DETAILED ACTION
Response to Amendment
1. This office action is in response to applicant’s communication filed on 06/16/2026 in response to PTO Office Action mailed 03/19/2026. The Applicant’s remarks and amendments to the claims and/or the specification were considered with the results as follows.
2. In response to the last Office Action, claims 25-34 are amended. No claims are added or canceled. As a result, claims 1-34 are pending in this office action.
3. The 35 USC 101 rejections have been withdrawn upon re-evaluation.
Response to Arguments
4. Applicant's arguments with respect to 35 USC 103 have been fully considered but are not persuasive and the details are as follows:
Applicant’s argument stated as “The Examiner’s proposed combination of Mehrasa and Villegas is predicted on the Examiner’s mischaracterization of Mehrasa as being directed to the same field of endeavor…one of ordinary skill in the art would not combine Mehrasa with Villegas because there would be no expectation of success”.
In response to Applicant’s argument, the Examiner disagrees because the Mehrasa reference discloses outputting an encoded history of context features r1, r2…. rt, wherein the encoded history is derived from event times (see Mehrasa, paragraphs [0140]-[0146]). Specifically, Mehrasa teaches that past actions (i.e., context features) are encoded into a vector representation. For example, prior LSTM 114A performs long short-term memory processing on the concatenated vector representation x.sub.n.sup.emb for past actions. Thus, Mehrasa teaches encoding historical context features into an encoded history as claimed.
The Examiner relies on Villegas only for its teaching of modifying the encoded history of Mehrasa by applying a local history window of size k, wherein the local history window excludes events occurring more than k events ago. The combination merely modifies the encoded history generated by Mehrasa using a known history-window technique taught by Villegas.
Applicant's argument that there is no teaching, suggestion, or motivation to combine the references is not persuasive. Obviousness may be established by combining or modifying the teachings of the prior art where there is some teaching, suggestion, or motivation to do so, either explicitly in the references themselves or implicitly from the knowledge generally available to one of ordinary skill in the art. See In re Fine, 837 F.2d 1071, 5 USPQ2d 1596 (Fed. Cir. 1988); In re Jones, 958 F.2d 347, 21 USPQ2d 1941 (Fed. Cir. 1992); and KSR Int'l Co. v. Teleflex Inc., 550 U.S. 398 (2007). In the present case, both Mehrasa and Villegas are directed to machine learning techniques for event prediction using encoder-based frameworks. Although the references may address different applications, they both process historical event information to improve prediction performance. Therefore, one of ordinary skill in the art would have recognized that the local history window technique disclosed by Villegas could be incorporated into the encoded history generated by Mehrasa to improve the processing of sequential event data by focusing on more relevant recent events while excluding less relevant historical events. Such a modification would have been a predictable use of prior art elements according to their established functions and would have yielded a reasonable expectation of success. Accordingly, the rejection under 35 U.S.C. § 103 is maintained.
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 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.
Claims 1-3, 12-15 and 24-26 are rejected under 35 U.S.C. 103 as being unpatentable over Mehrasa et al. (US 2020/0160176 A1), hereinafter Mehrasa and in view of Villegas et al. (US 2020/0082248 A1), hereinafter Villegas.
Referring to claims 1, 13 and 24, Mehrasa discloses a computer-implemented machine learning method for prediction for a temporal point process (See para. [0062], predicting a future event using a temporal point process, which is characterized by a conditional intensity function), the method comprising:
receiving, by at least one trained encoder (See para [0140] and Figure 6, receiving data representing irregularly spaced actions including past actions and a current action for training a variational auto-encoder model), an event series comprising a plurality of discrete event times τ₁, τ₂ τᵢ (See para. [0004], para. [0005], para. [0144] and Figure 6, step 430 modeling event data in continuous time using the temporal point process);
outputting, by the at least one trained encoder, an encoded history of context features r₁, r₂ r₁, wherein: the encoded history is derived from the event times (See para. [0140]-para. [0146], the past actions [e.g., context features] are encoded into a vector representation, for example, prior LSTM 114A performs long short-term memory on the concatenated vector representation x.sub.n.sup.emb for past actions) […];
generating a global feature G from the encoded history, wherein generating the global feature G is performed using a subset r₁, r₂ rᵢ₋₁ of the encoded history that excludes a most recent one r₁ of the context features (See para. [0140]-para. [0146], the past actions [e.g., context features] are encoded into a vector representation [e.g. global feature G], note in para. [0147], the current action [e.g. the most recent action] is not encoded in the vector representation 450A, the current action is encoded into a vector representation 450B for the current action);
providing, to a trained decoder (See para. [0151]-para. [0152], providing to an action decoder):
a representation of the global feature G (See para. [0140]-para. [0146] and Figure 6, the past actions [e.g., context features] are encoded into a vector representation [e.g. global feature G] in step 460A); and
the most recent one rₗ of the context features r₁, r₂ r (See para. [0140]- para. [0147] and Figure 6, the current action is obtained in step 460B);
outputting by the trained decoder, a prediction for a time τₗ₊₁ of a next event, wherein the prediction is derived from at least the representation of the global feature G and the most recent one rₗ of the context features r₁, r₂ r₁ (See para. [0148] – para. [0158] and Figure 6, the action decoder 118A generates a probability distribution over action categories for the current action, the time decoder generates a probability distribution over inter-arrival time for the current action, the system predicts probabilities of action categories and probabilities of inter-arrival times of a next action using the trained model).
