DETAILED ACTION
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
This final office action is in response to the amendment filed 10 March 2026.
Claims 1, 3-6, 8-9, 11-13, 15-16, and 18-21 are pending. Claims 1, 9, and 16 are independent claims. Claims 2, 7, 10, 14, and 17 are cancelled. Claim 21 is newly added.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 1, 4-6, 8-9, 12-13, 15-16, and 19-21 are rejected under 35 U.S.C. 103 as being unpatentable over Lei et al. (Time-Driven Feature-Aware Jointly Deep Reinforcement Learning for Financial Signal Representation and Algorithmic Trading, 2019, hereafter Lei) and further in view of Nagaraja (US 2018/0260692, published 13 September 2018) and further in view of Burhani et al. (US 11715017, filed 30 May 2019, hereafter Burhani) and further in view of Sanchez (US 11455524, filed 28 August 2018) and further in view of Tamaskar et al. (US 2024/0012400, filed 28 August 2020, hereafter Tamaskar) and further in view of Brim (Deep Reinforcement Learning Pairs Trading with a Double Deep Q-Network, 6-8 January 2020).
As per independent claim 1, Lei discloses a reinforcement machine learning system, comprising:
train a first model to approximate a state-action value function to estimate an expected cumulative return for an autonomous reinforcement learning agent to perform an action in a given state (Section 3.1, pages 3-6: Here, a first model is trained. This model includes historical stock trading data (page 3) and consists of temporal dimensions and feature dimensions (Figure 2; page 4). Further, a gate feature, used in both gated recurrent unit (GRU) and long short-term memory (LSTM) neural networks, is computed (pages 4-5). These features are used to create a final state vector (page 5))
train a second model to generate a simulated experience, the second network trained to predict a simulated state at a next time step after performing a given action, the second model being trained using real experience in a real environment (Section 3.2, pages 6-7: Here, a jointly supervised deep and reinforcement learning method is defined. The framework includes predicting a closing price (generating a simulated experience). This prediction is compared to the actual closing price and used to update parameters of the deep learning model (pages 6-7). This allows for training the model using real environmental data (actual closing price))
a deep Q-learning network (Section 2, pages 2-3)
However, Lei fails to specifically disclose:
a processor
a memory device coupled with the processor
the processor configured to at least:
execute an autonomous reinforcement learning agent, the reinforcement learning agent configured to uniformly explore an action by uniformly sampling an action from all possible remaining action state space combinations and performing the sampled action, the reinforcement learning agent further being configured to perform a selected action selected by a first neural network given a current state of a real environment
neural network
predict a simulated state at a next time step after performing a given action without predicting a reward associated with the simulated state at the next time step, wherein a one-step simulated experience is generated starting from a real state
the first neural network being trained based on the simulated experience generated by the second neural network and a real experience from a real environment
the first neural network and the second neural network being trained simultaneously and in parallel in the autonomous reinforcement agent
wherein the first neural network includes a deep double Q-learning network
However, Nagaraja, which is analogous to the claimed invention because it is directed toward a reinforcement learning agent, discloses:
a processor (paragraphs 0037 and 0057-0058)
a memory device coupled with the processor (Figure 4; paragraph 0135)
a processor configured to at least perform functions (paragraphs 0057-0058)
execute an autonomous reinforcement learning agent (paragraph 0037: Here, a reinforcement learning agent is created and used), the reinforcement learning agent configured to uniformly explore an action by uniformly sampling an action from all possible remaining action state space combinations (paragraph 0045: Here, the state action space combinations are explored based upon a trigger causing the reinforcement learning agent to explore the reinforcement learning environment)
neural network (paragraph 0065: Here, the reinforcement learning processor uses neural network data paths to communicate with a neural network which uses the actions, state-value functions, Q-values, and reward values generated by the reinforcement learning processor to approximate an optimal state-value function as well as an optimal reward function)
training a first neural network and a second neural network simultaneously and in parallel (paragraph 0057: Here, application-domain specific instructions sets are single instruction multiple agent based instructions. These instructions may be implemented in parallel by the reinforcement learning agent to train the reinforcement learning agents)
It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Nagaraja with Lei, with a reasonable expectation of success, as it would have allowed for training and implementing reinforcement learning agents using single instruction multiple agents to perform simultaneous instructions to leverage parallelism in reinforcement learning operations (Nagaraja: Abstract; paragraph 0009).
