DETAILED ACTION
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 .
Claims 1-2, 4-5, 7-17 and 19-20 have been amended. Claims 1-20 have been examined.
Response to Arguments
Applicant's arguments filed 8/10/2026 have been fully considered but they are not persuasive. On p. 11 of the 8/10/2026 remarks, Applicant argues that Eisner’s prediction is limited to an utterance being spoken with respect to a “single predicted data dataflow program for a single speaker,” and therefore fails to disclose “a probability that the entity will experience a first future event at a first future time within a future time period after the time period,” as provided in claim 11. However, as noted by Applicant: “Eisner then teaches re-parsing "each prefix from scratch after each partial utterance is received.” Thus, a prediction/probability is made for each utterance received. With each partial utterance, Eisner is predicting a full utterance that hasn’t yet occurred. That is, the full utterance is associated with a future time period. This disclosure applies to the claimed limitation.
Applicant’s remaining arguments, see the send half of the last paragraph on p. 11 through p. 13, filed 8/10/2026, with respect to the rejection(s) of claim(s) 11 under 35 USC 102(a)(2) and claims 1 and 17 under 35 USC 103, 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 Therefore, the rejection has been withdrawn.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 U.S. Patent Application Publication 20240070079 by BenHanokh et al.
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.
Claim(s) 1-5, 7 and 11-20 is/are rejected under 35 U.S.C. 103 as being unpatentable over U.S. Patent Application Publication 20230367602 by Eisner et al. ("Eisner") in view of U.S. Patent Application Publication 20240070079 by BenHanokh et al. (“BenHanokh”).
In regard to claim 1, Eisner discloses:
1. A service provider system comprising: a non-transitory memory; and one or more hardware processors coupled to the non-transitory memory and configured to read instructions from the non-transitory memory to cause the service provider system to perform operations comprising: See Fig. 6 along with ¶ 0118-0121, e.g. “Storage subsystem 604 includes one or more physical devices configured to temporarily and/or permanently hold computer information such as data and instructions executable by the logic subsystem.”
obtaining input feature data for an entity, wherein the input feature data is associated with sequence-based data for neural network features over a past time period; Eisner ¶ 0028, “each training example including an example error condition, and an example data-flow program fragment for responding to the error. For example, previously-trained code-generation machine 104 may be trained with regard to a large plurality of annotated dialogue histories …” Also ¶ 0029, “In examples, a context-specific dialogue history includes a plurality of events arranged in a temporal order, e.g., time-stamped events, …”
executing a predictive call execution framework comprising a deep neural network model trained for forecasting of future events associated with the entity, Eisner, Fig. 1, element 104 along with ¶ 0019, “The previously-trained code-generation machine 104 may be based on any suitable technology, such as state-of-the-art or future machine learning (ML), artificial intelligence (AI), and/or natural language processing (NLP) technologies.” Also ¶ 0123, “Non-limiting examples of techniques that may be incorporated in an implementation of one or more machines include … multi-layer neural networks, … recurrent neural networks (e.g., long short-term memory networks), …”
wherein the deep neural network model is trained using training data associated with the neural network features for event sequences from past events associated with … the entity; Eisner ¶ 0029, “In examples, a context-specific dialogue history includes a plurality of events arranged in a temporal order, e.g., time-stamped events, …”
Eisner does not expressly disclose events associated with a plurality of entities including the entity. This is taught by BenHanokh. See BenHanokh ¶ 0026, “The cache manager 110 can include a prefetch manager 116, which can identify data items that may be requested by client computing device 102 or other requesting device at a future time, and cause the identified data items to be loaded from persistent storage and stored in the cache memory 134.”
Eisner also discloses:
encoding the input feature data as a sequence processable by the deep neural network model; Eisner, ¶ 0053, “As an example, the utterance prefix is part of a token sequence u=(u1, u2, . . . , u|u|) and the context is encoded as a sequence c=(c1, c2, . . . , c|c|).”
determining, using the deep neural network model and the sequence, a first predictive forecast comprising a probability that the entity will experience one of the future events within a future time period subsequent to the past time period; Eisner, ¶ 0045, “As an alternative, latency may be reduced by predicting and speculatively executing function calls while the user is still speaking based on partial results from automatic speech recognition (ASR) and the current state of execution.” Also ¶ 0047, “For either approach, subprograms may be selected for execution as soon as the semantic parser predicts that the subprogram has a high probability of being in the correct parse.” Also ¶ 0057, “In some examples, after each successive utterance prefix u[m], the complete graph G is predicted based on the current prefix u[m] and context c. As one example, the single most probable G is predicted. Additionally or alternatively, an estimate of the distribution p(G|c, u|m|) is constructed.”
