Efficient multi-scale representation of visual objects using a biologically plausible spike-latency code and...

Area of Science:

• Computational Neuroscience
• Artificial Intelligence
• Computer Vision

Background:

• Deep neural networks excel at object recognition but are energy-intensive.
• Spiking neural networks (SNNs) offer a more biologically plausible and efficient alternative.

Purpose of the Study:

• To develop an efficient SNN model for object recognition using spike-latency coding and winner-take-all inhibition (WTA-I).
• To investigate the impact of spatial frequency (SF) processing and WTA-I schemes on visual stimulus representation.

Main Methods:

• A multi-scale parallel processing approach using three spatial frequency (SF) channels.
• Spiking neurons with synaptic weights updated via spike-timing-dependent plasticity (STDP).
• Implementation of winner-take-all inhibition (WTA-I) for efficient coding.

Main Results:

• The SNN model efficiently represents visual objects using spike-latency coding.
• A network of 200 spiking neurons achieved object representation with as few as 15 spikes per neuron.
• Performance was analyzed across different SF bands and WTA-I configurations.

Conclusions:

• Biologically plausible learning rules in SNNs can enable efficient object recognition.
• This approach advances understanding of brain function and facilitates novel artificial vision systems.