Stochastic Sparse Backpropagation

Investigating whether training can preserve model quality while reducing backward-pass computation. The project compares dense training with stochastic gradient selection, block sparsity, pruning-based child networks, and dynamic subnetwork refresh strategies.

• Dense forward passes with selective backward updates
• MLP and CNN benchmarks
• Accuracy, time, memory, and compute analysis

Wild Basin Trail Camera Platform

A research platform for managing and exploring large trail-camera collections through scientific review workflows, SpeciesNet predictions, OCR metadata extraction, filtering, and privacy-aware publication tools.

Trail Camera Image Analysis Tool

A desktop workflow for conservation teams that coordinates SpeciesNet inference, OCR extraction, resumable jobs, parallel downloads, and structured result exports.