Deep Learning in JS
summary
Shows that TensorFlow.js runs pre-trained models in the browser and supports transfer learning using poseNet and MobileNet (internal layer conv_pw_13_relu). Demonstrates a responsive photo grid that recenters faces by tracking the nose keypoint, and a Pac‑Man controller trained with webcam examples (flatten → dense 100 relu → dense softmax, Adam optimizer, categorical crossentropy). Performance: 8 FPS on a 2014 MacBook at low quality, under 1 FPS at higher quality; models are large but client‑side inference preserves user data.
AI-written summary · about
chapters (13)
- 0:00Introduction to deep learning
- 2:52TensorFlow and ecosystem
- 7:12What is a tensor?
- 9:01Models and training
- 15:51Activation functions
- 20:27Pre-trained models
- 23:40Pose detection demo
- 29:48Face centering with PoseNet
- 30:08Transfer learning basics
- 39:57Pac-Man with an elephant
- 43:53How neural networks work
- 46:51Ethics and future of AI
- 49:00Q&A session
related talks
32mDeep Learning in JS
20mTensorFlow.js Bringing Machine Learning to the Web and Beyond
41mTensorFlow.js 101: ML in the Browser and Beyond
35mThe Future of JavaScript & Machine Learning
34mA Powerful Introduction to Machine Learning with Js
34mA Powerful Introduction to Machine Learning with Js
15mMachine learning for next gen web apps with TensorFlow.js
31mDeep Learning at the Edge with TensorFlow.js