MesoNet: history-dependent synaptic updates

history beyond the last pre/post pair. extra state + its timescale need names, “second order” by itself is not enough to implement anything

experiment direction: add dependence on past activity rather than using only the immediate pre/post timing pair. The notes call this a second-order adaptation term and compare it with the current STDP and Hebbian alternatives.

qualitative observations: more diverse neuron activations under the nonlinear/history-dependent variant; stronger global patterns under Hebbian adaptation but a need to control runaway plasticity. No coefficient values or benchmark results are recorded in this dated section.

Local STDP and distributed updates

Variable plasticity

review questions: how long is the retained history, how is it decayed, and which state variable stores it? Compare memory cost and update cost as well as firing diversity. A useful experiment should determine whether the added history helps retention or merely changes firing rates.

References saved in this journal section

• https://www.jmlr.org/papers/volume22/20-1031/20-1031.pdf
• https://web.stanford.edu/group/brainsinsilicon/research/
• https://arxiv.org/pdf/2405.20694
• https://math.stackexchange.com/questions/214212/intuition-behind-jacobian-of-the-svd
• https://arxiv.org/pdf/1711.04735