Eisner does not expressly disclose the following limitations taught by BenHanokh:
determining, using the deep neural network model, additional predictive forecasts each comprising a respective probability that a corresponding other entity of the plurality of entities will experience the one of the future events within the future time period; BenHanokh ¶ 0026, “The cache manager 110 can include a prefetch manager 116, which can identify data items that may be requested by client computing device 102 or other requesting device at a future time, and cause the identified data items to be loaded from persistent storage and stored in the cache memory 134.” Also ¶ 0028, “The machine learning model 122 can be trained to generate one or more predicted data items 140 from a sequence of one or more data accesses represented by requested data items 108, where each of the predicted data items 140 is associated with a respective probability 142 generated by the machine learning model 122. The probability indicates a likelihood that the associated predicted data item 140 will be accessed subsequent to the sequence of requested data items 108.”
selecting a subset of the plurality of entities based on whether each of the probability and the respective probabilities meet or exceed a threshold, wherein the subset of the plurality of entities includes the entity; and BenHanokh ¶ 0020, “The probability criteria can be satisfied if, for example, the predicted data item is one of a threshold number T of predicted data items having the highest respective access probabilities in a group of predicted data items.” Also ¶ 0029, “For example, if one of the data items 140 is associated with a highest probability 142 that is greater than the probabilities of the other data items 140 in the group of predictions 124, then the data item 140 having the highest probability is most likely to be the next data item accessed. As another example, if each of the data items 140 is associated with the same probability, such as two data items each having 50% probability, then each data item is equally likely to be the next data item accessed.”
Eisner also discloses:
adding, based on the first predictive forecast, an application programming interface (API) call for data from an external computing service to a … processing event …, wherein the … processing event is executed prior to the future time period and the data from the external computing service is cached and usable during the one of the future events within the future time period. Eisner, ¶ 0012, “A storage server can provide access to a data store by receiving data access requests from clients …” ¶ 0017, “… invoking an API to perform an action using the API, e.g., ordering food from a restaurant, scheduling a ride with a ride-hailing service, scheduling a meeting in a calendar service, placing a phone call.” ¶ 0030, “The example data-flow program may include any suitable functions in any suitable sequence/arrangement, so that via training, the code-generation machine is configured to output suitable functions in a suitable sequence.” Also ¶ 0045, “As an alternative, latency may be reduced by predicting and speculatively executing function calls while the user is still speaking based on partial results from automatic speech recognition (ASR) and the current state of execution.”
Eisner does not expressly disclose: batch event that aggregates the API call with additional API calls for other entities of the subset of the plurality of entities. This is taught by BenHanokh. See BenHanokh ¶ 0016, “The storage server can more efficiently retrieve data items if the retrieval can be performed during idle periods in which the server is not processing other requests, or in batches, since batching retrieval of data items for retrieval can improve throughput, for example.” Also ¶ 0026, “The cache manager 110 can include a prefetch manager 116, which can identify data items that may be requested by client computing device 102 or other requesting device at a future time, and cause the identified data items to be loaded from persistent storage and stored in the cache memory 134.”
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use BenHanokh’s predictive batching service with Eisner’s model in order to improve performance as suggested by BenHanokh (see ¶ 0012).
In regard to claim 2, Eisner also discloses:
2. The service provider system of claim 1, wherein the deep neural network model comprises a long short-term memory (LSTM) architecture, gated recurrent unit (GRU) architecture, or other recurrent neural network (RNN) architecture trained using a plurality of event sequences encoded from occurrences of past events associated with the future events to be forecasted, and Also ¶ 0123, “Non-limiting examples of techniques that may be incorporated in an implementation of one or more machines include support vector machines, multi-layer neural networks, convolutional neural networks (e.g., including spatial convolutional networks for processing images and/or videos, temporal convolutional neural networks for processing audio signals and/or natural language sentences, and/or any other suitable convolutional neural networks configured to convolve and pool features across one or more temporal and/or spatial dimensions), recurrent neural networks (e.g., long short-term memory networks), …”
wherein the deep neural network model uses one-dimensional convolutional layers with the LSTM architecture, the GRU architecture, or the other RNN architecture. ¶ 0123, “… temporal convolutional neural networks …” Note that a temporal convolutional network utilizes 1D convolutional layers.
