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 .
Response to Amendment
The amendments and arguments filed 01/02/2026 have been entered. Claims 1-3, 5-13, 15-20 remain pending in the application.
Applicant’s amendments and arguments, with respect to claim rejections of claims 1-3, 5-13, 15-20, under 35 U.S.C 101 filed 10/02/2025 have been considered and are not persuasive. Therefore, the previous rejections as set forth in the previous office action will be maintain.
The applicant traverses the rejection under 35 U.S.C. 101 and initially maintains that independent claims 1, 11, and 20 are not directed to an abstract idea under Step 2A, Prong One. Nevertheless, even assuming without admitting that the claims recite an abstract idea or mental process, Applicant argues that the claims integrate the alleged abstract idea into a practical application under Step 2A, Prong Two when the limitations are considered both individually and as an ordered combination.
In particular, Applicant relies on the amended limitations requiring determining a current context of the autonomous vehicle, determining that the current context corresponds to a target context, selecting one or more neural-network parameter values from the target distribution for the target context to adjust the neural network of the AV, and operating the AV in its current context using the neural network. Applicant contends that these limitations, individually and collectively, go beyond merely determining or selecting information because they apply the selected parameter values to the neural network and result in the practical and physical operation of the AV in the current context.
Applicant therefore argues that operating the AV based on the adjusted neural network meaningfully limits the alleged abstract idea and constitutes a real-world technological application, rather than merely applying the abstract idea on a generic computer. Accordingly, Applicant submits that independent claims 1, 11, and 20, and the claims depending therefrom, are patent eligible at least under Step 2A, Prong Two, and requests withdrawal of the 101 rejection.
Applicant’s arguments have been considered and are not persuasive. Under Step 2A, Prong One, the limitations of determining the current context of the AV, determining whether the current context corresponds to the target context, and selecting parameter values based on that correspondence encompass observing information, comparing the current and target contexts, evaluating whether the contexts correspond, and selecting values based on that evaluation. As broadly recited, and without any particular computational technique, these observations, comparisons, evaluations, judgments, and selections can practically be performed in the human mind and therefore fall within the mental-process grouping of abstract ideas.
Under Step 2A, Prong Two, the additional elements do not integrate the mental process into a practical application. Obtaining the target distribution merely supplies the information used in performing the context comparison and parameter selection. Although the claim further applies the selected values to adjust a neural network and operates the AV using the neural network, these functions are recited only at a high level of generality. The claim does not specify a particular technical procedure for implementing the selected parameter values, a particular neural-network architecture or updating technique, or a particular manner in which the adjustment improves the functioning of the neural network or the operation of the AV. Thus, the additional elements generally apply the recited mental process using a neural network in the technological environment of an AV, rather than imposing a meaningful technological limitation on the mental process.
Considering the limitations as an ordered combination does not alter this conclusion. The claim broadly obtains context-associated parameter information, determines whether the current context corresponds to the target context, selects corresponding parameter values, adjusts a neural network using those values, and uses the neural network to operate an AV. However, the claim does not recite how the selected values technically change the neural network’s processing, what particular vehicle-control operation is performed based on the adjustment, or how the claimed combination provides a technological improvement over ordinary neural-network-based AV operation. Accordingly, the ordered combination merely instructs that the result of the abstract context evaluation and parameter selection be implemented through generally recited neural-network and AV functions.
Under Step 2B, as established by the evidence of record, updating neural-network parameters and using a neural network to operate an AV are well-understood, routine, and conventional functions performed according to their ordinary purposes. These additional elements, individually and in combination, merely implement the abstract context evaluation and parameter selection without supplying an inventive concept. Therefore, independent claims 1, 11, and 20, and the claims depending therefrom, remain ineligible under 35 U.S.C. 101.
Applicant’s amendments and arguments, with respect to claim rejections of claims 1-3, 5-13, 15-20, under 35 U.S.C 103 filed 10/02/2025 have been considered some of them are persuasive.
The applicant argues that Gendron-Bellemare does not disclose the amended limitation requiring a target distribution of internal neural-network parameter values based on internal parameter values collected from neural networks of one or more AVs. According to Applicant, Gendron-Bellemare instead discloses probability distributions of possible Q returns generated as outputs of a distributional Q network, which are technically distinct from the internal parameters, such as weights and biases, used by the network to produce those outputs. Applicant therefore contends that Gendron-Bellemare does not anticipate amended independent claims 1, 11, and 20. Applicant further contends that Mnih does not cure this alleged deficiency with respect to dependent claims 5 and 15, and accordingly requests withdrawal of the rejections under 35 U.S.C. 102 and 103
The Examiner respectfully agrees that Gendron-Bellemare’s disclosed probability distributions concern possible Q returns generated as neural-network outputs and are distinct from the claimed target distribution of internal neural-network parameter values based on parameter values collected from one or more AVs.
Nevertheless, Gendron-Bellemare remains applicable to substantial portions of amended independent claims 1, 11, and 20. Gendron-Bellemare teaches a reinforcement-learning system including a neural network having internal network parameter values, wherein the agent may be an autonomous or semi-autonomous vehicle operating in a real-world environment. Gendron-Bellemare further teaches receiving a current observation characterizing the current state of the vehicle’s environment, processing the observation using the neural network and its internal parameter values, selecting an action based on the neural-network output, and using the selected action as a control input for operating the vehicle in the current environment. Accordingly, the amendment does not negate the applicability of Gendron-Bellemare to the recited AV, neural network and internal parameters generally, determination of the AV’s current driving context, and operation of the AV using the neural network.
However, upon further consideration of the amended claims, a new ground of rejection necessitated by Applicant’s amendment is set forth below.
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.
Claim 3 is 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 3 recites the limitation "the exploration distribution". There is insufficient antecedent basis for this limitation in the claim. Claim 3 depends from claim 1, thus the rejection of claim 1 is incorporated. However, amended claim 1, from which claim 3 depends, no longer introduces or defines an exploration distribution. Therefore, it is unclear which distribution is referenced by “the exploration distribution”, rendering the scope of claim 3 indefinite. For examination purposes, the examiner interprets “the exploration distribution” as a broader collection of neural network parameter values associated with a plurality of autonomous vehicles or operating contexts, from which a context-specific target distribution is selected or derived.
