Dynamic Artificial Neural Network With STDP Synapse Adjustment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing artificial neural networks are based on antiquated models that ignore the temporal character of activation patterns and feedback/inhibition functions, leading to compromised neuron functions and lack of adaptive learning capabilities.
Innovation Solution
A hierarchical array of dynamic artificial neurons with synapse circuits and soma circuits that simulate biological neural networks, enabling autonomous learning through Synaptic Time Dependent Plasticity (STDP), where synapse strength values are adjusted based on the temporal difference between input and output pulses, allowing for complex task performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional artificial neural networks are used, then device complexity is reduced, but adaptability and autonomous learning capability are lost
Solution Approach 1:
The patent implements dynamic synapse strength adjustment through STDP mechanisms, where synaptic weights are not static but continuously adapt based on temporal relationships between pre-synaptic and post-synaptic spikes. This dynamic property enables autonomous learning while maintaining biological plausibility, resolving the contradiction between adaptability and complexity by using localized, simple update rules rather than complex centralized control
Solution Approach 2:
The neural network performs self-learning through STDP without external intervention or manual weight adjustment. The system autonomously modifies its own synaptic strengths based on temporal spike patterns, enabling adaptive behavior while avoiding the complexity of external training mechanisms. Each synapse serves itself by automatically adjusting its weight based on local temporal correlations
2Measurement precision
If temporal character of activation patterns is ignored, then processing speed is improved, but learning accuracy deteriorates
Solution Approach 1:
The patent models neural activation as periodic spike trains rather than continuous signals, capturing temporal patterns through discrete, rhythmic events. This periodic representation maintains temporal precision for accurate learning while enabling efficient processing through event-driven computation rather than continuous processing, resolving the speed-accuracy tradeoff
Solution Approach 2:
The patent replaces traditional mechanical or continuous processing systems with a discrete, event-driven spiking neural network model. By substituting continuous activation values with discrete spike events and using temporal patterns instead of continuous signals, the system achieves both high processing speed through efficient event handling and high learning accuracy through precise temporal encoding
3Adaptability or versatility
If feedback connections are removed, then device complexity is reduced, but adaptability deteriorates
Solution Approach 1:
The patent segments the neural network into distinct functional units (synapses, neurons, layers) with localized STDP mechanisms at each synapse. This segmentation allows feedback connections to be implemented in a modular, distributed manner rather than as a complex centralized system, enabling adaptability through local learning rules while keeping overall device complexity manageable through functional decomposition
Data Source
AI summary
A hierarchical information processing system is disclosed having a plurality of artificial neurons, comprised of binary logic gates, and interconnected through a second plurality of dynamic artificial synapses, intended to simulate or extend the function of a biological nervous system. The system is capable of approximation, autonomous learning and strengthening of formerly learned input patterns. The system learns by simulated Synaptic Time Dependent Plasticity, commonly abbreviated to STDP. Each artificial neuron consisting of a soma circuit and a plurality of synapse circuits, whereby the soma membrane potential, the soma threshold value, the synapse strength and the Post Synaptic Potential at each synapse are expressed as values in binary registers, which are dynamically determined from certain aspects of input pulse timing, previous strength value and output pulse feedback.