Christoph Stöckl

The paper links cognitive maps, successor-relation based reinforcement learning, and BTSP by showing how neural codes encode sequential structure and enable plan generation.

The paper shows that local synaptic plasticity enables neural networks to learn cognitive maps for planning, achieving performance comparable to AI offline planning algorithms without requiring backpropagation or large datasets.

The paper shows that networks of spiking neurons can have innate computational capabilities through their connectivity structure, not synaptic plasticity, when they include diverse neuron types.

This paper introduces a new learning approach for spiking neural networks that enables one-shot learning through local synaptic plasticity, inspired by dopamine-based learning signals in the brain.