Physics-Aware Auto Tiny Machine Learning for Edge Devices

Watch a technical talk exploring neurosymbolic auto tiny machine learning and its applications in resource-constrained edge devices. Discover how combining physics-based process models with neural operators can be automatically optimized based on platform constraints. Learn about the integration of symbolic techniques' context awareness with machine learning models' robustness through Bayesian optimization techniques. Explore real-world applications including onboard physics-aware neural-inertial navigation, on-device human activity recognition, on-chip fall detection, neural-Kalman filtering, and the co-optimization of neural and symbolic processes. Presented by Swapnil Sayan Saha, Algorithm Development Engineer at STMicroelectronics Inc., this hour-long presentation demonstrates how to achieve long-term high-level reasoning while maintaining physical constraints within limited platform resources.