Enhanced accuracy in first-spike coding using current-based adaptive LIF neuron.

Area of Science:

• Computational Neuroscience
• Artificial Intelligence
• Machine Learning

Background:

• Spiking neural networks (SNNs) utilize first spike (FS) timing for efficient information processing.
• Current FS coding methods show promise but lag behind advanced techniques for complex temporal data.
• Improving neural dynamics is key to unlocking the full potential of FS coding.

Purpose of the Study:

• To enhance the performance of FS coding in SNNs for auditory data classification.
• To improve neural dynamics for better temporal correlation and memory preservation.
• To reduce decision-making delays in SNNs utilizing FS coding.

Main Methods:

• Introduction of a current-based adaptive LIF neuron (CuAdLIF) with delayed responses and membrane potential adaptation.
• Development of strategies to minimize decision-making delays.
• Implementation of adaptive training for FS coding.
• Evaluation on auditory datasets.

Main Results:

• The CuAdLIF neuron significantly improved the extraction of temporal features.
• FS coding accuracy was substantially enhanced compared to previous methods.
• Proposed strategies effectively reduced output time delays in the SNNs.
• The enhanced SNN demonstrated superior performance in auditory classification tasks.

Conclusions:

• The CuAdLIF neuron and adaptive strategies represent a significant advancement for FS coding in SNNs.
• This approach offers a more efficient and accurate method for processing temporal information, particularly in auditory domains.
• The findings pave the way for more sophisticated and faster SNNs in real-world applications.