Deep Learning for Mobile Devices

Explore the challenges and solutions for implementing deep learning on mobile devices in this 44-minute conference talk. Learn about convolutional neural networks (CNNs) and their potential applications in smartphone and wearable device applications. Discover strategies to overcome memory and power constraints, including building mobile-friendly shallow CNN architectures. Gain insights from real-time demos and case studies from major tech companies like Google, Microsoft, and Facebook. Delve into topics such as model pruning, quantization, neural architecture search, on-device training, and federated learning. Understand the importance of latency, energy considerations, and hardware limitations in mobile deep learning implementations. Walk away with practical knowledge to leverage deep learning techniques in resource-constrained mobile environments.

Introduction
Introducing Seafood
Welcome
Why do we care
Latency
Push notifications
Training a model
Finetuning
Tensorflow
Hardware
Energy Considerations
Model Pruning
Quantization
Pocket Flow
Neural Architecture Search
Training Time
Comparison
Ondevice training
federated learning
recap
book
Questions