Classification Network Pre-Synaptic Inhibition Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing pattern recognition systems face inefficiencies in real-time analysis of images due to cycling strategies that can lead to mixed signals and require waiting for image changes to be recognized, especially in multiple layers of hierarchy, where de-inhibition is not well defined and recognition of new images is problematic.
Innovation Solution
A classification network that uses pre-synaptic inhibition for feedback connections, allowing for simultaneous pattern classification and continuous feedback, enabling dynamic adjustment of co-occurring features without extensive training, thereby avoiding combinatorial explosions and improving recognition of novel patterns.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If cycling strategy is used in ART for pattern recognition, then pattern classification can be performed, but real-time analysis becomes inefficient due to mixed signals and requirement to wait for image changes
Solution Approach 1:
The patent implements continuous feedback connections that operate asynchronously with image changes, allowing the network to continuously process patterns without waiting for discrete image updates. This continuous action enables real-time analysis while maintaining recognition accuracy through ongoing feedback loops between output and input layers.
Solution Approach 2:
The patent introduces dynamic feedback mechanisms that adjust network states in response to changing image data. The feedback connections create a dynamic system that can adapt its processing based on real-time image characteristics, enabling efficient real-time analysis without the rigid cycling strategy of traditional ART.
2Measurement precision
If lateral inhibition and winner-take-all paradigm are used, then singular response is generated, but simultaneous representations of multiple patterns are not allowed
Solution Approach 1:
The patent uses feedback connections to enable multiple output units to simultaneously represent different patterns. Instead of traditional lateral inhibition that forces one winner, feedback allows continuous adjustment where multiple patterns can coexist and be represented concurrently, maintaining specificity while enabling versatility.
Solution Approach 2:
The patent changes the connection parameters from fixed lateral inhibition to dynamic feedback connections. This parameter change allows the network to adjust activation levels continuously, enabling multiple patterns to be represented simultaneously with appropriate weightings rather than forcing a single winner-take-all outcome.
3Ease of operation
If de-inhibition is implemented in ART hierarchy, then cells can be released for new images, but the mechanism is not well defined and recognition of new images becomes problematic
Solution Approach 1:
The patent defines de-inhibition through feedback connections from higher hierarchy levels to lower levels. When a cell in a lower level is successfully inhibited, feedback signals its status, allowing controlled de-inhibition when needed. This feedback-based mechanism provides clear operational rules while maintaining reliable new image recognition through hierarchical coordination.
Solution Approach 2:
The patent implements preliminary de-inhibition preparation where cells are held in an inhibited state until a new image is detected. The feedback mechanism prepares the network by signaling when de-inhibition is appropriate, ensuring cells are released at the correct moment for new image recognition rather than randomly, improving both operational clarity and recognition reliability.
4Reliability
If extensive training is performed to optimize connection strengths, then pattern recognition accuracy improves, but system complexity and training time increase significantly
Solution Approach 1:
The patent implements self-organizing feedback mechanisms where the network automatically adjusts its own connection strengths through feedback from output to input layers. Instead of requiring external extensive training, the system performs self-training by continuously refining its weights based on recognition performance, reducing both training complexity and time while maintaining accuracy.
Solution Approach 2:
The feedback connections enable the network to learn from its own performance automatically. Output errors feed back to adjust input connections, creating a self-correcting system that reduces the need for extensive manual training. This feedback-driven learning simplifies the training process while improving recognition accuracy through continuous self-optimization.
Data Source
AI summary
Pattern classification system and methods are disclosed. The pattern classification systems and methods employ one or more classification networks that can parse multiple patterns simultaneously while providing a continuous feedback about its progress. Per-synaptic inhibition is employed to inhibit feedback connections to permit more flexible processing. Various additional improvements result in highly robust pattern recognition systems and methods that are suitable for use in research, development, and production.