Sequence anticipation and spike-timing-dependent plasticity emerge from a predictive learning rule

What if synaptic plasticity in single neurons could be explained by predictive processes?

paper | code

I got quite interested in predictive processing during my PhD. Working in the lab of Martin Vinck, we proposed an algorithmic perspective on synaptic plasticity in single neurons. We derived a gradient-based learning rule, where neurons amplify those synapses that maximally predict other synaptic inputs based on their temporal relations. We showed that this rule gives some interesting properties to single neurons, and is able to describe a variety of biological plasticity phenomena (STDP and the like) as neurons predicting their own future inputs. This was a fun project that has a long and quite intense story. It started right before the first covid wave and never really ended up.