In regard to claim 3, Eisner also discloses:
3. The service provider system of claim 1, wherein the sequence-based data comprises time-based data of computing events executed by the entity over the past time period using a computing service of the service provider system provided to the entity. Eisner, ¶ 0017, “… invoking an API to perform an action using the API, e.g., ordering food from a restaurant, scheduling a ride with a ride-hailing service, scheduling a meeting in a calendar service, placing a phone call.”
In regard to claim 4, Eisner also discloses:
4. The service provider system of claim 1, wherein the operations further comprise: executing the batch processing event with the external computing service for the API call; storing, in a cache, the data from the external computing service for the one of the future events detecting a computing operation occurring that is associated with the one of the future events; retrieving the stored data; and loading the stored data to at least one downstream computing service during the one of the future events. Eisner, ¶ 0045, “As an alternative, latency may be reduced by predicting and speculatively executing function calls while the user is still speaking based on partial results from automatic speech recognition (ASR) and the current state of execution.” Also ¶ 0051, “the user is not presented any output until their utterance is complete.” Note that the output of speculative execution is first received and stored until an associated utterance is complete. Also see BenHanokh, Fig. 1 elements 110 and 130, depicting cached data.
In regard to claim 5, Eisner also discloses:
5. The service provider system of claim 1, wherein the input feature data comprises data corresponding to at least one string of computing events processed for the entity during the past time period, and Eisner, ¶ 0020, “Accordingly, the user utterance 102, data-flow program 106 generated for the user utterance 102, and/or any relevant assistant response 120 by the automated assistant may be stored in the context-specific dialogue history 130.”
wherein the encoding the input feature data comprises: encoding the data corresponding to the at least one string of computing events from at least one sequence of computing events occurring during the past time period and the neural network features. Eisner, ¶ 0019, “the previously-trained code-generation machine 104 includes an encoder machine configured to encode the user utterance 102 as a semantic feature, e.g., a vector in a semantic vector space learned by the previously-trained code-generation machine 104 and a decoder machine configured to decode the semantic feature by outputting one or more functions from the plurality of pre-defined functions 110.”
In regard to claim 7, Eisner also discloses:
7. The service provider system of claim 1, wherein the predictive call execution framework comprises one or more modules that are implemented with downstream services of the service provider system that provides stored data in real-time during the one of the future events within the future time period. Eisner, ¶ 0045, “As an alternative, latency may be reduced by predicting and speculatively executing function calls while the user is still speaking based on partial results from automatic speech recognition (ASR) and the current state of execution.” Also ¶ 0051, “the user is not presented any output until their utterance is complete.”
In regard to claim 11, Eisner discloses:
11. A method comprising: See e.g. Eisner Fig. 2, broadly depicting a method.
determining, using a deep neural network model of a predictive framework configured to enable sequence-based forecasting of future events using external data calls, a plurality of past events for an entity over a time period, wherein the plurality of past events include feature data for features processed by the deep neural network model; Eisner ¶ 0028, “each training example including an example error condition, and an example data-flow program fragment for responding to the error. For example, previously-trained code-generation machine 104 may be trained with regard to a large plurality of annotated dialogue histories …” Also ¶ 0029, “In examples, a context-specific dialogue history includes a plurality of events arranged in a temporal order, e.g., time-stamped events, …”
encoding the feature data for the plurality of past events as a sequence processable by the deep neural network model; Eisner, ¶ 0053, “As an example, the utterance prefix is part of a token sequence u=(u1, u2, . . . , u|u|) and the context is encoded as a sequence c=(c1, c2, . . . , c|c|).”
determining, using the deep neural network model and the sequence, a first predictive forecast comprising a probability that the entity will experience a first future event at a first future time within a future time period after the time period based on the feature data; Eisner, ¶ 0045, “As an alternative, latency may be reduced by predicting and speculatively executing function calls while the user is still speaking based on partial results from automatic speech recognition (ASR) and the current state of execution.” Also ¶ 0047, “For either approach, subprograms may be selected for execution as soon as the semantic parser predicts that the subprogram has a high probability of being in the correct parse.” Also ¶ 0057, “In some examples, after each successive utterance prefix u[m], the complete graph G is predicted based on the current prefix u[m] and context c. As one example, the single most probable G is predicted. Additionally or alternatively, an estimate of the distribution p(G|c, u|m|) is constructed.”
