State-Dependent Learning in Spiking Neuron Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing spiking neuron networks face challenges with target-oriented learning, which may result in slow convergence and lack of accuracy, limiting their effectiveness in achieving rapid and precise state transitions.
Innovation Solution
A computerized spiking neuron apparatus and method that implement state-dependent learning by updating neuron excitability and connection efficacy based on internal states, using eligibility traces and response processes to adjust synaptic weights, thereby facilitating plasticity updates that transition the network towards target states.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If target-oriented learning is used in spiking neuron networks, then the learning process can be guided towards a specific output, but the convergence speed is slow and accuracy is insufficient
Solution Approach 1:
The patent changes the learning rate parameter dynamically based on neuron state. When the neuron is close to its target membrane potential, the learning rate is reduced to prevent overshooting and improve precision. When the neuron is far from the target, a higher learning rate accelerates convergence. This state-dependent parameter adjustment resolves the contradiction between convergence speed and learning accuracy.
Solution Approach 2:
The patent introduces dynamic adjustment of synaptic weights based on the neuron's current state rather than using static or uniformly dynamic updates. The learning process adapts its behavior in real-time according to the neuron's proximity to the target state, enabling both rapid initial convergence and precise final adjustment, thus resolving the speed-accuracy tradeoff.
2Productivity
If synaptic weights are strongly potentiated to improve learning speed, then convergence may be faster, but oscillations and over-potentiation occur reducing stability
Solution Approach 1:
The patent applies preliminary anti-action by reducing the learning rate before the neuron reaches its target state. This preemptive reduction prevents overshooting and oscillations that would otherwise occur due to strong potentiation. By anticipating the potential instability and counteracting it in advance, the system maintains both learning speed and network stability.
Solution Approach 2:
The patent uses feedback from the neuron's current membrane potential to modulate the learning rate. The system continuously monitors the neuron state and adjusts synaptic potentiation strength accordingly, reducing potentiation when the neuron approaches the target to prevent oscillations. This state-dependent feedback mechanism enables fast learning while maintaining stability.
Data Source
AI summary
State-dependent supervised learning framework in artificial neuron networks may be implemented. A framework may be used to describe plasticity updates of neuron connections based on a connection state term and a neuron state term. Connection states may be updated based on inputs and outputs to and/or from neurons. The input connections of a neuron may be updated using input traces comprising a time-history of inputs provided via the connection. Weight of the connection may be updated and connection state may be time varying. The updated weights may be determined using a rate of change of the input trace and a term comprising a product of a per-neuron contribution and a per-connection contribution configured to account for the state time-dependency. Using event-dependent connection change components, connection updates may be executed on a per neuron basis, as opposed to a per-connection basis.