Reproduce “Meta-learning three-factor plasticity rules for structured credit assignment with sparse feedback” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Dimitra Maoutsa
A meta-learning framework is introduced that discovers biologically plausible local, three-factor synaptic plasticity rules for recurrent neural networks. These rules enable structured credit assignment using sparse and delayed reinforcement feedback, and the framework's gradient computation method was rigorously validated.
Key insights
- It is possible to discover effective local plasticity rules through meta-optimization, rather than relying on hand-designed or biologically implausible global rules.
- The framework provides a computational means to bridge the gap between biological learning constraints (locality, sparse feedback) and the computational power of artificial neural networks.
- Tangent-propagation through learning significantly advances the feasibility of meta-learning over complex, temporally extended inner-loop learning processes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.