Eisner does not expressly disclose the following limitations taught by BenHanokh:
determining, using the deep neural network model, additional predictive forecasts each comprising a respective probability that a corresponding other entity of a plurality of entities including the entity will experience the first future event within the future time period; BenHanokh ¶ 0026, “The cache manager 110 can include a prefetch manager 116, which can identify data items that may be requested by client computing device 102 or other requesting device at a future time, and cause the identified data items to be loaded from persistent storage and stored in the cache memory 134.” Also ¶ 0028, “The machine learning model 122 can be trained to generate one or more predicted data items 140 from a sequence of one or more data accesses represented by requested data items 108, where each of the predicted data items 140 is associated with a respective probability 142 generated by the machine learning model 122. The probability indicates a likelihood that the associated predicted data item 140 will be accessed subsequent to the sequence of requested data items 108.”
selecting a subset of the plurality of entities based on whether each of the probability and the respective probabilities meet or exceed a threshold, wherein the subset of the plurality of entities includes the entity; BenHanokh ¶ 0020, “The probability criteria can be satisfied if, for example, the predicted data item is one of a threshold number T of predicted data items having the highest respective access probabilities in a group of predicted data items.” Also ¶ 0029, “For example, if one of the data items 140 is associated with a highest probability 142 that is greater than the probabilities of the other data items 140 in the group of predictions 124, then the data item 140 having the highest probability is most likely to be the next data item accessed. As another example, if each of the data items 140 is associated with the same probability, such as two data items each having 50% probability, then each data item is equally likely to be the next data item accessed.”
Eisner also discloses:
determining a first external data call required for the first future event based on at least one of the plurality of past events; executing the first external data call to a first external service in a … processing operation …, prior to the future time period, wherein the first external service includes data used during the first future event; Eisner, ¶ 0017, “… invoking an API to perform an action using the API, e.g., ordering food from a restaurant, scheduling a ride with a ride-hailing service, scheduling a meeting in a calendar service, placing a phone call.” ¶ 0030, “The example data-flow program may include any suitable functions in any suitable sequence/arrangement, so that via training, the code-generation machine is configured to output suitable functions in a suitable sequence.” Also ¶ 0045, “As an alternative, latency may be reduced by predicting and speculatively executing function calls while the user is still speaking based on partial results from automatic speech recognition (ASR) and the current state of execution.”
Eisner does not expressly disclose: batch event that aggregates the first external data call with additional external data calls for other entities of the subset of the plurality of entities … This is taught by BenHanokh. See BenHanokh ¶ 0016, “The storage server can more efficiently retrieve data items if the retrieval can be performed during idle periods in which the server is not processing other requests, or in batches, since batching retrieval of data items for retrieval can improve throughput, for example.” Also ¶ 0026, “The cache manager 110 can include a prefetch manager 116, which can identify data items that may be requested by client computing device 102 or other requesting device at a future time, and cause the identified data items to be loaded from persistent storage and stored in the cache memory 134.”
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use BenHanokh’s predictive batching service with Eisner’s model in order to improve performance as suggested by BenHanokh (see ¶ 0012).
Eisner also discloses:
receiving the data from the first external service based on the executing; and storing the data in a cache usable for the first future time of the first future event. Eisner, ¶ 0045, “As an alternative, latency may be reduced by predicting and speculatively executing function calls while the user is still speaking based on partial results from automatic speech recognition (ASR) and the current state of execution.” Also ¶ 0051, “the user is not presented any output until their utterance is complete.” Note that the output of speculative execution is first received and stored until an associated utterance is complete.
