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 .
Specification
The title of the invention is not descriptive. The title only states a data management system and method, which could refer to any computer system and method. A new title is required that is clearly indicative of the invention to which the claims are directed.
The following title is suggested: Compressing Old Machine Learning Dataset Data for Use by a Human.
Claim Rejections - 35 USC § 112
Claims 1-7 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Regarding Claim 1, it recites “a compressor/decompressor for human” (line 4). The term “human” lacks an article such as “a.” It is also unclear what this element is because it does not recite what it means for a compressor/decompressor to be “for human.” What makes data compression/decompression “for human”? Line 5 recites “by human,” which is confusing because it too lacks an article such as “a” or “the.” It is unclear what it means for verification to be performed “by human,” and it cannot be determined if “human” recited by line 5 is the same human as recited by line 3. The last line again recites “for human,” which is indefinite as described above. For the purposes of examination under prior art, the examiner can only interpret the terms “for human” and “by human” in a general sense to mean that humans are associated with the system in some way, as is necessarily the case for systems involving computers.
Regarding Claim 2, it recites “a compressor/decompressor for model” (line 3). Similarly to claim 1, the term “model” lacks an article such as “a.” It is also unclear what this element is because it does not recite what it means for a compressor/decompressor to be “for model.” Line 6 recites “the data used for learning and inference.” However, line 4 recites “data to be usable for learning and inference.” It is unclear if “the data used for learning and inference” refers to the same element as “data to be usable for learning and inference” because the wording differs. Finally, line 9 again recites “for model,” which lacks an article such as “the” and is indefinite as described above. For the purposes of examination under prior art, the examiner can only interpret the term “for model” in a general sense to mean that a compressor/decompressor is associated with a model in some way.
Regarding Claim 3, it recites “the compressor/decompressor for human” (line 3), which is indefinite as described for claim 1.
Regarding Claim 4, it recites “the compressor/decompressor for model” (line 3), which is indefinite as described for claim 2.
Regarding Claim 5, it recites “the compressor/decompressor for human” (line 2), which is indefinite as described for claim 1. In addition, lines 4-5 recite “a genuine one and a false one.” It is unclear what “one” refers to—a genuine what and a false what? And how does “a difficulty of distinguishing” get turned into an image quality evaluation index? Since so much of the present claim is indefinite, the examiner is unable to make a meaningful interpretation with respect to prior art.
Regarding Claim 6, it recites “the compressor/decompressor for model” (line 3), which is indefinite as described for claim 2. The present claim also recites “the accuracy” (lines 2-3). This term lacks antecedent basis because there has been no previous mention of accuracy. It is also unclear what “an image quality evaluation index” means in terms of the learning of (the) model; the index is never described, and the learning process has not been recited, so it cannot be determined what the evaluation index means or how “the accuracy” is used as an index. For all of these reasons, the examiner is unable to make a meaningful interpretation with respect to prior art.
Regarding Claim 7, it recites elements substantially similar to those of claim 1, so it is indefinite for the same reasons.
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 and 7 are rejected under 35 U.S.C. 103 as being unpatentable over icYou520 et al., AI Stack Exchange, “What happens to the training data after your machine learning model has been trained?”, 2018 (obtained from <<https://ai.stackexchange.com/questions/7739/what-happens-to-the-training-data-after-your-machine-learning-model-has-been-tra>> on 31 Dec. 2025 “icYou520”) in view of Mandrychenko (US 2020/0287920).
Regarding claim 1, icYou520 teaches wherein the processor specifies data that is no longer used for learning and inference in the machine learning model among the data stored in the storage device, and (icYou520, pp. 1 - 2, “In many cases, a production-ready model has everything it needs to make predictions without retaining training data {data that is no longer used for learning}. For example: a linear model might only need the coefficients, a decision tree just needs rules/splits, and a neural network needs architecture and weights. The training data isn’t required as all the information needed to make a prediction is incorporated into the model. … Having said that, where possible the training data would be retained. If additional data is received, a new model can be trained on the enlarged dataset. If it is decided a different approach is required, or if there are concerns about concept drift, then it’s good to have the original data still on hand. In many cases, the training data might comprise personal data or make a company’s competitive advantage, so the model and the data should stay separate.”; icYou520 teaches embodiments of models whose training data is not needed by a model after training, recommending that data should be retained “where possible”).
icYou520 does not explicitly teach:
A data management system including a processor connected to a storage device that stores data usable for learning and inference in a machine learning model, the system comprising:
a compressor/decompressor for human that compresses and decompresses data to be usable for verification performed by human, wherein
the processor compresses the specified data using the compressor/decompressor for human.