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 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an
abstract idea without significantly more.
Regarding claim 1,
Step 1:
Claim 1 recites a method, one of the four statutory categories of patentable subject matter.
Step 2A, Prong I:
Claim 1 further recites the limitations of:
“determining a current context of the AV, the current context comprising at least one of a driving environment associated with a location of the AV, a hardware configuration of the AV, a software configuration of the AV, or a task of the AV” This limitation recites a mental process. A person can mentally or manually determine the context of an AV such as determining the weather around the vehicle or the hardware of the vehicle. Such determination is a mental process and is capable to be performed by a human’s mind.
“in response to determining that the current context of the AV corresponds to the target context, selecting the one or more of the neural network parameter values in the target distribution for the target context to adjust the neural network of the AV” This limitation recites a mental process. A person can mentally or manually select a parameter value based on a distribution after determining a target context. The process of selecting a value is a mental process and is capable to be performed by a human’s mind.
“... the target context comprises at least one of a driving environment associated with a location, a hardware configuration of one or more AVs, a software configuration of the one or more AVs, and a task of the one or more Avs” This limitation recites a mental process. A person can mentally or manually determine the context of an AV such as determining the weather around the vehicle or the hardware of the vehicle or the software implemented by the vehicle, or what the vehicle is doing. Such determination is a mental process and is capable to be performed by a human’s mind.
Step 2A, Prong II:
Claim 1 recites the following additional elements:
“obtaining, at an autonomous vehicle (AV), for a target context, a target distribution of neural network parameter values of a neural network parameter of a neural network of the AV, ... collected from one or more Avs” This additional element recites an additional element of an insignificant extra-solution activity as identified in MPEP 2106.05(g) of mere data gathering, and does not provide integration into a practical application.
“... wherein the target distribution is based on internal neural network parameter values of one or more neural networks ...” This additional element recites an additional element of a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), and does not provide integration into a practical application. The limitation recites a black-box application of machine learning neural network. The element recites the application of using values to determine a target distribution without reciting the specific algorithm or unconventional method to obtain the distribution. The limitation simply recites the application of values from a neural network to determine a distribution without providing any improvement toward a machine learning algorithm, or a specific computer element.
“operating, using the neural network of the AV, the AV in the current context of the AV” This additional element recites an additional element of a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), and does not provide integration into a practical application. The limitation recites a black-box application of machine learning neural network. The limitation simply recites using a neural network to operate an automatic vehicle, without providing specific step or algorithm of the neural network to implement the operation of the vehicle. The limitation does not recite any improvement toward a neural network to operate a vehicle, nor any unconventional method to operate an automatic vehicle, or an improvement toward the automatic vehicle or the neural network itself.
Step 2B:
When considered individually or in combination, the additional limitations and elements of claim 1 does not amount to significantly more than the judicial exception for the same reasons discussed above as to why the additional limitations do not integrate the abstract idea into a practical application. The additional elements of outlined in Step 2A performing functions as designed simply accomplishes execution of the abstract ideas.
The additional element “obtaining, at an autonomous vehicle (AV), for a target context, a target distribution of neural network parameter values of a neural network parameter of a neural network of the AV, ...” further recites an additional element of a well-understood, routine, conventional activity as identified in MPEP 2106.05(d) of receiving or transmitting data over a network, and does not amount to significantly more than the judicial exception for the same reasons discussed above.
The additional element “... wherein the target distribution is based on internal neural network parameter values of one or more neural networks collected from one or more Avs” further recites an additional element of a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), and does not amount to significantly more than the judicial exception for the same reasons discussed above.
The additional element “operating, using the neural network of the AV, the AV in the current context of the AV” further recites an additional element of a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), and does not amount to significantly more than the judicial exception for the same reasons discussed above.
In conclusions from above for the elements considered as a mental process, elements reciting a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), elements reciting a well-understood, routine, conventional activity as identified in MPEP 2106.05(d), and elements reciting an insignificant extra-solution activity as identified in MPEP 2106.05(g) are carried over and do not provide significantly more than the abstract idea. Looking at the limitations in combination and the claims as a whole does not change this conclusion and the claim is ineligible.
Therefore, additional limitations of claim 1 do not amount to significantly more than the judicial exception.
Thus, claim 1 recites abstract ideas with additional elements rendered at a high level of generality resulting in claims that do not integrate the abstract idea into a practical application or amount to significantly more than the judicial exception.
Therefore, claim 1 is not patent eligible.
Regarding claim 2 depends on claim 1, thus the rejection of claim 1 is incorporated.
“The method of Claim 1, further comprising: implementing a subset of the neural network parameter values in the target distribution in the neural network of the AV in the target context” This additional element recites an additional element of a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), and does not provide integration into a practical application or significantly more than the judicial exception. The limitation recites black box application of a neural network values within a vehicle in a specific context. Such operating of a neural network according to a value in a specific context is a conventional practice of a neural network that operates using a value under a set rule. Simply implementing a value to operate a neural network does not integrate the abstract idea of determine a context of a vehicle and select a value of a neural network into a practical application or amount to significantly more than the judicial exception
Thus, claim 2 recites additional elements rendered at a high level of generality resulting in claims that do not integrate the abstract idea into a practical application or amount to significantly more than the judicial exception.
Regarding claim 3 depends on claim 1, thus the rejection of claim 1 is incorporated.
“The method of Claim 1, wherein the neural network parameter values in the target distribution comprise a subset of the neural network parameter values in the exploration distribution” This additional element merely indicating a field of use or technological environment in which to apply a judicial exception as identified in MPEP 2106.05(h), and does not provide integration into a practical application or significantly more than the judicial exception.
Thus, claim 3 recites additional elements rendered at a high level of generality resulting in claims that do not integrate the abstract idea into a practical application or amount to significantly more than the judicial exception.
Regarding claim 5 depends on claim 1, thus the rejection of claim 1 is incorporated.
“The method of Claim 1 further comprising: in response to determining that an amount of change in the current context of the AV exceeds a threshold, updating a neural network of a stack of the AV with the selected one or more of the neural network parameter values.” This additional element recites an additional element of a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), and does not provide integration into a practical application or significantly more than the judicial exception. The limitation recites black box application of conventional value update method via using a threshold without providing any improvement toward a machine learning algorithm, a neural network or a computer element.
