Lightning Talk: Deep Learning in the Wild: Embedded PyTorch for... Taraqur Rahman & Owen O'Donnell

This talk focuses on the use of deep learning in wildlife conservation, specifically through the deployment of embedded PyTorch systems for passive acoustic monitoring. The speakers, Taraqur Rahman and Owen O'Donnell from OWL Integrations, present an end-to-end pipeline for detecting the endangered three-wattled bellbird in remote rainforest environments, addressing strict constraints on power and compute. They discuss the full lifecycle, including audio preprocessing, model training, optimization, and deployment, while covering challenges such as extreme class imbalance and the trade-offs between accuracy, latency, and power consumption. The session emphasizes practical insights gained from deploying machine learning at the edge, providing attendees with valuable patterns for creating real-world impact through embedded AI applications.