Explore cutting-edge neuromorphic engineering concepts in this tinyML forum session focused on algorithms. Delve into hardware-friendly learning for edge ML with Ralph Etienne Cummings, investigate robustness and efficiency in neural systems using spike-based machine intelligence with Priya Panda, and learn about training spiking neural networks end-to-end using surrogate gradients from Friedemann Zenke. Discover how to enable neuromorphic learning machines through meta-learning with Emre Neftci. Gain insights into topics such as Batch Normalization Through Time (BNTT) for temporal learning, energy efficiency, and robustness in SNNs, Spike Activation Map (SAM) for interpretable SNNs, and bio-inspired homeostatic plasticity for optimal initialization. Understand the importance of end-to-end training in artificial neural networks and explore innovative solutions like surrogate gradients and Holomorphic Equilibrium Propagation for computing exact gradients through finite size oscillations.
Neuromorphic Engineering Algorithms for Edge ML and Spiking Neural Networks
Overview
Syllabus
Intro
JOHNS HOPKINS UNIVERSITY
Spiking for tinyML
Batch Normalization Through Time (BNTT) for Temporal Learning
BNTT: Energy Efficiency & Robustness
Training SNNs for edge with heterogeneous demands
Spike Activation Map (SAM) for interpretable SNN
Spiking neurons are binary units with timed outputs
End-to-end training is key for artificial neural networks
Solution: Replace the true gradient with a surrogate gradient
Surrogate gradients self-calibrate neuromorphic systems when they can access the analog substrate variables
Fluctuation-driven initialization and bio-inspired homeostatic plasticity ensure optimal initialization
Holomorphic Equilibrium Propagation Computes Exact Gradients Through Finite Size Oscillations
Technical Program Committee
Taught by
tinyML