However, Mandrychenko teaches:
A data management system including a processor connected to a storage device that stores data usable for learning and inference in a machine learning model, the system comprising: a compressor/decompressor for human that compresses and decompresses data to be usable for verification performed by human, (¶ [0069]—“In an embodiment, data aggregation module 215 can perform a time-based data aggregation on the collected network metadata to reduce transmission bandwidth and local storage requirements of the endpoint device. The collected network metadata can be aggregated on the endpoint using a predetermined or configurable fixed time interval (e.g., 1 minute, 5 minutes, 10 minutes, etc.). For example, within the fixed time interval all network communication metadata from the same device and user and application and sent to the same destination and host and port and protocol can be aggregated into one metadata transfer object with the count for bytes sent and received updated during the fixed time interval. In one embodiment, when the current time interval expires, data aggregation module 215 can create a data transfer object, compresses the data transfer object and writes the data transfer object to a database on the endpoint device [a compressor/decompressor for human that compresses and decompresses data]. When a network connection to the gateway (e.g., gateway device 104) associated with the cloud-based endpoint traffic analysis service is available (as represented by the endpoint device being connected to the enterprise network), the aggregated data can be sent via an API provided by the gateway device using an encrypted channel, for example, using Hyper Text Transfer Protocol Secure (HTTPS).”)
wherein a processor compresses the specified data using the compressor/decompressor for human (¶ [0106] and [0107]—“[0106] According to an implementation, data sets can be required to be split into training and detection sets. Initial training data set can be decided by a user or an external service, which can represent normal traffic to train the endpoint traffic analysis service to detect anomalies in normal traffic. The dataset can also be a representative sample of an attack, where the service can be trained to spot attacks in the network traffic. Training can run on a cluster of graphics processing units (GPUs) and can be integrated with cloud. The data can be fed to the cluster using API gateway and integration service. [0107] Once training stage is complete, the data can be saved in a compressed binary format using the storage service. Anomaly detection service can schedule a run on a time interval using the latest models and/or some specific models to detect network attacks. The output of the anomaly detection service can be a decision indicating whether the new data is representative of normal traffic or representative of anomalous or risky network behavior. Any detected anomalies can be saved into the database together with the context regarding when and how the anomaly was generated for further reference.”).
In view of the teachings of Mandrychenko it would have been obvious for a person of ordinary skill in the art to apply the teachings of Mandrychenko to icYou520 before the effective filing date of the claimed invention in order to minimize storage space utilize fewer computing resources for stored data. icYou520 teaches embodiments of models whose training data is not needed by a model after training, and suggests that data should be retained “where possible”. Mandrychenko teaches storing data utilized after training in a compressed format. Mandrychenko notes that it is desirable to minimize storage requirements (cf. paragraph 0007, “The agent reduces transmission bandwidth and local storage requirements”).
Regarding claim 7, icYou520 teaches causing the processor to specify data that is no longer used for learning and inference in the machine learning model among the data stored in the storage device, and (icYou520, pp. 1 - 2, “In many cases, a production-ready model has everything it needs to make predictions without retaining training data {data that is no longer used for learning}. For example: a linear model might only need the coefficients, a decision tree just needs rules/splits, and a neural network needs architecture and weights. The training data isn’t required as all the information needed to make a prediction is incorporated into the model. … Having said that, where possible the training data would be retained. If additional data is received, a new model can be trained on the enlarged dataset. If it is decided a different approach is required, or if there are concerns about concept drift, then it’s good to have the original data still on hand. In many cases, the training data might comprise personal data or make a company’s competitive advantage, so the model and the data should stay separate.”; icYou520 teaches embodiments of models whose training data is not needed by a model after training, recommending that data should be retained “where possible”).
icYou520 does not explicitly teach:
A data management method in a data management system including a processor connected to a storage device that stores data usable for learning and inference in a machine learning model, the data management system including a compressor/decompressor for human that compresses and decompresses data to be usable for verification performed by human, the method comprising:
compress the specified data using the compressor/decompressor for human..
