Summary

This repository provides shared code solutions and insights for Professor Hung-Yi Lee's Machine Learning course assignments (2023 edition). It targets students seeking to understand and achieve high-performance baselines ("Boss baseline") for practical ML tasks, offering a valuable reference for learning and implementation.

How It Works

The project structures solutions for each homework assignment (HW01-HW07) into tiered baselines: Medium, Strong, and Boss. This tiered approach allows users to progressively learn and implement more sophisticated techniques. Sample code and benchmark results are included, with code updated throughout the course to reflect evolving solutions and achieve target performance levels.

Quick Start & Requirements

• Clone the repository: git clone https://github.com/Hoper-J/HUNG-YI_LEE_Machine-Learning_Homework.git
• Prerequisites include Python and standard ML libraries (PyTorch, Hugging Face Transformers are implied). Specific homeworks leverage tools like Optuna, TensorBoard, and advanced architectures (Conformer, StyleGAN, LLMs), suggesting potential GPU requirements for training.
• Links to Kaggle competitions and related course materials are available within the assignment descriptions.

Highlighted Details

• HW01: Implements advanced ML techniques including optimizers (Adam, Momentum), regularization (L2), feature selection, cross-validation, and automated hyperparameter tuning via Optuna.
• HW02-HW04: Introduces TensorBoard, learning rate schedulers, BiLSTM, Conformer architectures, Self-Attention Pooling, and Additive Margin Softmax.
• HW05-HW07: Focuses on Transformer models, back-translation, LLM fine-tuning for QA, gradient accumulation, and image generation with StyleGAN, including FID/AFD metrics.

Maintenance & Community

Code is updated throughout the course duration. A related repository for "Generative AI Introduction" is linked. No specific community channels (Discord, Slack) or core maintainer details are provided.

Licensing & Compatibility

The repository's license is not specified in the provided README content. This lack of information may pose compatibility concerns for commercial use or integration into proprietary projects.

Limitations & Caveats

Some code may be adapted from previous years. Initial versions of HW03 did not reach the stated "Boss baseline." FID/AFD calculation functions in HW06 differ from the official JudgeBoi implementation. The repository serves as homework solutions rather than a general-purpose library.