In regard to claim 12, Eisner also discloses:
12. The method of claim 11, wherein the deep neural network model is trained using a long short-term memory (LSTM) algorithm, a gated recurrent unit (GRU) algorithm, or other recurrent neural network (RNN) algorithm with one-dimensional convolutional layers for the sequence-based forecasting of the future events based on past sequences of events associated with past executions of the external data calls, and Eisner ¶ 0123, “Non-limiting examples of techniques that may be incorporated in an implementation of one or more machines include support vector machines, multi-layer neural networks, convolutional neural networks (e.g., including spatial convolutional networks for processing images and/or videos, temporal convolutional neural networks for processing audio signals and/or natural language sentences, and/or any other suitable convolutional neural networks configured to convolve and pool features across one or more temporal and/or spatial dimensions), recurrent neural networks (e.g., long short-term memory networks), …” Also ¶ 0123, “… temporal convolutional neural networks …” Note that a temporal convolutional network utilizes 1D convolutional layers.
wherein, prior to the determining the plurality of past events, the method further comprises: executing the predictive framework for the entity based on the entity utilizing a computing service associated with the first future event. Eisner ¶ 0029, “In examples, a context-specific dialogue history includes a plurality of events arranged in a temporal order, e.g., time-stamped events, …”
In regard to claim 13, Eisner also discloses:
13. The method of claim 11, wherein the batch processing operation comprises using at least a portion of the external data calls that are processed prior to the future events. Eisner, ¶ 0017, “… invoking an API to perform an action using the API, e.g., ordering food from a restaurant, scheduling a ride with a ride-hailing service, scheduling a meeting in a calendar service, placing a phone call.” ¶ 0030, “The example data-flow program may include any suitable functions in any suitable sequence/arrangement, so that via training, the code-generation machine is configured to output suitable functions in a suitable sequence.” Also ¶ 0045, “As an alternative, latency may be reduced by predicting and speculatively executing function calls while the user is still speaking based on partial results from automatic speech recognition (ASR) and the current state of execution.” Also see BenHanokh as cited above.
In regard to claim 14, Eisner also discloses:
14. The method of claim 13, further comprising: determining, using the deep neural network model, a second predictive forecast comprising a second probability that the entity will experience a second future event at a second future time within the future time period based on the feature data, wherein the second future event occurs after the first future event with a computing service provided to the entity; and determining a second external data call required for the second future event based on at least one of the plurality of past events, wherein the executing the batch processing operation includes executing the second external data call with the first external data call. See Fig. 2, element 215, e.g. “For Each Sequentially Received Utterance Prefix:” Also see ¶ 0047, “For either approach, subprograms may be selected for execution as soon as the semantic parser predicts that the subprogram has a high probability of being in the correct parse.” Also Fig. 5 and ¶ 0111, “At 515, actions are described that apply to each sequentially received utterance prefix.” Each sequential prefix corresponds to a future time with respect to a previous prefix. Also see BenHanokh, ¶ 0028 as cited above.
In regard to claim 15, Eisner also discloses:
15. The method of claim 11, further comprising: retrieving, at the first future time and from the cache, the stored data for the first future event; and loading the stored data during a data processing operation for the first future event. Also ¶ 0051, “the user is not presented any output until their utterance is complete.” Also see BenHanokh, ¶ 0026 as cited above.
In regard to claim 16, Eisner also discloses:
16. The method of claim 11, wherein prior to the determining the first predictive forecast, the method further comprises: determining a plurality of vectors from the sequence and additional sequences of the plurality of past events during different time intervals over the time period, wherein each of the plurality of vectors comprise encoded data identifying each of the sequences, and Eisner, ¶ 0019, “the previously-trained code-generation machine 104 includes an encoder machine configured to encode the user utterance 102 as a semantic feature, e.g., a vector in a semantic vector space learned by the previously-trained code-generation machine 104 and a decoder machine configured to decode the semantic feature by outputting one or more functions from the plurality of pre-defined functions 110.”
wherein the probability of the first predictive forecast is one of a vector or a score provided as an output from the deep neural network model that identifies a likelihood of occurrence of the first future event occurring. Eisner, ¶ 0098, “If the same invocation appears in multiple plans with different embeddings, these embeddings can be pooled before converting to a score.” Also ¶ 0134, “In any of the preceding examples, or any other example, the change in execution cost is additionally or alternatively based on one or more of an expected latency, a total computing cost, and a likelihood distribution indicating a probability that the program node will appear in a final dataflow program.”
