Voltage slope guided learning in spiking neural networks

Area of Science:

• Machine Learning
• Computational Neuroscience
• Artificial Intelligence

Background:

• The temporal credit-assignment problem is a key challenge in machine learning, especially for systems with delayed feedback.
• Aggregate-label learning algorithms in spiking neural networks aim to address this by assigning feedback signals to relevant inputs.
• Existing methods face computational inefficiencies or difficulties in identifying learning adjustment points.

Purpose of the Study:

• To propose a novel algorithm, membrane voltage slope guided (VSG), to overcome limitations in aggregate-label learning for spiking neural networks.
• To enhance the efficiency and reliability of learning from delayed feedback signals in complex input activity.

Main Methods:

• Developed the membrane voltage slope guided (VSG) algorithm for spiking neural networks.
• Utilized membrane voltage dynamics to identify precise weight adjustment points, avoiding intensive computations.
• Evaluated VSG's performance on simulated data and real-world medical and speech classification tasks.

Main Results:

• The VSG algorithm effectively correlates delayed feedback signals with relevant clues in background spiking activity.
• VSG demonstrates superior performance in classification tasks compared to previous aggregate-label learning methods.
• The algorithm successfully avoids intensive calculations and ensures consistent identification of adjustment points.

Conclusions:

• The proposed VSG algorithm offers a computationally efficient and reliable solution for the temporal credit-assignment problem in spiking neural networks.
• VSG provides a meaningful advancement for aggregate-label learning, with demonstrated effectiveness on diverse datasets.
• This method represents a significant reference for future research in biologically inspired machine learning.