Thus, claim 5 recites additional elements rendered at a high level of generality resulting in claims that do not integrate the abstract idea into a practical application or amount to significantly more than the judicial exception.
Regarding claim 6 depends on claim 1, thus the rejection of claim 1 is incorporated.
“The method of Claim 1, wherein selecting the one or more of the neural network parameter values includes: searching the target distribution for determining the one or more of the neural network parameter values based on at least one of the target context and a search algorithm, ...” This additional element recites an additional element of a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), and does not provide integration into a practical application or significantly more than the judicial exception. The limitation recites black box application of a search algorithm, and a target context to search for a target distribution, without reciting the specific search algorithm, an improvement or an unconventional search algorithm or improvement toward a computer element.
“... the search algorithm comprising at least one of Bayesian optimization and reinforcement learning” This additional element merely indicating a field of use or technological environment in which to apply a judicial exception as identified in MPEP 2106.05(h), and does not provide integration into a practical application or significantly more than the judicial exception.
Thus, claim 6 recites additional elements rendered at a high level of generality resulting in claims that do not integrate the abstract idea into a practical application or amount to significantly more than the judicial exception.
Regarding claim 7 depends on claim 1, thus the rejection of claim 1 is incorporated.
“The method of Claim 1, wherein selecting the one or more of the neural network parameter values includes: searching the target distribution for the one or more of the neural network parameter values based on a search cost function” This additional element recites an additional element of a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), and does not provide integration into a practical application or significantly more than the judicial exception. The limitation recites black box application of a search cost function, without reciting the specific search cost algorithm, an improvement or an unconventional search cost algorithm or improvement toward a computer element.
Thus, claim 7 recites additional elements rendered at a high level of generality resulting in claims that do not integrate the abstract idea into a practical application or amount to significantly more than the judicial exception.
Regarding claim 8 depends on claim 1, thus the rejection of claim 1 is incorporated.
“The method of Claim 1, wherein selecting the one or more of the neural network parameter values includes: searching the target distribution for the one or more of the neural network parameter values based on at least one of a safety metric associated with one or more searched neural network parameter values in the target distribution, a comfort metric associated with one or more searched neural network parameter values in the target distribution, and a performance metric associated with one or more searched neural network parameter values in the target distribution” This additional element recites an additional element of a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), and does not provide integration into a practical application or significantly more than the judicial exception. The limitation recites black box application of various metrics such as a safety metric, a comfort metric, or a performance metric as a mean to evaluate the output of a neural network algorithm, without providing any improvement toward the neural network search algorithm or an unconventional search algorithm or improvement toward a computer element.
Thus, claim 8 recites additional elements rendered at a high level of generality resulting in claims that do not integrate the abstract idea into a practical application or amount to significantly more than the judicial exception.
Regarding claim 9 depends on claim 1, thus the rejection of claim 1 is incorporated.
“The method of Claim 1, further comprising: assigning a weight to each neural network parameter value in at least one of the exploration distribution and the target distribution based on a likelihood of usage of such neural network parameter value” This additional element recites an additional element of a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), and does not provide integration into a practical application or significantly more than the judicial exception. The limitation recites black box application of weight assignment onto parameter value, which is a conventional application of weight in machine learning practice to indicate the importance of a value. The limitation does not recite any specific weight handling technique or unconventional weight evaluation algorithm or improvement toward a computer element.
Thus, claim 9 recites additional elements rendered at a high level of generality resulting in claims that do not integrate the abstract idea into a practical application or amount to significantly more than the judicial exception.
Regarding claim 10 depends on claim 1, thus the rejection of claim 1 is incorporated.
“The method of Claim 1, wherein the neural network parameter includes at least one of a layer width, a kernel size, a number of layers, a depth of layers, a learning rate, and a number of layers in a block.” This additional element merely indicating a field of use or technological environment in which to apply a judicial exception as identified in MPEP 2106.05(h), and does not provide integration into a practical application or significantly more than the judicial exception.
Thus, claim 10 recites additional elements rendered at a high level of generality resulting in claims that do not integrate the abstract idea into a practical application or amount to significantly more than the judicial exception.
Regarding claim 11,
Step 1:
Claim 11 recites a system, one of the four statutory categories of patentable subject matter.
Step 2A, Prong I:
Claim 11 further recites the limitations of:
“determine an exploration distribution of an internal neural network parameter for one or more neural networks of one or more autonomous vehicles (AVs), the exploration distribution of the neural network parameter comprising neural network parameter values” This limitation recites a mental process. A person can review a collection of neural-network parameter values and mentally organize or identify the value as an exploration distribution, with or without pen and paper
“determine, for a target context, a target distribution of neural network parameter values of the neural network parameter from the exploration distribution, the target context comprising at least one of a driving environment associated with a location, a hardware configuration of one or more AVs, a software configuration of the one or more AVs, and a task of the one or more AVs” This limitation recites a mental process. A person can evaluate the target context and mentally select or group parameter values from the exploration distribution that are associated with the target context to form the target distribution.
“determine a current context of the AV, the current context comprising at least one of a driving environment associated with a location of the AV, a hardware configuration of the AV, a software configuration of the AV, or a task of the AV” This limitation recites a mental process. A person can mentally or manually determine the context of an AV such as determining the weather around the vehicle or the hardware of the vehicle or the software implemented by the vehicle, or what the vehicle is doing. Such determination is a mental process and is capable to be performed by a human’s mind.
“in response to determining that the current context of the AV corresponds to the target context, select the one or more of the neural network parameter values in the target distribution for the target context to adjust the neural network of the AV” This limitation recites a mental process. A person can mentally or manually select a parameter value based on a distribution after determining a target context. The process of selecting a value is a mental process and is capable to be performed by a human’s mind.
Step 2A, Prong II:
Claim 11 recites the following additional elements:
“a memory”, “one or more processors coupled to the memory, the one or more processors being configured”, “an AV configured”, “a computer of an AV” these additional elements are a high-level recitation of generic computer components used as a tool, and does not provide integration into a practical application.