In regard to claim 17, Eisner discloses:
17. A non-transitory machine-readable medium having stored thereon machine-readable instructions executable to cause a machine to perform operations comprising: See Eisner Fig. 6 element 604 and ¶ 0119, “Storage subsystem 604 includes one or more physical devices configured to temporarily and/or permanently hold computer information such as data and instructions executable by the logic subsystem.”
receiving training data for a recurrent neural network (RNN) model, wherein the training data comprising feature data for model features selected for the RNN model over a time period associated with a plurality of past computing events for an entity; Eisner ¶ 0028, “each training example including an example error condition, and an example data-flow program fragment for responding to the error. For example, previously-trained code-generation machine 104 may be trained with regard to a large plurality of annotated dialogue histories …” Also ¶ 0029, “In examples, a context-specific dialogue history includes a plurality of events arranged in a temporal order, e.g., time-stamped events, …” Also ¶ 0123, “Non-limiting examples of techniques that may be incorporated in an implementation of one or more machines include … recurrent neural networks (e.g., long short-term memory networks), …”
encoding the plurality of past computing events using the feature data as a sequence processable by the RNN model; Eisner, ¶ 0019, “the previously-trained code-generation machine 104 includes an encoder machine configured to encode the user utterance 102 as a semantic feature, e.g., a vector in a semantic vector space learned by the previously-trained code-generation machine 104 and a decoder machine configured to decode the semantic feature by outputting one or more functions from the plurality of pre-defined functions 110.” Also see Eisner, ¶ 0053, “As an example, the utterance prefix is part of a token sequence u=(u1, u2, . . . , u|u|) and the context is encoded as a sequence c=(c1, c2, . . . , c|c|).”
training the RNN model using the training data and the encoding of the plurality of past computing events, wherein the RNN model is trained to enable predictive forecasting of a future computing event; Eisner ¶ 0028, “each training example including an example error condition, and an example data-flow program fragment for responding to the error. For example, previously-trained code-generation machine 104 may be trained with regard to a large plurality of annotated dialogue histories …”
predicting, using the RNN model and the sequence, a probability that the entity will experience the future computing event within a future time period subsequent to the time period; Eisner, ¶ 0045, “As an alternative, latency may be reduced by predicting and speculatively executing function calls while the user is still speaking based on partial results from automatic speech recognition (ASR) and the current state of execution.” Also ¶ 0047, “For either approach, subprograms may be selected for execution as soon as the semantic parser predicts that the subprogram has a high probability of being in the correct parse.” Also ¶ 0057, “In some examples, after each successive utterance prefix u[m], the complete graph G is predicted based on the current prefix u[m] and context c. As one example, the single most probable G is predicted. Additionally or alternatively, an estimate of the distribution p(G|c, u|m|) is constructed.”
Eisner does not expressly disclose the following limitations taught by BenHanokh:
predicting, using the RNN model, respective probabilities that corresponding other entities of a plurality of entities will experience the future computing event within the future time period, wherein the plurality of entities includes the entity; BenHanokh ¶ 0026, “The cache manager 110 can include a prefetch manager 116, which can identify data items that may be requested by client computing device 102 or other requesting device at a future time, and cause the identified data items to be loaded from persistent storage and stored in the cache memory 134.” Also ¶ 0028, “The machine learning model 122 can be trained to generate one or more predicted data items 140 from a sequence of one or more data accesses represented by requested data items 108, where each of the predicted data items 140 is associated with a respective probability 142 generated by the machine learning model 122. The probability indicates a likelihood that the associated predicted data item 140 will be accessed subsequent to the sequence of requested data items 108.”
selecting a subset of the plurality of entities based on whether each of the probability and the respective probabilities meet or exceed a threshold, wherein the subset of the plurality of entities includes the entity; BenHanokh ¶ 0020, “The probability criteria can be satisfied if, for example, the predicted data item is one of a threshold number T of predicted data items having the highest respective access probabilities in a group of predicted data items.” Also ¶ 0029, “For example, if one of the data items 140 is associated with a highest probability 142 that is greater than the probabilities of the other data items 140 in the group of predictions 124, then the data item 140 having the highest probability is most likely to be the next data item accessed. As another example, if each of the data items 140 is associated with the same probability, such as two data items each having 50% probability, then each data item is equally likely to be the next data item accessed.”
