# US 2023/0066478 A1 · Method and system for learning behavior of highly complex and non-linear systems > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 2023/0066478 A1 - Document ID: US20230066478A1 - Status: Published application - Kind: A1 - Application number: 17/822,013 - Filed: 2022-08-24 - Published: 2023-03-02 - Earliest claimed: 2021-08-24 - Assignees: Jio Platforms Limited - Applicants: Jio Platforms Limited - Inventors: Shailesh KUMAR; Palash SETHI - Web page: https://patentgrep.com/patent/US20230066478A1 ## CPC - G06N 3/047: Probabilistic or stochastic networks - G06N 3/048: Activation functions - G06N 3/08: Learning methods - G06N 5/01: Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound - G06N 20/00: Machine learning - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract The present disclosure generally relates to handling data of non-linear, multi-variable complex systems. More particularly, the present disclosure relates to methods and systems for training machine learning-based computing devices to ensure adaptive sampling of highly complex data packets. The present invention provides a robust and effective solution to implement a complexity-based sampling methodology that trains the neural network in complex mapping regions, by iteratively sampling the DBMS function and training the neural network in complex regions. The system (110) for training the complex and non-linear neural network may be equipped with a Machine Learning (ML) Engine (214) to solve the problem efficiently. ## Claims ### Claim 1 A system (110) for training a complex and non-linear neural network, the system comprising: a processor (202); a memory (204) coupled to the processor (202), wherein the memory (204) comprises processor-executable instructions, which on execution, causes the processor (202) to: receive a set of data packets from the neural network; execute a first set of instructions based on the received set of data packets; determine a complexity of a region based on the executed first set of instructions; determine a plurality of sample points proportional to the determined complexity of the region; determine a plurality of regions of constant complexity based on the plurality of sample points; and train the non-linear neural network based on the determined plurality of regions of constant complexity. ### Claim 2 (depends on 1) The system as claimed in claim 1, wherein the neural network comprises non-linear, multi-variable complex computing devices. ### Claim 3 (depends on 1) The system as claimed in claim 1, wherein the sample points are uniform and random. ### Claim 4 (depends on 1) The system as claimed in claim 1, wherein the complexity of the region in the set of data packets is determined by curriculum sampling. ### Claim 5 (depends on 1) The system as claimed in claim 1, wherein the plurality of the regions of constant complexity is determined by a regression tree approach. ### Claim 6 (depends on 5) The system as claimed in claim 5, wherein the regression tree approach comprises K-dimensional (KD) trees. ### Claim 7 (depends on 5) The system as claimed in claim 5, wherein the regression tree approach comprises feeding a regression tree with errors (508) in data sampling after a Z score normalization (510) to identify one or more n-dimensional hypercubes. ### Claim 8 (depends on 7) The system as claimed in claim 7, wherein the one or more n-dimensional hypercubes comprise a volume of data points, a number of data points and an average error value of data points. ### Claim 9 (depends on 5) The system as claimed in claim 5, wherein the Z score normalization (510) is calculated for the one or more identified n-dimensional hypercubes. ### Claim 10 (depends on 1) The system as claimed in claim 1, wherein the neural network is trained by feeding a DBMS function with the plurality of sample points in a parallelizable fashion as an input. ### Claim 11 (depends on 7) The system as claimed in claim 7, wherein the DBMS function generates a training dataset (406) of sample data points and a test dataset (404) of sample points as an output. ### Claim 12 (depends on 1) The system as claimed in claim 1, wherein the neural network is trained based on a training dataset (406) of sample data points and a test dataset (404) of sample points. ### Claim 13 (depends on 1) The system as claimed in claim 1, wherein the neural network is trained by an adaptive sampling approach. ### Claim 14 A method for training a complex and non-linear neural network, the method comprising: receiving, by a processor (202), a set of data packets from the neural network; executing, by the processor (202), a first set of instructions based on the received set of data packets; determining, by the processor (202), a complexity of a region in the set of data packets received based on the executed first set of instructions; determining, by the processor (202), a plurality of sample points proportional to the determined complexity of the region; determining, by the processor (202), a plurality of regions of constant complexity based on the plurality of sample points; and training, by the processor (202), the non-linear neural network based on the determined plurality of regions of constant complexity. ### Claim 15 (depends on 14) The method as claimed in claim 14, wherein the neural network comprises non-linear, multi-variable complex computing devices. ### Claim 16 (depends on 14) The method as claimed in claim 14, wherein the sample points are uniform and random. ### Claim 17 (depends on 14) The method as claimed in claim 14, wherein the complexity of the region in the set of data packets is determined by curriculum sampling. ### Claim 18 (depends on 14) The method as claimed in claim 14, wherein the plurality of the regions of constant complexity is determined by a regression tree approach. ### Claim 19 (depends on 18) The method as claimed in claim 18, wherein the regression tree approach comprises K-dimensional (KD) trees. ### Claim 20 (depends on 18) The method as claimed in claim 18, wherein the regression tree approach comprises feeding a regression tree with errors (508) in data sampling after a Z score normalization (510) to identify one or more n-dimensional hypercubes. ### Claim 21 (depends on 20) The method as claimed in claim 20, wherein the one or more n-dimensional hypercubes comprise a volume of data points, a number of data points and an average error value of data points. ### Claim 22 (depends on 20) The method as claimed in claim 20, wherein the Z score normalization (510) is calculated for the one or more identified n-dimensional hypercubes. ### Claim 23 (depends on 14) The method as claimed in claim 14, wherein the neural network is trained by feeding a DBMS function with the plurality of sample points in a parallelizable fashion as an input. ### Claim 24 (depends on 14) The method as claimed in claim 14, wherein the DBMS function generates a training dataset (406) of sample data points and a test dataset (404) of sample points as an output. ### Claim 25 (depends on 14) The method as claimed in claim 14, wherein the neural network is trained based on a training dataset (406) of sample data points and a test dataset (404) of sample points. ### Claim 26 (depends on 14) The method as claimed in claim 14, wherein the neural network is trained by an adaptive sampling approach.