MesoNet: experiment-wrapper contract

wrapper: load → encode → run → outputs. keep the model bits swappable, this does not need to become one gigantic script

architecture choice: begin with convolutional SNNs for visual datasets while keeping feedforward and recurrent alternatives in the comparison notes.

wrapper responsibilities: dataset loading/preprocessing, encoder selection, simulation execution and comparable outputs. The initial wrapper is intended to cover MNIST, SHD and DVS Gesture with Poisson, TSC or direct-spike input as appropriate.

learning ideas: activity/state-dependent learning rates and reward-like signals for unsupervised STDP. These are design directions; the entry does not prove they solve deep-network scaling.

• review questions: define the interface between encoder and simulator—neuron indices, timestamps, sample boundaries and labels. Store model configuration and evaluation configuration separately. Which transformations are fitted on training data, and which parameters are held fixed across datasets?

• References saved in this journal section

• https://brian2.readthedocs.io/en/stable/advanced/functions.html#default-functions
• https://www.sciencedirect.com/science/article/abs/pii/S0925231299000958
• https://medium.com/@theagipodcast/spiking-neural-network-architectures-e6983ff481c2

Iterative development