MesoNet: CIFAR-100 encoding and invariant features
CIFAR-100: what survives the image → spike conversion, esp color + spatial structure? save for later: shift the image one pixel, how much should the spike pattern change
motivation: a more complex image dataset tests spatial structure beyond MNIST. The notes consider time-to-first-spike encoding and invariant information clustering as possible directions.
unresolved translation: an ANN clustering objective does not directly define a local spike-timing update. Specify what makes two encoded inputs equivalent and which activity statistic represents that equivalence.
review questions: does encoding preserve relative spatial information and color information? How does the time horizon affect spike count and representation? Compare encoders with the same train/test partition and readout rather than changing all components together.
related constraint: homeostatic plasticity is considered as a way to maintain usable activity, but the entry does not establish an improved CIFAR-100 score.
References saved in this journal section