MesoNet: benchmark axes
accuracy, spikes/inference, convergence, wall time. keep separate columns—moving to CIFAR also changes the encoding problem
Target measures: classification accuracy, computational efficiency and convergence speed. The notes propose spikes per inference and epochs to convergence alongside accuracy; retain wall-clock time as a separate execution measurement.
Datasets: MNIST as the baseline, then CIFAR-100 or an ImageNet subset for more complex structure. Encoding remains a major confound when moving from one dataset to another.
Learning alternatives surveyed: contrastive objectives, clustering, and energy-based approaches. These are candidate connections to STDP rather than completed implementations.
Energy comparison
Model comparison plots
Review questions: compare against other SNNs, conventional ANNs, or both? What compute budget, input duration, label access and stopping rule are shared? Define the target before tuning so a faster but less accurate model is not silently treated as the same outcome.
References saved in this journal section