Mehrasa does not explicitly the encoded object is restricted to a local window of size k, where the local window excludes those of event times that are more than k events ago.
Villegas discloses the encoded object is restricted to a local window of size k, where the local window excludes those of event times that are more than k events ago (See para. [0037], the encoder generates encoded state feature corresponding to the object, the encoded object can be tracked over a predefined number of time steps, the oldest object is removed from consideration, the state information is stored in a rolling buffer such that, at each time instance, an oldest state information is removed from the buffer and state information corresponding to a current state minus one time step is added to the buffer).
Therefore, it 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 was made to modify the size of the encoded history of Mehrasa to be restricted to a buffer size k, where the buffer size excludes the oldest information, as taught by Villegas. Skilled artisan would have been motivated to provide a more accurate representation of each of the objects, and help to predict the next or future movements of the objects as they move through the environment (See Villegas, para. [0038]). Furthermore, both references (Mehrasa and Villegas) teach features that are directed to analogous art and the same field of endeavor, namely machine learning techniques using encoder-based frameworks for event prediction. Accordingly, the close relationship between the two references supports a reasonable expectation of success.
As to claims 2, 14 and 25, Mehrasa discloses wherein the representation of the global feature G is the global feature G itself (See para. [0140]-para. [0146] and Figure 6, the past actions [e.g., context features] are encoded into a vector representation [e.g. global feature G] in step 460A).
As to claims 3, 15 and 26, Mehrasa discloses wherein the representation of the global feature G is a global latent variable Z for the global feature G (See para. [0011], determining a conditional posterior distribution for the current action, based at least in part on the current vector representation; sampling a latent variable from the conditional prior distribution).
As to claim 12, Mehrasa discloses prior to receiving the event series at the trained encoder: building the at least one trained encoder; and building the trained decoder (See para. [0056] and para. [0092], a variational auto-encoder (VAE) [13] describes a generative process with simple prior p.sub.θ(z) [usually chosen to be a multivariate Gaussian] and complex likelihood p.sub.θ(x|z) [the parameters of which are produced by neural networks], a latent variable z.sub.n may be sampled from the posterior [or prior during testing] distribution, and is fed to decoder networks action decoder 118A and time decoder 1186 for generating distributions over action category a.sub.n and inter-arrival time τ.sub.n]).
Claims 4-9, 16-21 and 27-32 are rejected under 35 U.S.C. 103 as being unpatentable over Mehrasa (US 2020/0160176 A1) and in view of Villegas (US 2020/0082248 A1) and further in view of Condessa (US 2024/0201668 A1).
As to claims 4, 16 and 27, Mehrasa does not explicitly disclose applying cross-attention to the subset of the encoded history to generate an attention feature.
Condessa discloses applying cross-attention to the subset r₁, r₂ rᵢ₋₁ of the encoded history to generate an attention feature ri; and providing the attention feature ri to the trained decoder; wherein the prediction is further derived from the attention feature (See para. [0029] and para. [0030], the machine learning system 140 includes at least (i) a transformer encoder 300 and (ii) a transformer decoder 302. The encoder 300 is applied to the history embedding sequence 308 to produce intermediate history features 312. The encoder 300 uses a linear network followed by multiple layers of causal transformer blocks to apply self-attention to the history embedding sequence 308. In the decoder network, the input embedding sequence 310 and the intermediate history features 312 are combined using a cross-attention mechanism. The input embedding sequence 310 goes through a series of causal self-attention layers to produce the queries for the cross-attention layer. At the same time, the intermediate history features 312 are fed to the cross-attention layer as both keys and values. The output of the cross-attention layer goes through further processing, such as getting combined with the residual connection and passing through a fully-connected network to convert to the final output [i.e., the predicted measurement data]).
