MesoNet: unsupervised-learning scope
unsupervised STDP, see how far this can go w/ depth. also catastrophic forgetting—more layers running ≠ more useful features learned
starting direction: unsupervised STDP, with the project journal and deadlines established on September 15. The immediate research question is whether a local spike-timing rule can support useful hierarchical representations as convolutional SNN depth increases.
separate questions: computational scaling with synapse count; representation quality in deeper layers; catastrophic forgetting under ongoing plasticity. A network that runs at greater depth has not necessarily learned higher-level features.
planned baseline: establish an existing STDP image-classification model before introducing a new rule. Keep input encoding, neuron dynamics, inhibition and learning rule separate so an experiment can identify which change caused an effect.
ANN and SNN neuron schematic
ANN and SNN model comparison
review question: what is the layer-wise evidence that the next layer adds information rather than merely reproducing low-level clusters? The starting notes identify this limitation; they do not provide a depth-scaling result.
References saved in this journal section