Adaptive Neural Network Group Control Across Varying Machine Settings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional neural networks face accuracy loss and catastrophic memory issues when operating in varying environments, leading to suboptimal performance as they are trained with increasingly large datasets to cover multiple settings.
Innovation Solution
The system collects training data for specific settings, associates characteristics with neural network coefficients and structures, and stores them in a database management system, allowing for adaptive switching between different neural network configurations based on changing settings to maintain optimal performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If training data set size is increased to cover multiple different settings, then the neural network can operate in more settings, but accuracy is lost and catastrophic memory loss occurs
Solution Approach 1:
The patent segments the training data by creating separate training data sets for different settings or environments. Instead of using one large diverse training set, the system divides data into multiple specialized sets, each tailored to a specific setting, thereby maintaining high accuracy within each setting while achieving broad adaptability through selective deployment.
Solution Approach 2:
The system dynamically selects and switches between different neural network configurations based on the current setting or environment. This dynamic adaptation allows the system to maintain optimal accuracy for the current context while being versatile across multiple settings, resolving the contradiction between adaptability and reliability.
2Adaptability or versatility
If training data set size is increased to cover multiple different settings, then the neural network can operate in more settings, but catastrophic memory loss occurs
Solution Approach 1:
By segmenting training data into setting-specific subsets, the system avoids the memory loss associated with training on large diverse datasets. Each segmented training set is manageable in size and focused on a specific setting, preventing catastrophic forgetting while maintaining the ability to operate across multiple settings through configuration switching.
Solution Approach 2:
The system performs preliminary actions by pre-training separate neural network configurations on different setting-specific training data sets before deployment. This allows the system to have multiple pre-adapted configurations ready for selective use, avoiding the need to maintain large amounts of diverse training data in memory during operation.
3Adaptability or versatility
If a single neural network is trained with large datasets to cover multiple settings, then versatility is improved, but device complexity increases
Solution Approach 1:
The patent segments the neural network system into multiple simpler configurations, each trained on specific setting data. This segmentation replaces the need for one complex universal network with multiple simpler specialized networks, reducing individual model complexity while maintaining overall system versatility through selective configuration deployment.
Data Source
AI summary
Methods and systems that allow neural network systems to control adaptively groups of automated machines to achieve a common goal. In particular, a number of automated machines can be divided into multiple groups. Each group having one or more automated machines, and the automated machines belong to each group is to use an identical neural network (that is, having the same neural network structure and the same set of coefficients). As the groups of automated machines operate to achieve the common goal, some groups may perform better compared with other groups. If so, the neural network of the better performing group is copied to automated machines in other groups to ensure more automated machines also improve the results in achieving the common goal.