Please log in to watch this conference skillscast.
Getting machine learning models ready for use on-device is a major challenge. Drag-and-drop training tools can get you started, but the models they produce aren’t small enough or fast enough to ship. In this talk, you’ll learn optimization, pruning, and compression techniques that keep app sizes small and inference speeds high. You’ll start by discovering flexible model architectures that meet performance and accuracy requirements across devices and platforms. You’ll then learn pruning and distillation techniques to optimize model performance and use quantization tools to compress models to a fraction of their original size. Finally, you’ll explore a practical example of this process as you create an artistic style transfer model that’s just 17kb. All of these techniques will be applied to mobile machine learning frameworks for Android, i.e. TensorFlow Lite.
YOU MAY ALSO LIKE:
- Deep Learning Fundamentals with Leonardo De Marchi (Online Course on 8th - 11th February 2021)
- Android Architecture with Jorge Ortiz-Fuentes (Online Course on 22nd - 24th February 2021)
- Winter is Coming – Even Faster (SkillsCast recorded in November 2020)
- Building a Custom Type Provider (SkillsCast recorded in October 2020)
Creating smaller, faster, production-worthy mobile machine learning models for Android
Jameson Toole
Dr. Jameson Toole is the CEO and cofounder of Fritz—helping developers teach devices how to see, hear, sense, and think.