However, Burhani, which is analogous to the claimed invention because it is directed toward training a neural network to perform trades, discloses:
the agent is configured to perform a selected action selected by the first neural network given a current state of the real environment (column 2, lines 65-67: Here, trade orders (selected action) are performed based upon output from the neural network)
training in the autonomous reinforcement agent (column 6, lines 1-18)
It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Burnhani with Lei-Nagaraja, with a reasonable expectation of success, as it would have allowed for using a deep learning neural network to identify trades. This would have allowed a user to identify trends and place trades to improve financial gains.
Additionally, Sanchez, which is analogous to the claimed invention because it is directed toward improving predicted models, discloses predict a simulated state at a next time step after performing a given action without predicting a reward associated with the simulated state at the next time step, wherein a one-step simulated experience is generated starting from a real state (column 2, line 57- column 3, line 35: Here, a real world labeled training dataset is received (initial data set including a first plurality of facts) and used for generating an unlabeled simulate (predicted) data set. This simulated dataset is used for training a neural network to include a plurality of intermediate predictions). It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Sanchez with Lei-Nagaraja-Burnhani, with a reasonable expectation of success, as it would have allowed for training a network with the combination of real world and simulated data, thereby improving the accuracy of the model (Sanchez: column 28-35).
Next, Tamaskar, which is analogous to the claimed invention because it is directed toward using simulated experiences to train a neural network, discloses the first neural network being trained based on the simulated experience and a real experience from a real environment (Figure 4; paragraphs 0014 and 0034: Here, a set of simulated parameters are used to train a neural network). It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Tamaskar with Lei-Nagaraja-Burnhani-Sanchez, with a reasonable expectation of success, as it would have allowed for training using historical training data in combination with generated scenarios. This would have provided a user with a more robust neural network.
Further, Brim, which is analogous to the claimed invention because it is directed toward using a double deep Q-learning network for trading, discloses wherein the first neural network includes a deep double Q-learning network (Section I: Here, a double deep Q-learning network is used to decorrelate training samples, reduce error, and achieve better performance by extracting features with a specific trading strategy and pairs trading in mind). It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Brim with Lei-Nagaraja-Burnhani-Sanchez-Tamaskar, with a reasonable expectation of success, as it would have provided improved performance by reducing error.
As per dependent claim 4, Lei, Nagaraja, Burnhani, Sanchez, Tamaskar, and Brim disclose the limitations similar to those in claim 1, and the same rejection is incorporated herein. Tamaskar disclose wherein the processor is configured to interleave using of the simulated experiences generated by the neural network and the real experience from the real environment (Figure 4; paragraph 0033: Here, both simulated parameters and real-world historical data is used to train the neural network). It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Tamaskar with Lei-Nagaraja-Burnhani-Sanchez-Tamaskar-Brim, with a reasonable expectation of success, as it would have allowed for training using historical training data in combination with generated scenarios. This would have provided a user with a more robust neural network.
As per dependent claim 5 Lei, Nagaraja, Burnhani, Sanchez, Tamaskar, and Brim disclose the limitations similar to those in claim 4, and the same rejection is incorporated herein. Lei discloses performing multiple updates using the real experience received from the real environment (Section 1, pages 1-2: Here, a framework to jointly construct and iteratively train the supervised deep learning model and the reinforcement learning model is disclosed).
Lei fails to specifically disclose wherein the interleaving includes using the simulated experience generated by the neural network and updating using the real experience received from the real environment. However, Tamaskar disclose wherein the interleaving includes using the simulated experience generated by the neural network and updating using the real experience received from the real environment (Figure 4; paragraph 0033: Here, both simulated parameters and real-world historical data is used to train the neural network). It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Tamaskar with Lei-Nagaraja-Burnhani-Sanchez-Tamaskar-Brim, with a reasonable expectation of success, as it would have allowed for training using historical training data in combination with generated scenarios. This would have provided a user with a more robust neural network.