Eisner also discloses:
executing a computing call for a data processing of the future computing event on behalf of the entity, wherein the executing comprises executing a … processing event … prior to the future time period; and Eisner, ¶ 0045, “As an alternative, latency may be reduced by predicting and speculatively executing function calls while the user is still speaking based on partial results from automatic speech recognition (ASR) and the current state of execution.” Eisner, ¶ 0012, “A storage server can provide access to a data store by receiving data access requests from clients …” ¶ 0017, “… invoking an API to perform an action using the API, e.g., ordering food from a restaurant, scheduling a ride with a ride-hailing service, scheduling a meeting in a calendar service, placing a phone call.” ¶ 0030, “The example data-flow program may include any suitable functions in any suitable sequence/arrangement, so that via training, the code-generation machine is configured to output suitable functions in a suitable sequence.” Also ¶ 0045, “As an alternative, latency may be reduced by predicting and speculatively executing function calls while the user is still speaking based on partial results from automatic speech recognition (ASR) and the current state of execution.”
Eisner does not expressly disclose: batch event that aggregates the computing call with additional computing calls for other entities of the subset of the plurality of entities … This is taught by BenHanokh. See BenHanokh ¶ 0016, “The storage server can more efficiently retrieve data items if the retrieval can be performed during idle periods in which the server is not processing other requests, or in batches, since batching retrieval of data items for retrieval can improve throughput, for example.” Also ¶ 0026, “The cache manager 110 can include a prefetch manager 116, which can identify data items that may be requested by client computing device 102 or other requesting device at a future time, and cause the identified data items to be loaded from persistent storage and stored in the cache memory 134.”
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use BenHanokh’s predictive batching service with Eisner’s model in order to improve performance as suggested by BenHanokh (see ¶ 0012).
preloading and caching, based on the executing the computing call, data for the data processing of the future computing event prior to an occurrence of the future computing event within the future time period. Eisner, ¶ 0045, “Identifying and pre-executing partial programs while the user is still speaking provides the technical benefit of expediting the final response to the user, thus improving user satisfaction due to improved computer-human interaction.” Also ¶ 0051, “the user is not presented any output until their utterance is complete.” Note that the output of speculative execution is first received and stored until an associated utterance is complete. Also see BenHanokh, Fig. 1 elements 110 and 130, depicting cached data.
In regard to claim 18, Eisner also discloses:
18. The non-transitory machine-readable medium of claim 17, wherein a predictive framework includes the RNN model and one or more operations that integrate outputs by the RNN model into a plurality of downstream services that utilize the data and additional data for predictive computing call executions by the predictive framework, and Eisner, ¶ 0045, “As an alternative, latency may be reduced by predicting and speculatively executing function calls while the user is still speaking based on partial results from automatic speech recognition (ASR) and the current state of execution.” Also ¶ 0051, “the user is not presented any output until their utterance is complete.”
wherein the RNN model comprises one of a long short-term memory (LSTM) model or a gated recurrent unit (GRU) model that use one-dimensional convolutional layers. Eisner, ¶ 0123, “… temporal convolutional neural networks … recurrent neural networks (e.g., long short-term memory networks).” Note that a temporal convolutional network utilizes 1D convolutional layers.
In regard to claim 19, Eisner also discloses:
19. The non-transitory machine-readable medium of claim 18, wherein the outputs comprise at least one of a regression output type or a classification output type associated with the future computing event, and Eisner, ¶ 0098, “Even if a function invocation is classified as positive, the actual invocation must be delayed until its arguments have been computed.”
wherein the outputs utilize at least one recurrent cell for the RNN model. ¶ 0123, e.g. “… recurrent neural networks (e.g., long short-term memory networks), …” Note that recurrent cells are elements of an RNN and are thereby an inherent element.
In regard to claim 20, Eisner also discloses:
20. The non-transitory machine-readable medium of claim 17, wherein the operations further comprise: detecting the occurrence of the future computing event; and loading the cached data from a local storage for a computing service requiring the data for the data processing. Eisner, ¶ 0045, “Identifying and pre-executing partial programs while the user is still speaking provides the technical benefit of expediting the final response to the user, thus improving user satisfaction due to improved computer-human interaction.” Also see Fig. 6 and ¶ 0115-0116 describing local implementation.
Claim(s) 6 is/are rejected under 35 U.S.C. 103 as being unpatentable over Eisner and BenHanokh as applied above, and further in view of U.S. Patent 9697524 to Fischer et al. ("Fischer").
