Node.js <=> Machine Learning

on YouTube ↗31m

" summary

Client-side image captioning runs inference in the browser by converting Python models into JSON and optimized weights with webDNN, loading a CNN (example: ResNet-50 → LSTM pipeline) and generating translated captions inside a Chrome extension. Training stays on servers; inference moves to WebGL/WebAssembly/WebGPU where supported (WebGPU currently limited). Message passing uses content and background scripts. Cost example: server GPUs ran about $24/day GPU. Performance, compatibility, privacy, and model size decide architecture.

AI-written summary · about

" chapters (11)

1. 0:00Start
2. 6:33Server-side ML: The typical setup
3. 6:53Challenges with server-side ML
4. 8:06But there's good news End-user devices are becoming increasingly powerful
5. 9:04client-side ML: A new approach
6. 10:27Let's try out client-side ML Example: Image captioning
7. 17:36Architecture overview
8. 17:51Convert model
9. 19:19Load model into browser and run inference
10. 21:17Calling the model from a chrome extension
11. 24:07Possibilities for client side ML

" related talks

• 23mNode.js = Machine LearningNikhila Ravi · NodeConf EU 2017
• 39mMachine Learning Magic for Your JavaScript ApplicationYannick Assogba, Sandeep Gupta · Google I/O 2019
• 26mMachine Learning and AI for Angular DevelopersJerry Kurata · NG-MY 2019
• 20mTensorFlow.js Bringing Machine Learning to the Web and BeyondNick Kreeger, Nikhil Thorat · JSConf EU 2019
• 15mThe future of machine learning using JavaScriptDoni Rubiagatra · CityJS Singapore 2023
• 24mTensorFlow.js + NodeNick Kreeger, Yannick Assogba · EmpireJS 2018
• 34mA Powerful Introduction to Machine Learning with JsOliver Alejandro López · JSConf Mexico 2020 Virtual Edition
• 31mIntroduction to Machine Learning in Tensorflow.jsNick Kreeger · ForwardJS 2019