src.canns.trainer.stdp

STDP (Spike-Timing-Dependent Plasticity) trainer.

Classes

STDPTrainer

STDP (Spike-Timing-Dependent Plasticity) trainer.

Module Contents

class src.canns.trainer.stdp.STDPTrainer(model, learning_rate=0.01, A_plus=0.005, A_minus=0.00525, weight_attr='W', w_min=0.0, w_max=1.0, compiled=True, **kwargs)

Bases: src.canns.trainer._base.Trainer

STDP (Spike-Timing-Dependent Plasticity) trainer.

STDP is a biologically-inspired learning rule that adjusts synaptic weights based on the precise timing of pre- and post-synaptic spikes. Synapses are strengthened when pre-synaptic spikes precede post-synaptic spikes (LTP), and weakened when the order is reversed (LTD).

Trace-based Learning Rule:

ΔW_ij = A_plus * trace_pre[j] * spike_post[i] - A_minus * trace_post[i] * spike_pre[j]

where:

• W_ij is the weight from input j to neuron i

• spike_pre[j] is the presynaptic spike (0 or 1)

• spike_post[i] is the postsynaptic spike (0 or 1)

• trace_pre[j] is the exponential trace of presynaptic spikes

• trace_post[i] is the exponential trace of postsynaptic spikes

• A_plus controls LTP (long-term potentiation) magnitude

• A_minus controls LTD (long-term depression) magnitude

The spike traces evolve as:

trace = decay * trace + spike

This provides a temporal window for spike-timing correlations.

References

• Gerstner & Kistler (2002): Spiking Neuron Models

• Morrison et al. (2008): Phenomenological models of synaptic plasticity

• Bi & Poo (1998): Synaptic modifications in cultured hippocampal neurons

Initialize STDP trainer.

Parameters:

• model (src.canns.models.brain_inspired.BrainInspiredModel) – The spiking model to train (typically SpikingLayer)

• learning_rate (float) – Global learning rate multiplier (default: 0.01)

• A_plus (float) – LTP magnitude (default: 0.005)

• A_minus (float) – LTD magnitude (default: 0.00525, slightly > A_plus for stability)

• weight_attr (str) – Name of model attribute holding the connection weights

• w_min (float) – Minimum weight value (default: 0.0 for excitatory synapses)

• w_max (float) – Maximum weight value (default: 1.0)

• compiled (bool) – Whether to use JIT-compiled training loop (default: True)

• **kwargs – Additional arguments passed to parent Trainer

predict(pattern, *args, **kwargs)

Predict output spikes for a single input spike pattern.

Parameters:

pattern – Input spike pattern of shape (input_size,)

Returns:

Output spike pattern of shape (output_size,) with binary values (0 or 1)

train(train_data)

Train the model using STDP rule.

Parameters:

train_data (collections.abc.Iterable) – Iterable of input spike patterns (each of shape (input_size,)) Each pattern should contain binary values (0 or 1)

A_minus = 0.00525

A_plus = 0.005

compiled = True

learning_rate = 0.01

w_max = 1.0

w_min = 0.0

weight_attr = 'W'