Optical spike-timing-dependent plasticity with weight-dependent learning window and reward modulation.

Area of Science:

• Neuromorphic Engineering
• Computational Neuroscience
• Photonics

Background:

• Spike-timing-dependent plasticity (STDP) is crucial for learning in biological and artificial neural systems.
• Biological STDP synapses often exhibit multiplicative boundaries and modulation by third factors like dopamine.
• Existing optical STDP synapses lack these biological complexities.

Purpose of the Study:

• To introduce a third factor into optical STDP synapses for enhanced learning capabilities.
• To emulate biological STDP synapse behavior more closely in photonic neuromorphic systems.
• To enable reward-based reinforcement learning in optical neuromorphic systems.

Main Methods:

• Implemented a third factor, current-injection modulation of semiconductor optical amplifiers, into optical STDP.
• Introduced optical weight-dependent STDP using local feedback of synaptic weight.
• Incorporated global reward signals for adaptive modulation.
• Conducted simulation studies with scalable photonic neurons.

Main Results:

• Optical weight-dependent STDP more closely emulates biological STDP, acting as an intermediate between additive and multiplicative plasticity.
• This configuration balances synaptic stability and competition.
• Optical STDP with reward modulation successfully enables reward-based reinforcement learning.

Conclusions:

• The proposed third factor significantly enhances the biological plausibility and functionality of optical STDP synapses.
• This approach offers a novel pathway for developing advanced photonic neuromorphic systems capable of complex learning.
• The findings pave the way for more sophisticated, adaptive, and biologically inspired neuromorphic computing.