Stability versus neuronal specialization for STDP: long-tail weight distributions solve the dilemma

Area of Science:

• Computational Neuroscience
• Neuroscience
• Artificial Intelligence

Background:

• Spike-timing-dependent plasticity (STDP) is critical for synaptic weight modification and neural network structure.
• Existing STDP models face a dilemma between stability and functional synaptic specialization.
• Weight dependence in STDP is physiologically observed but functionally unclear.

Purpose of the Study:

• To address the stability-versus-function dilemma in STDP.
• To introduce a novel STDP model, log-STDP, with sublinear weight dependence for depression.
• To investigate the functional implications of log-STDP on synaptic weight distributions and network dynamics.

Main Methods:

• Development of the log-STDP model featuring sublinear weight dependence.
• Analysis of weight distributions and synaptic competition induced by log-STDP.
• Simulation of log-STDP in recurrently connected networks.

Main Results:

• Log-STDP generates broad weight distributions with no hard upper bound, mimicking experimental observations.
• The model induces graded synaptic competition, pushing correlated inputs towards larger weights.
• Log-STDP demonstrates stable dynamics and robust competition in recurrent networks, enhancing information processing.

Conclusions:

• Log-STDP offers a solution to the STDP stability-versus-function dilemma.
• The model's unique weight dependence promotes effective synaptic specialization and stable network structures.
• Log-STDP shows promise for advanced spike-based information processing.