“provide, to a computer of an AV, the target distribution for implementing one or more of the neural network parameter values in the target distribution to adjust a neural network of the AV for operation in the target context” This additional element recites an additional element of an insignificant extra-solution activity as identified in MPEP 2106.05(g) of mere data gathering, and does not provide integration into a practical application.
“operate, using the neural network of the AV, the AV in the current context of the AV” This additional element recites an additional element of a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), and does not provide integration into a practical application. The limitation recites a black-box application of machine learning neural network. The limitation simply recites using a neural network to operate an automatic vehicle, without providing specific step or algorithm of the neural network to implement the operation of the vehicle. The limitation does not recite any improvement toward a neural network to operate a vehicle, nor any unconventional method to operate an automatic vehicle, or an improvement toward the automatic vehicle or the neural network itself.
Step 2B:
When considered individually or in combination, the additional limitations and elements of claim 1 does not amount to significantly more than the judicial exception for the same reasons discussed above as to why the additional limitations do not integrate the abstract idea into a practical application. The additional elements of outlined in Step 2A performing functions as designed simply accomplishes execution of the abstract ideas.
The additional element “a memory”, “one or more processors coupled to the memory, the one or more processors being configured”, “an AV configured” are a high-level recitation of generic computer components used as a tool, and does not provide integration into a practical application.
The additional element “provide, to a computer of an AV, the target distribution for implementing one or more of the neural network parameter values in the target distribution to adjust a neural network of the AV for operation in the target context” further recites an additional element of a well-understood, routine, conventional activity as identified in MPEP 2106.05(d) of receiving or transmitting data over a network, and does not amount to significantly more than the judicial exception for the same reasons discussed above.
The additional “operate, using the neural network of the AV, the AV in the current context of the AV” further recites an additional element of a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), and does not amount to significantly more than the judicial exception for the same reasons discussed above.
In conclusions from above for the elements considered as a mental process, elements reciting a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), elements reciting a well-understood, routine, conventional activity as identified in MPEP 2106.05(d), and elements reciting an insignificant extra-solution activity as identified in MPEP 2106.05(g) are carried over and do not provide significantly more than the abstract idea. Looking at the limitations in combination and the claims as a whole does not change this conclusion and the claim is ineligible.
Therefore, additional limitations of claim 11 do not amount to significantly more than the judicial exception.
Thus, claim 11 recites abstract ideas with additional elements rendered at a high level of generality resulting in claims that do not integrate the abstract idea into a practical application or amount to significantly more than the judicial exception.
Therefore, claim 11 is not patent eligible.
Regarding claim 12 depends on claim 11, thus the rejection of claim 11 is incorporated. Claim 12 is further rejected under the same rationale as claim 2. Applicant is directed to the rejection of claim 2 above because the claim recites similar limitation and concepts.
Regarding claim 13 depends on claim 11, thus the rejection of claim 11 is incorporated. Claim 13 is further rejected under the same rationale as claim 3. Applicant is directed to the rejection of claim 3 above because the claim recites similar limitation and concepts.
Regarding claim 15 depends on claim 11, thus the rejection of claim 11 is incorporated. Claim 15 is further rejected under the same rationale as claim 5. Applicant is directed to the rejection of claim 5 above because the claim recites similar limitation and concepts.
Regarding claim 16 depends on claim 11, thus the rejection of claim 11 is incorporated. Claim 16 is further rejected under the same rationale as claim 6. Applicant is directed to the rejection of claim 6 above because the claim recites similar limitation and concepts.
Regarding claim 17 depends on claim 11, thus the rejection of claim 11 is incorporated. Claim 17 is further rejected under the same rationale as claim 7. Applicant is directed to the rejection of claim 7 above because the claim recites similar limitation and concepts.
Regarding claim 18 depends on claim 11, thus the rejection of claim 11 is incorporated. Claim 18 is further rejected under the same rationale as claim 9. Applicant is directed to the rejection of claim 9 above because the claim recites similar limitation and concepts.
Regarding claim 19 depends on claim 11, thus the rejection of claim 11 is incorporated. Claim 19 is further rejected under the same rationale as claim 10. Applicant is directed to the rejection of claim 10 above because the claim recites similar limitation and concepts.
Regarding claim 20 recites a machine, one of the four statutory categories of patentable subject matter.
Claim recites the following additional elements:
“A non-transitory computer-readable medium having stored thereon instructions which, when executed by one or more processors, cause the one or more processors” these additional elements are a high-level recitation of generic computer components used as a tool, and does not provide integration into a practical application.
Claim 20 is further rejected under the same rationale as claim 1. Applicant is further directed to the rejection of claim 1 above because the claim recites similar limitation and concepts.
Claim Rejections - 35 USC § 103
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 1-4, 6-14, 16-20 are rejected under 35 U.S.C. 103 as being unpatentable over Gendron-Bellemare et.al (US 12056593 B2) in view of Himayat et.al (US 20210188306 A1)
Regarding claim 1,
Gendron-Bellemare teaches or at least suggests a part of the limitation “A method comprising: ... the target context comprises at least one of a driving environment associated with a location, a hardware configuration of one or more AVs, a software configuration of the one or more AVs, and a task of the one or more AVs” (Col. 5 – Lines 46-57 “In some implementations, the environment 106 is a real-world environment and the agent 104 is a mechanical agent interacting with the real-world environment ... the agent 104 may be an autonomous or semi-autonomous vehicle navigating through the environment 106 ... the observations 108 may be generated by or derived from sensors of the agent 104. For example, the observations 108 may be captured by a camera of the agent 104. As another example, the observations 108 may be derived from data captured from a laser sensor of the agent”. Gendron-Bellemare teaches that the reinforcement-learning agent may be an autonomous or semi-autonomous vehicle navigating through a real-world environment and that observations characterizing the environment may be generated from vehicle sensors, such as cameras or laser sensors. Thus, the disclosed real-world driving environment and vehicle-navigation task correspond to at least one of the recited types of target context, as claimed.)