Therefore, it 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 was made to modify the training model of Mehrasa to apply cross-attention to the subset of the encoded history to generate an attention feature, as taught by Condessa. Skilled artisan would have been motivated to provide a relatively accurate prediction or estimation of the target data (See Condessa, para. [0019]). In addition, all references (Condessa, Mehrasa and Villegas) teach features that are directed to analogous art and they are directed to the same field of endeavor, such as machine learning techniques using encoder-based frameworks for event prediction. This close relation between both references highly suggests an expectation of success.
As to claims 5, 17 and 28, Mehrasa discloses wherein the representation of the global feature G is a global latent variable Z for the global feature G (See para. [0011], determining a conditional posterior distribution for the current action, based at least in part on the current vector representation; sampling a latent variable from the conditional prior distribution).
As to claims 6, 18 and 29, Mehrasa discloses wherein the at least one encoder is a single encoder (See para. [0140], training a variational auto-encoder).
As to claims 7, 19 and 30, Mehrasa discloses wherein the at least one encoder is a first encoder and a second encoder (See para. [0082] and Figure 2, the previous and current actions and their inter-arrival times (i.e., x.sub.n−1 and x.sub.n) are embedded by separate embedders 113 into respective vector representations).
As to claims 8, 20 and 31, Mehrasa discloses wherein the first encoder and the second encoder are different encoders (See para. [0082] and Figure 2, the previous and current actions and their inter-arrival times (i.e., x.sub.n−1 and x.sub.n) are embedded by separate embedders 113 into respective vector representations).
As to claims 9, 21 and 32, Mehrasa discloses wherein the first encoder and the second encoder share at least some model parameters (See para. [0082] and Figure 2, the previous and current actions and their inter-arrival times (i.e., x.sub.n−1 and x.sub.n) are embedded by separate embedders 113 into respective vector representations).
Claims 10, 22 and 33 are rejected under 35 U.S.C. 103 as being unpatentable over Mehrasa (US 2020/0160176 A1) and in view of Villegas (US 2020/0082248 A1) and Condessa (US 2024/0201668 A1) and further in view of Ogawa (US 2016/0344429 A1).
As to claims 10, 22 and 33, Mehrasa does not explicitly wherein the first encoder and the second encoder are duplicate encoders.
Ogawa discloses wherein the first encoder and the second encoder are duplicate encoders (See para. [0076] it is common practice to duplicate the encoder 40 itself to eliminate or minimize resulting increases in cost).
Therefore, it 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 was made to modify the encoders of Mehrasa to include duplicate encoders, as taught by Ogawa. Skilled artisan would have been motivated to provide a relatively accurate prediction or estimation of the target data (See Ogawa, para. [0076]). In addition, all references (Condessa, Ogawa, Mehrasa and Villegas) teach features that are directed to analogous art and they are directed to the same field of endeavor, such as encoding information data. This close relation between both references highly suggests an expectation of success.
Claims 11, 23 and 34 are rejected under 35 U.S.C. 103 as being unpatentable over Mehrasa (US 2020/0160176 A1) and in view of Villegas (US 2020/0082248 A1) and further in view of Page (US 2020/0090357 A1).
As to claims 11, 23 and 34, Mehrasa does not explicitly disclose the global feature G is a permutation-invariant operation incorporating all members.
Page discloses wherein the global feature G is a permutation-invariant operation incorporating all members (See para. [0050], the system uses permutation invariant operations (maximum operation) which can effectively capture global features).
Therefore, it 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 was made to modify the system to include a permutation-invariant operation, as taught by Page. Skilled artisan would have been motivated to capture global features effectively (See Page, para. [0050]). In addition, all references (Page, Mehrasa and Villegas) teach features that are directed to analogous art and they are directed to the same field of endeavor, such as encoding information data. This close relation between both references highly suggests an expectation of success.
Conclusion
THIS ACTION IS MADE FINAL. 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 YUK TING CHOI whose telephone number is (571) 270-1637. The examiner can normally be reached Monday-Friday 9am-6pm.
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, AMY NG can be reached at 5712701698. 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.
/YUK TING CHOI/Primary Examiner, Art Unit 2164
Prosecution Insights