Theory on reward-modulated learning in recurrent neural networks

We study learning dynamics with local reward-modulated plasticity rules both in linear and non-linear recurrent neural networks (RNNs).

The first part of this work on linear RNNs has been accepted in NeurIPS 2026 main conference and will appear soon online see here.

Extensions on this work and theory on reward-modulated learning in nonlinear RNNs will appear here and online as they become ready. Stay tuned!