Machine Learning Theory and Application

Overview

This repository is inspired by the Hung-yi Lee Machine Learning Course 2021. In that course, professor mainly focuses on the theory knowledge of machine learning and TAs will assign homework to students.

However, the explanations on homework is not quite enough for real-world application. Meanwhile, there are many practices can only be understood via coding. As a result, this repository is constructed.

This repository is not only about homework itself, but also contains many useful technical tutorials to show how these machine learning methods can be applied in daily situations.

If you are also looking for theory notes, please refer to this page.

Last but not least, there will be code update along with the lecture process.

Content

• COVID-19 Cases Prediction: basic application of regression
• Phoneme Classification: use simple neural network to classify phoneme
• Image Classification: use CNN to classify 11 kinds of food images
• Advanced Phoneme Classification: use transformer to classify phoneme again
• Machine Translation: use transformer to complete seq2seq task
• Anime Face Generation: use GAN to generate images
• Question Answering: fine-tune BERT for its adaptation to QA task
• Anomaly Detection: use Auto-encoder to detect fake human faces
• Explainable AI: explain AI modules to convince people
• Adversarial Attack: generate images to disable well-trained models
• Domain Adaptation: classify images in two different plotting styles
• Reinforcement Learning: implement RL game models of OpenAI with Policy Gradient and Actor Critic
• Network Compression: use knowledge distillation and design architecture to help small model perform as well as big ones
• Life-long Learning: compare regularization-based methods including EWC, MAS etc., to help model go against catastrophic forgetting
• Meta Learning: use MAML as meta algorithm to find a good initialization for CNN-based classifier

Reference

• Homework of Hung-yi Lee Machine Learning Course, 2021.