MesoNet: graph-directed plasticity

graph context for local STDP. how much extra work per spike before the “selective” update is basically visiting the whole network anyway

draft mechanism: represent connections as a graph and use significance/connectivity information to select where learning updates propagate. The intent is to give local STDP a wider context without updating every connection indiscriminately.

Local STDP and distributed updates

Triangulated attribution

problems identified: computational cost, limited hierarchical representation, long-range dependencies and retention/transfer. The notes consider adaptive rules, sparse connectivity and hybrid STDP/surrogate-gradient methods as alternatives.

review questions: what defines a “critical” synapse, how often is the graph recomputed, and does the selection cost exceed the updates saved? Compare update count and graph-maintenance cost separately.

• related entry: December 25 develops inverse-weight path distance and an incidence weighting idea. This December 23 note is the framing of that direction, not a measured optimization result.

• References saved in this journal section

• https://iopscience.iop.org/article/10.1088/2634-4386/ad3a95/pdf
• https://www.frontiersin.org/journals/neuroinformatics/articles/10.3389/fninf.2018.00002/full
• https://www.frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2018.00774/full

Plasticity notebook page

Graph-plasticity sketches

Equations and response sketches