Machine Learning

• We have a $50 prize for anyone who builds a really cool ML project. The prize may not go to anyone or may go to multiple people. You can seek out Vihaan Sondhi, Kyle Lee, and/or Dean Chen at the event to see if your project idea might qualify or email one of them in advance at first_name dot last_name @teenhacksli.com.

• Fast.ai is definitely a good choice for a library if you're just getting started with machine learning.

• Kaggle competitions are great for finding datasets and getting ideas for things to work on.

• Image Classification

  • Fast.ai is definitely what you want to use here. Look through their tutorials to get an idea.

• Language Modeling (aka generating text)

  • Learn using Andrej Karpathy's great introductory course. It goes from scratch to building a transformer. Extending on any of these significantly and/or implementing all of the coding exercises would be a valid project that would be both fun and an incredible learning experience.

• Reinforcement Learning

  • The following two resources are great for both learning and finding ideas for projects to work on.

  • https://arena3-chapter2-rl.streamlit.app/ -> if you're already comfortable with rl, look at the end of each section for bonus exercises if you want ideas.

  • https://web.stanford.edu/class/psych209/Readings/SuttonBartoIPRLBook2ndEd.pdf -> implement algorithms from the canonical tabular RL textbook.

• General ideas if you already have experience:

  • Get started on an autodiff library like this

  • Implement a neural net class from scratch in Numpy. Include normalization (batch norm, layer norm), gradient clipping, lr-scheduling, dropout, several different optimizers, activation functions, and more.

• For relatively cheap GPU rental, check out Vast.ai and Lambda Labs, but Google Colab should be sufficient for most projects.

  • If you anticipate that paying for GPU rental will be a bottleneck for you, email us at [e-mail] (cc [e-mail]) and we'll see if we can help you out.