Entity Identification via Thresholded Activity Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neural network solutions for entity identification suffer from poor performance, low reliability, high computational requirements, and inefficient use of resources such as storage space and energy.
Innovation Solution
A method involving a network of nodes that generates activity levels based on sensor input, compares these levels to a threshold, iterates until a local minimum is reached, and uses the distribution of activity levels to identify measurable characteristics, allowing for dynamic network structure and improved identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If neural networks are used for entity identification, then identification capability is achieved, but performance and reliability are poor
Solution Approach 1:
The patent changes the fundamental parameters of the neural network by using only positive scalar weights instead of traditional negative weights, and by introducing an activity threshold mechanism. This transforms the network into a system that naturally converges to stable states, improving both reliability and measurement precision of entity identification.
Solution Approach 2:
The network performs self-organization through iterative activity updates where nodes automatically adjust their activity levels based on threshold comparisons. This self-service mechanism eliminates the need for external training data or complex weight adjustments, resulting in more reliable and precise identification without requiring manual intervention.
2Productivity
If traditional neural networks are trained, then entity identification is achieved, but training time is significant and computational power required is high
Solution Approach 1:
The patent applies partial action by using only positive scalar weights and threshold-based updates instead of full neural network training. This reduced-action approach achieves entity identification without requiring extensive training iterations or computational resources, dramatically reducing both training time and energy consumption.
Solution Approach 2:
The patent replaces the mechanical training process of traditional neural networks with a simpler iterative activity update mechanism. Instead of adjusting weights through backpropagation and gradient descent, the system uses threshold-based activity comparisons that naturally converge to stable states, eliminating the need for computationally intensive training procedures.
3Productivity
If existing neural network solutions are used, then entity identification is performed, but storage space requirements are high
Solution Approach 1:
The patent extracts the essential function of entity identification from the complex neural network structure, retaining only the core activity update and threshold comparison mechanisms. This extraction eliminates the need for large storage spaces required by traditional neural network weight matrices while maintaining identification efficiency.
Solution Approach 2:
By changing the weight parameter space from real numbers (allowing negative values) to positive scalars only, and introducing threshold-based updates, the patent dramatically reduces the storage requirements. The simplified parameter representation enables efficient entity identification with minimal storage space compared to traditional neural networks.
Data Source
AI summary
Methods and apparatus, including computer program products, for identifying characteristics in time-varying information data. A network of nodes receives input comprising time-varying information data. An activity level is generated by each node based on the input and/or an input combination of time-varying information data. The activity level of each node is compared to a threshold level. Based on the comparing, the activity level is set for each node to a preset value, or the generated activity level is kept. A total activity level is calculated as the sum of all activity levels of the nodes of the network. If a condition for the total activity level being a local minimum of the total activity level has not been met, the providing, generating, comparing, setting, calculating and determining are repeated. Otherwise, a distribution of activity levels is utilized at the local minimum to identify a measurable characteristic of the data.