Gendron-Bellemare teaches or at least suggests the limitation “determining a current context of the AV, the current context comprising at least one of a driving environment associated with a location of the AV, a hardware configuration of the AV, a software configuration of the AV, or a task of the AV” (Col. 5 – Lines 36-42 “At each time step, the system 100 receives data characterizing the current state of the environment 106, e.g., an image of the environment 106, and selects an action 102 to be performed by the agent 104 in response to the received data. Data characterizing a state of the environment 106 will be referred to in this specification as an observation 108”, and Col. 5 – Lines 46-57 “In some implementations, the environment 106 is a real-world environment and the agent 104 is a mechanical agent interacting with the real-world environment ... the agent 104 may be an autonomous or semi-autonomous vehicle navigating through the environment 106 ... the observations 108 may be generated by or derived from sensors of the agent 104. For example, the observations 108 may be captured by a camera of the agent 104. As another example, the observations 108 may be derived from data captured from a laser sensor of the agent” Gendron-Bellemare teaches receiving, at each time step, data characterizing the current state of the environment, referred to as an observation, wherein the observation may be generated or derived from sensors of the autonomous vehicle. Accordingly, receiving and processing the sensed information characterizing the environment in which the AV is presently navigating corresponds to determining a current context of the AV comprising at least a driving environment or a task of the AV, as claimed.)
Gendron-Bellemare teaches or at least suggests the limitation “operating, using the neural network of the AV, the AV in the current context of the AV” (Col. 5 – Lines 34-37 “The reinforcement learning system 100 selects actions 102 to be performed by an agent 104 interacting with an environment 106 at each of multiple time steps.” Gendron-Bellemare teaches that the reinforcement-learning system selects actions to be performed by the agent while the agent interacts with the environment at successive time steps. Because the agent may be an autonomous or semi-autonomous vehicle and the actions are selected using the reinforcement-learning neural network in response to the current environmental observation, performing those actions corresponds to operating, using the neural network of the AV, the AV in its current context, as claimed.)
Gendron-Bellemare does not teach “... obtaining, at an autonomous vehicle (AV), for a target context, a target distribution of neural network parameter values of a neural network parameter of a neural network of the AV, wherein the target distribution is based on internal neural network parameter values of one or more neural networks collected from one or more AVs, ...” However, Himayat teaches or at least suggest this part of the limitation (paragraph 17 “A method of implementing distributed AI or ML learning for autonomous vehicles is disclosed. A plurality of vehicles is identified based on a context of an operation of each of the plurality of vehicles. The context of the operation includes at least one of a location and a type of the location. Each of the plurality of vehicles has generated an AI or ML model specific to the location or the type of the location based on training data collected from one or more sensors of the vehicle while the vehicle was present at the location or the type of the location”, paragraph 28 “the distributed AI or ML modules, described in more detail below, are configured to implement one or more of the operations described herein (e.g., to provide for distribution and aggregation of AI or ML models that are specific to the location or the type of location)”, paragraph 31 “each vehicle driving by an intersection or close to a Road-side Unit (RSU) is configured to contribute parameters of a locally learned AI or ML model to a server hosted on road-side infrastructure (e.g., and RSU) and/or hosted in another vehicle. Several types of models may be trained using this approach, each of which can provide different benefits. For example, AI or ML-based models for driving profiles and behavior may be trained in a distributed manner without sacrificing privacy”. Himayat discloses identifying a plurality of vehicles according to a context of operation and generating respective AI or ML models specific to that context. Himayat further teaches distributing and aggregating the context-specific models and expressly teaches that each vehicle contributes parameters of its locally learned AI or ML model to a server or another vehicle. Thus, the parameters contributed from the respective context-specific vehicle models are collected and used to form aggregated, context-specific model information that is distributed for use by an AV. Accordingly, Himayat teaches or at least suggests obtaining, at an AV and for a target context, a target distribution of neural-network parameter values based on internal neural-network parameter values collected from neural networks of one or more Avs, as claimed.)
Gendron-Bellemare does not teach “in response to determining that the current context of the AV corresponds to the target context, selecting the one or more of the neural network parameter values in the target distribution for the target context to adjust the neural network of the AV” However, Himayat teaches or at least suggest this part of the limitation (paragraph 31 “each vehicle driving by an intersection or close to a Road-side Unit (RSU) is configured to contribute parameters of a locally learned AI or ML model to a server hosted on road-side infrastructure (e.g., and RSU) and/or hosted in another vehicle. Several types of models may be trained using this approach, each of which can provide different benefits.”, paragraph 51 “At operation 504, an AI or ML model is caused to be transmitted from the RSU or one of the plurality of vehicles to an additional vehicle... In example embodiments, this transmission is based on a determination that the additional vehicle is present or anticipated to be present at the location or the type of the location”, paragraph 52 “At operation 605, the transmitted AI or ML model is caused to be deployed in the additional vehicle such that at least one assessment (e.g., of a safe speed, a driving behavior, and so on) made by an AI or ML autonomous driving application that is configured to control the additional vehicle is optimized for the location or the type of the location”, and paragraph 56 “the deploying of the transmitted AI or ML model in the additional vehicle includes aggregating the transmitted AI or ML model with an additional AI or ML model that was previously generated locally by an AI or ML learning application installed on the additional vehicle”. Himayat discloses that vehicles contribute parameters of their locally learned AI or ML models and that a corresponding context-specific model is transmitted to an additional vehicle upon determining that the additional vehicle is present or anticipated to be present at the associated location or type of location. Himayat further teaches deploying the transmitted model in the additional vehicle to optimize the autonomous-driving application for that context and aggregating the transmitted model with the model locally generated by the additional vehicle. Accordingly, Himayat teaches or at least suggests that, in response to determining that the current context of the AV corresponds to the target context, one or more neural-network parameter values associated with that target context are selected and implemented to adjust the neural network of the AV, as claimed.)