However, Mandrychenko teaches:
A data management method in a data management system including a processor connected to a storage device that stores data usable for learning and inference in a machine learning model, the data management system including a compressor/decompressor for human that compresses and decompresses data to be usable for verification performed by human, the method comprising: (¶ [0069]—“[0069] In an embodiment, data aggregation module 215 can perform a time-based data aggregation on the collected network metadata to reduce transmission bandwidth and local storage requirements of the endpoint device. The collected network metadata can be aggregated on the endpoint using a predetermined or configurable fixed time interval (e.g., 1 minute, 5 minutes, 10 minutes, etc.). For example, within the fixed time interval all network communication metadata from the same device and user and application and sent to the same destination and host and port and protocol can be aggregated into one metadata transfer object with the count for bytes sent and received updated during the fixed time interval. In one embodiment, when the current time interval expires, data aggregation module 215 can create a data transfer object, compresses the data transfer object and writes the data transfer object to a database on the endpoint device [a compressor/decompressor for human that compresses and decompresses data]. When a network connection to the gateway (e.g., gateway device 104) associated with the cloud-based endpoint traffic analysis service is available (as represented by the endpoint device being connected to the enterprise network), the aggregated data can be sent via an API provided by the gateway device using an encrypted channel, for example, using Hyper Text Transfer Protocol Secure (HTTPS).”), and
compress the specified data using the compressor/decompressor for human (¶ [0106] and [0107]—“[0106] According to an implementation, data sets can be required to be split into training and detection sets. Initial training data set can be decided by a user or an external service, which can represent normal traffic to train the endpoint traffic analysis service to detect anomalies in normal traffic. The dataset can also be a representative sample of an attack, where the service can be trained to spot attacks in the network traffic. Training can run on a cluster of graphics processing units (GPUs) and can be integrated with cloud. The data can be fed to the cluster using API gateway and integration service. [0107] Once training stage is complete, the data can be saved in a compressed binary format using the storage service. Anomaly detection service can schedule a run on a time interval using the latest models and/or some specific models to detect network attacks. The output of the anomaly detection service can be a decision indicating whether the new data is representative of normal traffic or representative of anomalous or risky network behavior. Any detected anomalies can be saved into the database together with the context regarding when and how the anomaly was generated for further reference.”).
In view of the teachings of Mandrychenko it would have been obvious for a person of ordinary skill in the art to apply the teachings of Mandrychenko to icYou520 before the effective filing date of the claimed invention in order to minimize storage space utilize fewer computing resources for stored data. icYou520 teaches embodiments of models whose training data is not needed by a model after training, and suggests that data should be retained “where possible”. Mandrychenko teaches storing data utilized after training in a compressed format. Mandrychenko notes that it is desirable to minimize storage requirements (cf. paragraph 0007, “The agent reduces transmission bandwidth and local storage requirements”).
Claims 2-3 are rejected under 35 U.S.C. 103 as being unpatentable over icYou520 in view of Mandrychenko, as applied to claim 1, above, and further in view of Zhang, Zhao, et al. (“Efficient I/O for neural network training with compressed data,” 2020 IEEE International Parallel and Distributed Processing Symposium (IPDPS). IEEE, 2020; hereinafter “Zhang”).
Regarding Claim 2, icYou520/Mandrychenko does not specifically teach a compressor/decompressor for model that compresses and decompresses data to be usable for learning and inference in the machine learning model, wherein the data used for learning and inference in the machine learning model stored in the storage device is data obtained by compressing data, which is acquired from a data generation source, using the compressor/decompressor for model by the processor and stored in the storage device.
However, Zhang teaches a compressor/decompressor for model that compresses and decompresses data to be usable for learning and inference in the machine learning model, wherein the data used for learning and inference in the machine learning model stored in the storage device is data obtained by compressing data, which is acquired from a data generation source, using the compressor/decompressor for model by the processor and stored in the storage device (Abstract and p. 410, first paragraph—a compressor/decompressor compresses data to be usable for training of deep learning {DL} models. See also section IV. B for a description of the compression and IV. C for accessing and storing compressed data).
All of the claimed elements were known in icYou520/Mandrychenko and Zhang and could have been combined by known methods with no change in their respective functions. It therefore would have been obvious to a person of ordinary skill in the art at the time of filing of the applicant’s invention to combine the compressor/decompressor for model of Zhang with the system of icYou520/Mandrychenko to yield the predictable result of a compressor/decompressor for model that compresses and decompresses data to be usable for learning and inference in the machine learning model, wherein the data used for learning and inference in the machine learning model stored in the storage device is data obtained by compressing data, which is acquired from a data generation source, using the compressor/decompressor for model by the processor and stored in the storage device. One would be motivated to make this combination for the purpose of facilitating training while meeting I/O throughput needs, as described by Zhang, section I.