As per dependent claim 6, Lei, Nagaraja, Burnhani, Sanchez, Tamaskar, and Brim disclose the limitations similar to those in claim 1, and the same rejection is incorporated herein. Lei discloses wherein the first neural network includes a deep Q-learning network and the state-action value function includes a Q-function (Section 2, pages 2-3: Here, a deep Q-learning network (DQN) utilizes the advantages of deep learning to learn distributed features for the original financial signals layer by layer, and serves Q learning network to improve the accuracy of the training model).
As per dependent claim 8, Lei, Nagaraja, Burnhani, Sanchez, Tamaskar, and Brim disclose the limitations similar to those in claim 1, and the same rejection is incorporated herein. Burnhani discloses wherein the action includes buying and selling a security share in order completion (column 2, lines 65-67: Here, trade orders (selected action) are performed based upon output from the neural network)
It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Burnhani with Lei-Nagaraja-Burnhani-Sanchez-Tamaskar-Brim, with a reasonable expectation of success, as it would have allowed for using a deep learning neural network to identify trades. This would have allowed a user to identify trends and place trades to improve financial gains.
With respect to claims 9,12-13, and 15, the applicant discloses the limitations substantially similar to those in claims 1, 4, 6, and 8, respectively. Claims 9 and 12-13 are similarly rejected.
With respect to claims 16 and 19-21, the applicant discloses the limitations substantially similar to those in claims 1, 4, 6 and 8, respectively. Claims 16 and 19-21 are similarly rejected.
Claims 3, 11, and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Lei, Nagaraja, Burnhani, Sanchez, Tamaskar, and Brim and further in view of Wulfmeier et al. (WO 2020/239841, published 3 December 2020, hereafter Wulfmeier).
As per dependent claim 3, Lei, Nagaraja, Burnhani, Sanchez, Tamaskar, and Brim disclose the limitations similar to those in claim 1, and the same rejection is incorporated herein. Lei discloses wherein the processor is further configure to retrain the first model (Section 1, pages 1-2: Here, a framework to jointly construct and iteratively train the supervised deep learning model and the reinforcement learning model is disclosed) using as additional training data a reward associated with the sampled action received from the environment (Section 3.2.1, page 6: Here, the total income is used as the reward function of the reinforcement learning model).
Lei fails to specifically disclose wherein the data includes the sampled action and a state of the real environment after the sampled action is taken. However, Wulfmeier, which is analogous to the claimed invention because it is directed toward selection actions, discloses wherein the data includes the sampled action and a state of the real environment after the sampled action is taken (page 4, lines 26-28: Here, the system controls an agent interacting with the environment to select actions to be performed by the agent in response to observations that cause the agent to perform the selected action).
It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Wulfmeier with Lei-Nagaraja-Burnhani -Sanchez-Tamaskar-Brim, with a reasonable expectation of success, as it would have allowed for sampling possible action-state pairs to identify the best action to perform based upon the environment. This would have allowed for retraining the model with the best data in order to improve task selection.
With respect to claim 11, the applicant discloses the limitations substantially similar to those in claim 3. Claim 11 is similarly rejected.
With respect to claims 18, the applicant discloses the limitations substantially similar to those in claim 3. Claim 18 is similarly rejected.
Response to Arguments
Applicant’s arguments have been fully considered and are persuasive. Therefore, the rejection has been withdrawn. However, upon further consideration, a new ground(s) of rejection is made in view of Lei, Nagaraja, Burnhani, Sanchez, Tamaskar, and Brim.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure:
Soleyman et al. (US 12061673): Discloses a reinforcement learning agent using a deep learning network to produce actions corresponding to each controlled platform and obtaining a reward value (Abstract)
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 KYLE R STORK whose telephone number is (571)272-4130. The examiner can normally be reached 8am - 2pm; 4pm - 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, Omar Fernandez Rivas can be reached at 571/272-2589. 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.
/KYLE R STORK/Primary Examiner, Art Unit 2128
Prosecution Insights