Before the effective filing date, it would have been obvious to one of ordinary skill in the art to combine the teaching of distributional reinforcement learning, wherein autonomous vehicle operates in a specific environment using neural network model by Gendron-Bellemare with the teaching of distributed learning to learn context-specific driving patterns by Himayat. The motivation to do so is referred to in Himayat’s disclosure (paragraph 13 “the solution includes training AI or ML models in a distributed manner using data collected for a particular context and/or a particular location from a cluster of vehicles that have operated and/or are currently operating in that particular context and/or that particular location”, paragraph 15 “Per vehicle or per user profiles (e.g., of driving behaviors) can then be used to train AD models in a correlated manner, accounting for behaviors of multiple agents at a given instance/location. Once the correlated, time and location specific multi-agent driving behavioral models are collected, many different methods may be used to develop optimal AD algorithms targeting different objectives or key performance indicators, such as (a) driving efficiency subject to safety rules such as RSS, (b) defensive driving that emphasizes safety with relaxed requirements on driving time, and so on”, paragraph 16 “In example embodiments, distributed training, wherein AD models or AI- or ML-based user profile models are trained locally within a vehicle and then combined in a coordinated manner ... Advantages include one or more of the following: privacy-preserving and bandwidth-efficient approach to profiling user behavior, collecting training data for AI models for profiling driving patterns as a function of location and time across multiple agents, or enhanced AD solutions for capturing location/time specific driving profiles of users or vehicles.” Himayat discloses that distributed, context-specific learning provides benefits including accounting for the behaviors of multiple agents, improving driving efficiency and safety, preserving privacy, reducing communication bandwidth, and enhancing location- and time-specific autonomous-driving solutions. In view of these identified benefits, a person of ordinary skill in the art would have been motivated to modify Gendron-Bellemare’s reinforcement-learning system for operating an autonomous vehicle in a current environment by incorporating Himayat’s technique of training local AI or ML models using information obtained from multiple vehicles operating in a particular context or location and combining the local models in a distributed manner. Such a modification would have predictably enabled Gendron-Bellemare’s AV neural network to use parameter information learned from multiple vehicles operating in the same or similar context, thereby adapting and improving operation of the AV for the particular driving environment.)
Regarding claim 2 depends on claim 1, thus the rejection of claim 1 is incorporated.
Gendron-Bellemare in view of Himayat teaches or at least suggest “The method of Claim 1, further comprising: implementing a subset of the neural network parameter values in the target distribution in the neural network of the AV in the target context” (Col. 1 - Lines 51-64 “The distributional Q network is a deep neural network that is configured to process the action and the current observation in accordance with current values of the network parameters to generate a network output that defines a probability distribution over possible Q returns for the action-current observation pair. Each possible Q return is an estimate of a return that would result from the agent performing the action in response to the current observation. For each action, a measure of central tendency of the possible Q returns with respect to the probability distribution for the action-current observation pair is determined. An action is selected to be performed by the agent in response to the current observation using the measures of central tendency for the actions” Gendron-Bellemare discloses that the distributional Q network processes the current observation and action using current values of its internal network parameters to generate an output and select an action for the agent in the current environment. As established in the incorporated rejection of claim 1, Himayat teaches or suggests obtaining and deploying context-specific parameter values derived from parameters contributed by vehicle models. Thus, in the combination, the context-specific parameter values taught by Himayat are implemented as the current internal network parameter values used by Gendron-Bellemare’s neural network to process the observation and operate the AV in the corresponding target context. Accordingly, Gendron-Bellemare in view of Himayat teaches or at least suggests implementing a subset of the neural-network parameter values in the target distribution in the neural network of the AV in the target context)
Regarding claim 3 depends on claim 1, thus the rejection of claim 1 is incorporated.
Gendron-Bellemare in view of Himayat teaches or at least suggest “wherein the neural network parameter values in the target distribution comprise a subset of the neural network parameter values in the exploration distribution” (Col. 8 – Lines 52-58 “The system 100 may periodically (e.g., after given numbers of training iterations) update the values of the target distributional Q network parameters based on the current values of the distributional Q network parameters. For example, the system 100 may update the values of the target distributional Q network parameters to match the current values of the distributional Q network parameters” Gendron-Bellemare discloses deriving the target-network parameter values from the corresponding current-network parameter values. Under the Examiner’s interpretation above, and the incorporated teachings of Himayat, a broader collection of neural network parameter values associated with a plurality of autonomous vehicles or operating contexts corresponds to the parameter values in the exploration distribution, from which a context-specific target distribution is selected or derived. Gendron-Bellemare further teaches deriving/updating the target-network parameter values based on corresponding current-network parameter values, including setting the target-network parameter values to match the current-network parameter values. Thus, in the combined system, the context-specific target-network parameter values are selected, copied, or otherwise derived from the broader collection of parameter values and therefore comprise at least a subset thereof. Accordingly, Gendron-Bellemare in view of Himayat teaches or at least suggests that the neural-network parameter values in the target distribution comprise a subset of the neural-network parameter values in the exploration distribution, as claimed.)
Regarding claim 6 depends on claim 1, thus the rejection of claim 1 is incorporated.
Gendron-Bellemare in view of Himayat teaches or at least suggest “wherein selecting the one or more of the neural network parameter values includes: searching the target distribution for determining the one or more of the neural network parameter values based on at least one of the target context and a search algorithm, the search algorithm comprising at least one of Bayesian optimization and reinforcement learning.” (Col. 1 - Lines 51-64 “The distributional Q network is a deep neural network that is configured to process the action and the current observation in accordance with current values of the network parameters to generate a network output that defines a probability distribution over possible Q returns for the action-current observation pair. Each possible Q return is an estimate of a return that would result from the agent performing the action in response to the current observation. For each action, a measure of central tendency of the possible Q returns with respect to the probability distribution for the action-current observation pair is determined. An action is selected to be performed by the agent in response to the current observation using the measures of central tendency for the actions”, and Col. 7 - Lines 45-48 “The system 100 includes a training engine 124 that is configured to train the distributional Q network 112 over multiple training iterations using reinforcement learning techniques” Gendron-Bellemare discloses training the distributional Q network over multiple training iterations using reinforcement-learning techniques. During this reinforcement-learning process, the network processes a current observation and possible actions using current network-parameter values, evaluates the possible Q returns, and selects an action based on the evaluation. Thus, Gendron-Bellemare teaches using reinforcement learning as a search and evaluation technique responsive to the current operating context. As established in the incorporated rejection of claim 1, Himayat teaches or at least suggests a context-specific target distribution of neural-network parameter values. A person of ordinary skill in the art would have understood that Gendron-Bellemare’s reinforcement-learning technique could be applied to search or evaluate candidate parameter values within Himayat’s context-specific target distribution to determine parameter values suitable for the target context. Accordingly, Gendron-Bellemare in view of Himayat teaches or at least suggests searching the target distribution to determine one or more neural-network parameter values based on the target context using a search algorithm comprising reinforcement learning, as claimed.)
