Neuromodulation + Eligibility Traces (Three-Factor Learning)

Three-factor learning gates eligibility traces with delayed neuromodulatory signals to drive online plasticity. This Skill orchestrates reward-based updates by bridging local synaptic activity with modulatory signals to solve delayed credit assignment in PyTorch.

Accumulating, replacing, and Dutch trace variants are methods for managing eligibility traces during online learning. This Skill supports all three variants with rate-based and STDP kernels, enabling flexible biologically inspired plasticity gated by neuromodulation.

Yes, you can gate weight updates using DA, ACh, NE, and 5-HT neuromodulatory signals. This Skill applies these modulatory signals to eligibility traces, accelerating reward-guided adaptation in cognitive agents during online learning.

You configure neuromodulated STDP modes using EligibilityConfig, NeuromodConfig, ThreeFactorConfig, and PlasticityFullConfig. These end-to-end configurations support online, hybrid, and auxiliary_loss modes for flexible three-factor learning.

Yes, PyTorch is required as the core dependency. The Skill uses PyTorch to execute minimal online three-factor updates by gating eligibility traces with neuromodulatory signals for reward-based plasticity.

Use three-factor learning when tasks require reward-based plasticity, attention gating, or delayed credit assignment. Standard STDP lacks neuromodulatory gating, whereas this approach bridges local synaptic activity with delayed modulatory signals to drive adaptive online learning.