FEM: Machine Learning with JavaScript
April 18, 2024Course: https://frontendmasters.com/workshops/machine-learning/
https://fem-ml-workshop.netlify.app/
https://github.com/charliegerard/fem-ml-workshop
https://charliegerard.dev/projects
Basics
https://github.com/tensorflow/tfjs-models
Overfitting: model is too close to training data, value of 1. Can't predict.
- Supervised learning: given labelled data.
- Unsupervised learning: given a dataset without labels. Use is clustering.
- Semi-supervised learning: combines supervised and unsupervised. Can achieve higher accuracy.
- Reinforcement learning: training provides rewards and punishment. Used in gaming.
Pre-trained Models
Finding pre-trained models
- Kaggle and more specifically TensorFlow.js models
- TensorFlow.js models
- Hugging face
Choosing a Model
- What are you building?
- Quality of the training data
- Quantity of the training data
- Can you have access to the list of classes/labels?
Tools
Transfer Learning
Reusing pre-trained model for a new task.
Quick start
https://teachablemachine.withgoogle.com/train/image
https://experiments.withgoogle.com/tiny-motion-trainer
Building a Model
https://cs231n.github.io/convolutional-networks/
Split training and test data.
https://neurosity.co/tech-specs
https://www.youtube.com/playlist?list=PLOU2XLYxmsILr3HQpqjLAUkIPa5EaZiui
https://www.linkedin.com/pulse/welcome-tensorflowjs-monthly-newsletter-jason-mayes/