Regarding claim 7 depends on claim 1, thus the rejection of claim 1 is incorporated.
Gendron-Bellemare in view of Himayat teaches or at least suggest “wherein selecting the one or more of the neural network parameter values includes: searching the target distribution for the one or more of the neural network parameter values based on a search cost function.” (Col. 2 - Lines 35-60 “For each action, the action and the next training observation are processed using a target distributional Q network (or, in some cases, the distributional Q network) and in accordance with current values of target network parameters of the distributional Q network to generate a next network output for the action-next training observation pair. The next network output defines a next probability distribution over possible Q returns for the action-next training observation pair. The target distributional Q network has the same neural network architecture as the distributional Q network but the current values of the target network parameters are different from the current values of the network parameters. For each action, a measure of central tendency of the possible Q returns with respect to the respective next probability distribution for the action-next training observation pair is determined. An argmax action is determined, where the argmax action is an action for which the measure of central tendency of the possible Q returns is highest. A respective projected sample update is determined for each of the possible Q returns using the current reward and the argmax action. A gradient is determined with respect to the network parameters of a loss function that depends on the projected sample updates for the possible Q returns and the current probabilities for the possible Q returns. The current values of the network parameters are updated using the gradient.” Gendron-Bellemare discloses determining a gradient with respect to the neural-network parameters of a loss function that depends on the projected Q-return updates and current Q-return probabilities, and updating the current network-parameter values using that gradient. A person of ordinary skill in the art would have understood the loss function to correspond to a search cost function and the gradient-based parameter update to correspond to searching for parameter values that reduce or optimize the search cost function. As established in the incorporated rejection of claim 1, Himayat teaches or suggests the context-specific target distribution of neural-network parameter values. Accordingly, in the combination, applying Gendron-Bellemare’s loss-function-guided parameter search to Himayat’s target parameter distribution teaches or at least suggests searching the target distribution to determine one or more neural-network parameter values based on a search cost function, as claimed.)
Regarding claim 8 depends on claim 1, thus the rejection of claim 1 is incorporated.
Gendron-Bellemare in view of Himayat teaches or at least suggest “wherein selecting the one or more of the neural network parameter values includes: searching the target distribution for the one or more of the neural network parameter values based on at least one of a safety metric associated with one or more searched neural network parameter values in the target distribution, a comfort metric associated with one or more searched neural network parameter values in the target distribution, and a performance metric associated with one or more searched neural network parameter values in the target distribution.” (Col. 2 - Lines 35-60 “For each action, the action and the next training observation are processed using a target distributional Q network (or, in some cases, the distributional Q network) and in accordance with current values of target network parameters of the distributional Q network to generate a next network output for the action-next training observation pair. The next network output defines a next probability distribution over possible Q returns for the action-next training observation pair. The target distributional Q network has the same neural network architecture as the distributional O network but the current values of the target network parameters are different from the current values of the network parameters. For each action, a measure of central tendency of the possible Q returns with respect to the respective next probability distribution for the action-next training observation pair is determined. An argmax action is determined, where the argmax action is an action for which the measure of central tendency of the possible Q returns is highest. A respective projected sample update is determined for each of the possible Q returns using the current reward and the argmax action. A gradient is determined with respect to the network parameters of a loss function that depends on the projected sample updates for the possible Q returns and the current probabilities for the possible Q returns. The current values of the network parameters are updated using the gradient.” Gendron-Bellemare discloses evaluating possible actions using measures of central tendency of the possible Q returns, selecting the action having the highest measure, determining projected sample updates using the current reward, and updating the neural-network parameter values based on a loss-function gradient. A person of ordinary skill in the art would have understood the Q-return, reward, and loss-function evaluation to correspond to a performance metric, because these values indicate how well the neural network and its associated parameter values perform in producing a desired result. The parameter values are then updated based on that performance evaluation. As established in the incorporated rejection of claim 1, Himayat teaches or suggests a context-specific target distribution of neural-network parameter values. Accordingly, applying Gendron-Bellemare’s performance-based evaluation and gradient update to the candidate parameter values within Himayat’s target distribution teaches or at least suggests searching the target distribution for one or more neural-network parameter values based on a performance metric associated with the searched parameter values, as claimed.)
Regarding claim 9 depends on claim 1, thus the rejection of claim 1 is incorporated.
Gendron-Bellemare in view of Himayat teaches or at least suggest “assigning a weight to each neural network parameter value in at least one of the exploration distribution and the target distribution based on a likelihood of usage of such neural network parameter value. (Col. 12 - Lines 35-37 “The system updates the current values of the distributional Q network parameters using the gradient of the loss function (316)”, and Col. 7 – Lines 31-37 “The system 100 selects an action 102 to be performed by the agent 104 at the time step based on the measures of central tendency 122 corresponding to the actions. In some implementations, the system 100 selects an action having a highest corresponding measure of central tendency 122 from amongst all the actions in the set of actions that can be performed by the agent 104.” Gendron-Bellemare discloses evaluating possible Q returns, selecting an action based on that evaluation, determining parameter-specific gradients of a loss function, and updating the neural-network parameters using the respective gradients. A person of ordinary skill in the art would have understood the gradients to correspond to weights reflecting the relative contribution and degree of use of the respective parameter values in producing the selected action. In view of Himayat’s target distribution of neural-network parameter values, the combination teaches or at least suggests assigning a weight to each parameter value based on a likelihood of usage of that parameter value, as claimed.)
Regarding claim 10 depends on claim 1, thus the rejection of claim 1 is incorporated.
