Projects

These are some of the projects you can be involved in as a member of the club.

•   AI Fencing Referee

  Train a model which can visually indicate priority in a bout to aid in refereeing, analysis, and viewership. Benchmark model metrics against existing implementations. Future goals: Priority + Fencer Move Classification, Natural Language Reasoning

•   Fast Upscaler

  Create a model that can upscale video feed in real time. Use less computational power to achieve higher image quality

•   Block Graph

  Utilize graph structures to explore alternative arrangements of transformer blocks. Train models more effectively on multiple datasets through pre-training components on common subtasks.

•   Agar.io Bot

  Train ML algorithm to play Agar.io through reinforcement learning.

•   Multivariate Information Bottleneck: RL

  Information theoretic tools for state analysis in RL.

•   Motion Prediction in Traffic

  Given 1 second traffic video and agent(s) in a traffic scene, the model will predict the agents' trajectories 8 seconds into the future. Use transformers to encode traffic context and predict bounding box

•   Face Recognition

  Simple Facial Recognition app for beginners.

•   Bugs, Be Gone!

  Art of Conventional Bug-Squashing with LLMs.

•   Forward

  Job searching assistance which uses text extraction methods to help with resume writing based on job descriptions.

•   Stanford's Ribonanza

  Create a model that predicts structures of any RNA molecules for Stanford's Kaggle competiton.

•   Predictive Crime Mapping

  Given police service calls for a specific city, assess how safe your current area is. Identify which crimes are most likely to occur by taking into account spacial and temporal features.

•   Medical LLM

  This project aimed to deploy an LLM-based application, with a subsequent expansion to incorporate CI/CD pipelines using GCP and medical data. We explored LLM frameworks such as LangChain and Llama Index, then prompt engineering, fine-tuned LLM, and grounded the model using RAG to mitigate hallucination.

•   Breast Cancer Classification

  Classify subtypes of cancer. Trained on histopathology images.

•   3D Point Cloud Segmentation

  Object classification and segmentation using KITTI 3D Benchmark, a dataset for Semantic scene understanding using LiDAR Sequences.

•   LUX AI

  Train AI agents using reinforcement learning to compete and collaborate in resources gathering/ allocation problem. The agent is then sent off to fight other competitors within a Kaggle competition.

•   rideCare

  Multi-label sentiment analysis to identify threats. Makes use of Natural Language Processing, and Time Series Analysis. Deployed on a raspberry Pi as a ride safety application.

•   Image Retouching Service with GAN

  An image retouching software, which can change overcast weather picture to sunny day, change season of your picture, and apply character morphing.