Neural Network Apparatus Learning Threshold Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Neuromorphic processors face challenges in maintaining normal learning processes due to synaptic weights constantly being reduced when the firing threshold of neuron circuits is lowered, leading to ineffective learning.
Innovation Solution
A neural network apparatus with a control circuit that adjusts the learning threshold in synaptic circuits based on the frequency of firing signals from neuron circuits, ensuring the synaptic weights are updated appropriately and maintaining a balanced firing frequency range.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the firing threshold of neuron circuits is lowered to adjust firing frequency to a narrow range, then firing frequency control is improved, but synaptic weights are constantly reduced leading to ineffective learning
Solution Approach 1:
The control circuit monitors the firing frequency of post-synaptic neuron circuits and dynamically adjusts the learning threshold based on this feedback. When firing frequency exceeds the target range, the learning threshold is increased to prevent excessive weight reduction; when firing frequency is too low, the learning threshold is decreased to allow normal learning. This closed-loop feedback mechanism resolves the contradiction by adapting the learning threshold to maintain both firing frequency control and learning effectiveness.
Solution Approach 2:
The invention dynamically changes the learning threshold parameter based on the firing frequency of post-synaptic neurons. By adjusting this parameter in response to observed firing patterns, the system maintains synaptic weights within an effective range while keeping firing frequencies controlled, thus resolving the contradiction between frequency control and learning effectiveness.
2Adaptability or versatility
If the learning threshold is kept low to enable continuous learning, then learning capability is improved, but firing frequency becomes uncontrolled and spreads over a wide range
Solution Approach 1:
The learning threshold is transformed from a static parameter to a dynamic one that automatically adjusts based on firing frequency measurements. This dynamic adaptation allows the system to maintain learning capability when needed while controlling firing frequency within the target range, resolving the contradiction between adaptability and precision.
Solution Approach 2:
The control circuit uses feedback from firing frequency measurements to regulate the learning threshold. This feedback mechanism ensures that learning capability is preserved when firing frequencies are appropriate while preventing frequency spread when thresholds are too low, thus balancing adaptability and precision.
3Productivity
If synaptic weight updates are performed based on low internal potential, then learning speed is improved, but weights are constantly reduced leading to learning degradation
Solution Approach 1:
The learning threshold parameter is dynamically changed based on firing frequency to control synaptic weight updates. By adjusting this parameter, the system maintains learning speed while preventing the pathological constant reduction of weights, thus resolving the contradiction between productivity and reliability.
Solution Approach 2:
The control circuit performs preliminary anti-action by increasing the learning threshold before synaptic weights can be excessively reduced. This preventive measure counteracts the tendency toward weight degradation while preserving learning speed, resolving the contradiction between productivity and reliability.
Data Source
AI summary
A neural network apparatus according to an embodiment includes neuron circuits, synaptic circuits, and a control circuit. A firing circuit of each neuron circuit outputs a firing signal when absolute value of the internal potential is larger than a firing threshold. A firing threshold adjustment circuit of each neuron circuit changes the firing threshold in accordance with frequency of the firing signal. When the firing signal is output from a pre-synaptic neuron circuit, the synaptic circuit changes the synaptic weight in accordance with a contrast between a learning threshold and the absolute value of the internal potential held in a post-synaptic neuron circuit. The control circuit changes the learning threshold in accordance with frequency of the firing signal from a target neuron circuit. The learning threshold is used for changing the synaptic weight stored in one or more synaptic circuits each outputting the output signal to the target neuron.