Gendron-Bellemare in view of Himayat teaches or at least suggest “wherein the neural network parameter includes at least one of a layer width, a kernel size, a number of layers, a depth of layers, a learning rate, and a number of layers in a block.” (Col. 8 - Lines 38-41 “Generally, the target distributional O network 128 has the same neural network architecture (e.g., number of layers, neuron topology, and the like) as the distributional Q network 112”, and Col. 12 - Line 43 “where r is a positive learning rate hyper-parameter.” Gendron-Bellemare discloses that the distributional Q network and target distributional Q network have neural-network architectures defined by features such as a number of layers and neuron topology. Gendron-Bellemare further teaches using a learning-rate hyperparameter when updating the neural-network parameter values. Thus, the disclosed number of layers and learning rate correspond to at least one of the recited neural-network parameters.)
Regarding claim 12 depends on claim 11, thus the rejection of claim 11 is incorporated. Claim 12 is further rejected under the same rationale as claim 2. Applicant is directed to the rejection of claim 2 above because the claim recites similar limitation and concepts.
Regarding claim 13 depends on claim 11, thus the rejection of claim 11 is incorporated. Claim 13 is further rejected under the same rationale as claim 3. Applicant is directed to the rejection of claim 3 above because the claim recites similar limitation and concepts.
Regarding claim 16 depends on claim 11, thus the rejection of claim 11 is incorporated. Claim 16 is further rejected under the same rationale as claim 6. Applicant is directed to the rejection of claim 6 above because the claim recites similar limitation and concepts.
Regarding claim 17 depends on claim 11, thus the rejection of claim 11 is incorporated. Claim 17 is further rejected under the same rationale as claim 7. Applicant is directed to the rejection of claim 7 above because the claim recites similar limitation and concepts.
Regarding claim 18 depends on claim 11, thus the rejection of claim 11 is incorporated. Claim 18 is further rejected under the same rationale as claim 9. Applicant is directed to the rejection of claim 9 above because the claim recites similar limitation and concepts.
Regarding claim 19 depends on claim 11, thus the rejection of claim 11 is incorporated. Claim 19 is further rejected under the same rationale as claim 10. Applicant is directed to the rejection of claim 10 above because the claim recites similar limitation and concepts.
Regarding claim 20
Gendron-Bellemare teaches “A non-transitory computer-readable medium having stored thereon instructions which, when executed by one or more processors, cause the one or more processors” (Col. 14 - Lines 55-61 “Embodiments of the subject matter described in this specification can be implemented as one or more computer programs, i.e., one or more modules of computer program instructions encoded on a tangible non-transitory storage medium for execution by, or to control the operation of, data processing apparatus ... The term “data processing apparatus” refers to data processing hardware and encompasses all kinds of apparatus, devices, and machines for processing data, including by way of example a programmable processor”)
Claim 20 is further rejected under the same rationale as claim 1. Applicant is further directed to the rejection of claim 1 above because the claim recites similar limitation and concepts.
Claims 5, 15 are rejected under 35 U.S.C. 103 as being unpatentable over Gendron-Bellemare et.al (US 12056593 B2) in view of Himayat et.al (US 20210188306 A1), further in view of Mnih et al. (US 11783182 B2)
Regarding claim 5 depends on claim 1, thus the rejection of claim 1 is incorporated.
Gendron/Himayat does not teach the limitation “in response to determining that an amount of change in the context of the AV exceeds a threshold, updating a neural network of a stack of the AV with the selected one or more of the neural network parameter values” However, Mnih teaches or at least suggest this (Col. 7 - Lines 37-46 “the criteria can be the same as the criteria described above with reference to step 216, but with the specified threshold value or the specified number being greater than the value or number used for the determination of whether to update the parameters stored in the shared memory. Thus, the worker synchronizes the values of the parameters of the target network less frequently than the worker updates the parameter values stored in the shared memory.” Mnih discloses determining whether to update or synchronize the parameters of a target neural network based on a specified threshold criterion and, when the criterion is satisfied, updating the target-network parameters using parameter values stored in shared memory. As established in the incorporated rejection of claim 1, Gendron-Bellemare in view of Himayat teaches determining the AV’s operating context and selecting context-specific neural-network parameter values for adjusting the AV’s neural network. A person of ordinary skill in the art would have understood that Mnih’s threshold-triggered update technique could be applied to the detected change in the AV’s operating context so that the selected context-specific parameter values are implemented when the amount of context change exceeds the threshold. Accordingly, Gendron-Bellemare in view of Himayat and Mnih teaches or at least suggests, in response to determining that an amount of change in the AV context exceeds a threshold, updating the neural network of the AV stack using the selected neural-network parameter values, as claimed.)
Before the effective filing date, it would have been obvious to one of ordinary skill in the art to combine the teaching of distributional reinforcement learning, wherein autonomous vehicle operates in a specific environment using neural network model by Gendron-Bellemare, and the teaching of distributed learning to learn context-specific driving patterns by Himayat, with the teaching of threshold feature by Mnih. The motivation to do so is referred to in Mnih’s disclosure (Col. 7 - Lines 37-46 “For example, the criteria can be the same as the criteria described above with reference to step 216, but with the specified threshold value or the specified number being greater than the value or number used for the determination of whether to update the parameters stored in the shared memory. Thus, the worker synchronizes the values of the parameters of the target network less frequently than the worker updatesthe parameter values stored in the shared memory”. Mnih teaches that use of the threshold-controlled synchronization criterion causes the target-network parameters to be synchronized less frequently than the parameters stored in shared memory. In view of this teaching, a person of ordinary skill in the art would have been motivated to incorporate Mnih’s threshold-controlled updating technique into the context-specific parameter-update system of Gendron-Bellemare and Himayat to reduce unnecessary synchronization operations while permitting the AV neural network to be updated when a sufficiently meaningful change in context satisfies the threshold criterion)
Regarding claim 15 depends on claim 11, thus the rejection of claim 11 is incorporated. Claim 15 is further rejected under the same rationale as claim 5. Applicant is directed to the rejection of claim 5 above because the claim recites similar limitation and concepts.
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 DUY TU DIEP whose telephone number is (703)756-1738. The examiner can normally be reached M-F 8-4:30.
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, Alexey Shmatov can be reached at (571) 270-3428. 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.
/DUY T DIEP/ Examiner, Art Unit 2123
/ALEXEY SHMATOV/ Supervisory Patent Examiner, Art Unit 2123
Prosecution Insights