and produces a per-day regularity vector Htd. The per-day regularity vector Htd corresponds to the representation of each segment.).
constructing the representations into a sequence (Page 5 Section 3.3.2 of Gui, "Local temporal attention obtains the regularity vector sequence {Ht1 , Ht2 . . . Ht7 } of a week.", Page 4 Section 3.1, “We use day-ID, an enumeration value ranging from one to seven, to denote the day of a week.”
Gui assembles the per-day regularity vectors into a week-level sequence of seven daily vectors. The regularity vector sequence {Ht1 … Ht7} corresponds to the constructing of the representations into a sequence, and Gui's day-ID embedding corresponds to the time identifier associated with each segment.).
inputting the constructed sequence into a sequence model, and outputting the representation of the whole sequence (Page 4 Section 3.2 of Gui, "After T time steps, the encoder summarizes the whole input sequence into the final vectors cT and hT", Page 5 Section 3.3.2, "H′t is the week-level vector that summarizes all correlated information of last week for target location prediction."
Gui feeds the week-level sequence of daily vectors into its LSTM encoder-decoder and produces a week-level vector that summarizes the whole sequence. The LSTM encoder-decoder corresponds to the claimed sequence model, and the week-level summarizing vector corresponds to the claimed representation of the whole sequence.).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date to combine the teaching of Zhou in view of Li in view of Trivedi’s teaching with Gui's division of the location sequence into day-interval segments according to periodicity in order to capture daily and weekly travel regularities for improved long-term prediction (Page 5 and Introduction of Gui).
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to MAJD MAHER HADDAD whose telephone number is (571)272-2265. The examiner can normally be reached Mon-Friday 8-5 pm.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Kamran Afshar, can be reached at (571) 272-7796. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/M.M.H./Examiner, Art Unit 2125
/KAMRAN AFSHAR/Supervisory Patent Examiner, Art Unit 2125
Prosecution Insights