In regard to claim 6, Eisner does not expressly disclose:
6. The service provider system of claim 1, wherein the future events are associated with one of a transaction processing request, an account authentication request, or a network token fetching request, and wherein the API call comprises one of a bank account balance inquiry from an open source banking data platform, a network token prefetch request from a network token service provider, an available fund balance inquiry, an authentication confirmation, or a risk analysis. This is taught by Fischer. See Fischer, col. 25, line 64 – col. 26 line 7, “Requests for information such as requests for the status of an application, account balance, status of a deposit, copies of media (e.g., image copies of checks), delivery of correspondence and statements, servicing of non-conforming images, and providing of contact and transaction history, may also be automatically processed. This category of service requests may also include service requests that contain minimum financial risks to the customer or the bank, such as those requests for adjustment of amounts posted for less than $200 due to misreads, unreadability or other related claims.” It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use Fischer’s service requests with Eisner’s API in order to automatically process financial services as suggested by Fischer.
Claim(s) 8-10 is/are rejected under 35 U.S.C. 103 as being unpatentable over Eisner and BenHanokh as applied above, and further in view of U.S. Patent Application Publication 20230326193 by Zia et al. ("Zia").
In regard to claim 8, Eisner also discloses:
8. The service provider system of claim 1, wherein prior to obtaining the input feature data, the operations further comprise: training the deep neural network model using a first portion of the training data associated with the neural network features and a long short-term memory (LSTM) architecture, a gated recurrent unit (GRU) architecture, or a recurrent neural network (RNN) architecture with one-dimensional convolutional layers; and Eisner ¶ 0028, “each training example including an example error condition, and an example data-flow program fragment for responding to the error. For example, previously-trained code-generation machine 104 may be trained with regard to a large plurality of annotated dialogue histories …” Also ¶ 0029, “In examples, a context-specific dialogue history includes a plurality of events arranged in a temporal order, e.g., time-stamped events, …”
Eisner does not expressly disclose:
testing the deep neural network model for deployment with the predictive call execution framework using a second portion of the training data. This is taught by Zia. See ¶ 0119, “As shown in FIG. 4G, some training methods may divide the training data into a training data subset, 435a, a validation data subset 435b, and a test data subset 435c. … Once the architecture has been trained, the method may assess the architecture's effectiveness by applying the architecture to all or a portion of the test data subsets 435c. … Testing at block 430c may be used to confirm the effectiveness of the trained architecture. Once the model is trained, inference 430d may be performed on a newly received inference input.” It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use Zia’s testing with Eisner’s model in order to confirm training effectiveness as suggested by Zia.
In regard to claim 9, Eisner also discloses:
9. The service provider system of claim 8, wherein the training and the testing are performed using the training data specific to at least one of the plurality of entities or an event type that is designated for the forecasting of the future events based on sequences of other events occurring in a temporal association with each of the other events. Eisner, ¶ 0020, “Accordingly, the context-specific dialogue history 130 defines a plurality of concepts, e.g., concepts 130A through 130N. “Concept” is used herein to refer to any relevant or potentially relevant aspects of the interaction between the user and the automated assistant. For example, a concept may include an entity (e.g., a person, place, thing, number, date), …”
In regard to claim 10, Eisner also discloses:
10. The service provider system of claim 8, wherein prior to the training, the operations further comprise: generating training sequences of events from the training data and the neural network features; and converting the training sequences to vectors, wherein the training is performed using at least the vectors. ¶ 0019, “In some examples, the previously-trained code-generation machine 104 includes an encoder machine configured to encode the user utterance 102 as a semantic feature, e.g., a vector in a semantic vector space learned by the previously-trained code-generation machine 104.”
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
U.S. Patent Application Publication 20180150125 by HomChaudhuri et al. See Fig. 1, elements 106A-106D along with ¶ 0049, “For example, there may be two types of devices: an access point (“APs”) 104 which communicates with clients (also referred to as stations, or “STAs”) 106.” Also see ¶ 0051, “STAs 106 may include, for example, a laptop computer 106c, a television 106b, a wireless device 106d, and a mobile communication device 106a.” Also ¶ 0141, “The above-described training and resulting transition probabilities models 746 relate primarily to dynamic pre-fetching of pages, that is, pre-fetching that occurs in response to current page requests or in-use pages.”
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 James D Rutten whose telephone number is (571)272-3703. The examiner can normally be reached M-F 9:00-5:30 ET.
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, Li B Zhen can be reached at (571)272-3768. 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.
/James D. Rutten/Primary Examiner, Art Unit 2121
Prosecution Insights