Dynamic Computation Graph Selection for Neural Network Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current computation graphs used in neural networks are not necessarily optimal for different datasets and problems, leading to suboptimal learning performance.
Innovation Solution
An information processing method that involves performing learning using a neural network represented by a computation graph, changing the data and/or the computation graph, and generating a predictive model through supervised learning to output an appropriate computation graph for given data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a fixed computation graph is used for neural networks, then the model structure is simple and easy to implement, but the learning performance is suboptimal for different datasets and problems
Solution Approach 1:
The patent applies dynamics by transitioning from a fixed computation graph to a dynamic selection mechanism. The computation graph is no longer static but is instead selected dynamically based on the characteristics of the input data and problem type. This allows the system to adapt the computation graph configuration to match the specific learning task requirements, thereby improving learning performance while maintaining implementation simplicity through automated selection.
Solution Approach 2:
The patent changes the parameter of computation graph configuration based on data characteristics and problem types. By varying parameters such as the number of layers, type of activation functions, and connection patterns according to the specific learning task, the system optimizes the computation graph for each problem domain. This parameter change approach enables the system to achieve optimal learning performance without requiring manual complex configuration.
2Reliability
If the computation graph is manually configured for different problems, then the learning performance can be optimized, but the ease of operation and implementation is reduced
Solution Approach 1:
The patent implements self-service by enabling the system to automatically select and configure the appropriate computation graph based on the input data characteristics and problem type. Instead of requiring manual configuration by the user, the system performs self-configuration through automated analysis of the learning task requirements. This maintains optimal learning performance while significantly improving ease of operation, as users simply need to provide the data and problem description without worrying about complex graph configuration.
Solution Approach 2:
The patent employs feedback mechanisms where the system analyzes the learning task characteristics and adjusts the computation graph configuration accordingly. The feedback loop involves evaluating the problem type and data features, selecting the appropriate computation graph, and validating the configuration. This automated feedback process ensures optimal learning performance is achieved while eliminating the need for manual setup, thereby improving ease of operation.
3Adaptability or versatility
If a single computation graph is used for all datasets, then the device complexity is minimized, but the adaptability to different problem types is reduced
Solution Approach 1:
The patent applies universality by creating a multi-functional system that can handle various problem types and datasets through a single unified framework. Instead of maintaining separate computation graphs for different problems, the system uses one versatile computation graph selection mechanism that adapts to different tasks. This universal approach improves adaptability to different datasets while avoiding the complexity of maintaining multiple specialized graphs, as the same selection framework handles all problem types.
Solution Approach 2:
The patent uses dynamics to enable the computation graph configuration to change adaptively based on the specific learning task. The system dynamically selects the appropriate computation graph variant according to the problem type and data characteristics, providing versatility without requiring a fixed separate graph for each problem. This dynamic selection mechanism achieves high adaptability while keeping the overall system complexity manageable through a single unified selection framework.
Data Source
AI summary
Provided is an information processing method, includes, by one or a plurality of processors included in an information processing device, executing: performing learning by inputting prescribed data into a prescribed learning model that uses a neural network represented by a prescribed computation graph; changing the prescribed data and/or the prescribed computation graph, the changing including changing, from among a function, which is associated with a prescribed node in a prescribed layer within the prescribed computation graph, and an activation function, which receives input of an output value from the function, the function; obtaining a learning result from the learning using the changed prescribed data and/or prescribed computation graph; performing supervised learning using learning data that includes any data and any computation graph with which the learning has been performed, as well as a learning result obtained when learning is performed using the any data and the any computation graph; and generating a predictive model that is generated through the supervised learning, the predictive model outputting a specific computation graph when receiving input of prescribed data.