Overview
Explore the fundamentals of lightweight neural network architectures in this comprehensive technical talk that delves into the design and optimization of efficient networks for low-power devices like phones and TVs. Learn about specialized networks such as MobileNetV3, FBNet, and BlazeFace while understanding the engineering principles behind Fire modules and Squeeze-and-Excitation layers. Discover state-of-the-art methods for determining model parameters, architecture layer width, and depth using approaches like EfficientNet and Model Rubik's Cube algorithms. Master techniques for accelerating and optimizing neural network layers, making this 77-minute presentation essential for engineers and developers interested in creating efficient, lightweight neural networks for resource-constrained environments.
Syllabus
tinyML Talks: Lightweight Neural Network Architectures
Taught by
EDGE AI FOUNDATION