Research notes, methods, and practical write-ups.
Talk slides on biologically- and hardware-plausible alternatives to backpropagation through time — RTRL, feedback alignment, e-prop, OSTL, OTTT, S-TLLR, and TESS.
An exploration of spiking neural networks and their application to time series forecasting — from neuron dynamics and spike encoding to energy-efficient temporal modeling.
A concise overview of forecasting augmentation methods, from frequency-domain perturbations and wavelets to patch-based Temporal Patch Shuffle.