Regarding Claim 3, icYou520/Mandrychenko/Zhang teaches wherein the data generated by the data generation source is an image (section II. C—compression formats may be JPEG2000 or TIFF, which are for images. Section VII. B describes datasets of images), and the compressor/decompressor for human is a neural network that has been learned with an emphasis on a compression ratio of an image rather than a period of time required for compression and decompression of an image while satisfying image quality that is able to withstand verification performed by human (Abstract and p. 410, second of the three contributions—a compression algorithm is selected to provide the highest possible storage capacity, i.e. with an emphasis on compression ratio rather than time required to perform compression).
Claim 4 is rejected under 35 U.S.C. 103 as being unpatentable over icYou520 in view of Mandrychenko in view of Zhang, as applied to claim 2, above, and further in view of Johnston, Nick, et al. (“Computationally efficient neural image compression,” arXiv preprint arXiv:1912.08771 (2019); hereinafter “Johnston”).
Regarding Claim 4, icYou520/Mandrychenko teaches wherein the data generated by the data generation source is an image (section II. C—compression formats may be JPEG2000 or TIFF, which are for images. Section VII. B describes datasets of images), but does not specifically teach the compressor/decompressor for model is a neural network that has been learned with an emphasis on a period of time required for compression and decompression of an image rather than a compression ratio of an image while satisfying image quality that is able to withstand learning and inference in the machine learning model.
However, Johnston teaches a compressor/decompressor is a neural network that has been learned with an emphasis on a period of time required for compression and decompression of an image rather than a compression ratio of an image while satisfying image quality that is able to withstand learning and inference in the machine learning model (Abstract and section 1—a neural network is designed and trained to perform image compression while reducing run-time at no performance loss {i.e. with an emphasis on a period of time required for compression and decompression of an image rather than a compression ratio of an image while satisfying image quality}. Maintaining image quality indicates that the images are able to withstand learning and inference in the machine learning model. Section 4.1 measures inference performance {period of time to perform compression and decompression}).
All of the claimed elements were known in icYou520/Mandrychenko/Zhang and Johnston and could have been combined by known methods with no change in their respective functions. It therefore would have been obvious to a person of ordinary skill in the art at the time of filing of the applicant’s invention to combine the neural network performing image compression and decompression of Johnston with the compressor/decompressor of icYou520/Mandrychenko/Zhang to yield the predictable result of the data generated by the data generation source is an image, and the compressor/decompressor for model is a neural network that has been learned with an emphasis on a period of time required for compression and decompression of an image rather than a compression ratio of an image while satisfying image quality that is able to withstand learning and inference in the machine learning model. One would be motivated to make this combination for the purpose of enabling high-performance decoding to account for images being decompressed many times on compute constrained devices (Johnston, section 1).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. This art includes:
Singh et al. (U.S. Patent 10,715,176) teaches using machine learning to compress data to meet user requirements, estimating compression time and a compression ratio
Dejean-Servières, Mathieu, et al. (“Study of the impact of standard image compression techniques on performance of image classification with a convolutional neural network,” Diss. INSA Rennes; Univ Rennes; IETR; Institut Pascal, 2017) compares performance of machine learning classifiers using different types and levels of image compression
Castelli, Vittorio, et al. (“Progressive search and retrieval in large image archives,” IBM Journal of Research and Development 42.2 (1998): 253-268) teaches compressing images for searching in a database, thus compressing data for use by a human
Kuchnik, Michael, George Amvrosiadis, and Virginia Smith. (“Progressive compressed records: Taking a byte out of deep learning data,” arXiv preprint arXiv:1911.00472 (2019)) teaches compressing a dataset to multiple size and quality levels, and determining which compression level is best for training a particular machine learning model
Calhoun, S. Patrick, et al. (“Large scale research data archiving: Training for an inconvenient technology,” Journal of Computational Science 36 (2019): 100523) teaches archiving research datasets to less expensive storage using compression, thus compressing data for use by humans doing research
Any inquiry concerning this communication or earlier communications from the examiner should be directed to HAL W SCHNEE whose telephone number is (571) 270-1918. The examiner can normally be reached M-F 7:30 a.m. - 6:00 p.m.
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, Michael Huntley can be reached at 303-297-4307. 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.
/HAL SCHNEE/ Primary Examiner, Art Unit 2129