MesoNet: vanishing spikes in Spiking-Inception

vanishing spikes, ugh. check activity layer by layer, then encoding magnitude / inhibition / thresholds instead of changing everything together

code reference: Spiking-Inception src/Train.py. Initial training encountered a vanishing-spike problem and sensitivity to learning-rate settings.

Convolutional spiking architecture

research directions: compare update strategies, STDP variants and homeostatic mechanisms. The notebook records reading and hyperparameter adjustments, not a definitive fix.

proposed diagnostic sequence: examine activity after each layer, then input encoding magnitude, thresholds/reset behavior, inhibition and weight distributions. A final-layer accuracy alone cannot reveal where activity disappears.

• review questions: is the loss of spikes present before training or introduced by updates? Are silent units recovering under homeostasis, or being kept silent by initialization/normalization? Preserve the failed configuration so a claimed fix can be tested against it.

• References saved in this journal section

• https://arxiv.org/pdf/2001.01680
• https://pmc.ncbi.nlm.nih.gov/articles/PMC6639412/
• https://doi.org/10.1007/978-3-030-01264-9_9
• https://www.sciencedirect.com/science/article/pii/S2542435123003653
• https://www.frontiersin.org/journals/computational-neuroscience/articles/10.3389/fncom.2010.00019/full#B9