Spiking Neural Network Learning via Selective Neuron Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Spiking neural networks face challenges in achieving high accuracy and efficient learning due to issues like crosstalk between input subjects and inadequate determination of synaptic weights, leading to incomplete learning and decreased performance.
Innovation Solution
A learning method for spiking neural networks that involves first and second learning phases, where intermediate neurons are determined based on spike output signals, and synaptic weights are adjusted using the spike-timing-dependent plasticity (STDP) algorithm, with additional learning focused on neurons that failed in the initial phase to improve accuracy and speed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If spiking neural networks use pulse type spike signals for information transfer, then power consumption is reduced, but learning accuracy and efficiency deteriorate
Solution Approach 1:
The learning process is segmented into multiple phases (first learning phase and second learning phase). In the first phase, all intermediate neurons undergo initial learning. In the second phase, only specific intermediate neurons that require additional training are selected and subjected to further learning. This segmentation allows the network to achieve high learning accuracy while maintaining energy efficiency by avoiding unnecessary learning operations on already-sufficient neurons.
Solution Approach 2:
Instead of applying learning operations to all intermediate neurons uniformly, the patent applies partial action by selectively targeting only those neurons that need additional learning based on their spike output characteristics. This prevents excessive learning operations on neurons that have already achieved sufficient learning, thereby reducing overall power consumption while maintaining learning accuracy.
2Reliability
If conventional learning methods are applied to all intermediate neurons, then learning completeness is improved, but learning efficiency and power consumption deteriorate
Solution Approach 1:
The patent implements a feedback mechanism where the spike output signals of intermediate neurons are monitored and evaluated. Based on the number of spikes in output signals, the system determines which neurons require additional learning. This feedback-driven approach ensures that learning is applied only where needed, maintaining learning completeness while significantly improving learning efficiency and reducing power consumption.
Solution Approach 2:
The patent changes the parameter of learning application from a static uniform approach to a dynamic selective approach. By using the number of spikes as a criterion parameter, the system adaptively determines which neurons should undergo learning in the second phase. This parameter-based selection ensures complete learning of necessary neurons while avoiding redundant operations, thus improving efficiency without sacrificing completeness.
3Stability of the object's composition
If uniform learning is applied to all neurons, then learning consistency is improved, but learning speed and adaptability deteriorate
Solution Approach 1:
The patent introduces dynamics into the learning process by making the learning application adaptive rather than static. The second learning phase dynamically selects which intermediate neurons to train based on their individual spike output characteristics. This dynamic approach allows the network to maintain consistency in learning methodology while adapting to the specific needs of different neurons, thereby improving overall learning speed without sacrificing consistency.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach enhances the learning efficiency and accuracy of spiking neural networks by selectively targeting neurons for additional training, reducing crosstalk and ensuring complete learning, thereby improving overall network performance.
Implementation Method 1
synaptic weights are adjusted using the spike-timing-dependent plasticity (STDP) algorithm
Data Source
AI summary
Disclosed is a learning method of a neural network which includes a first intermediate neuron layer and a second intermediate neuron layer. The method includes performing first learning, which is based on a first synaptic weight layer, with respect to input subjects and the first intermediate neuron layer, determining intermediate neurons, which will perform second learning, from among intermediate neurons of the first intermediate neuron layer, based on the number of spikes of each of spike output signals of the intermediate neurons of the first intermediate neuron layer, and performing the second learning, which is based on a second synaptic weight layer, with respect to the intermediate neurons determined to perform the second learning.