cs231n Deep Learning for Computer Vision Review

Goal: aid people in building project portfolios for job search in AI/ML.

Demos are much easier to build in the era of LLMs

You get access to :
Free GPU time from AMD,
Professor OH
online content
other members to compare answers with the caveat LLMs may work better than any human collaboration.

Structure:
Review the lectures 1-2 per week and do a short 10m presentation in front of a group. This material composes more than 50% of the interview questions for AI/ML entry or mid level jobs if you dont have substantial open source contributions. Best to practice NN implementation and basics before interviews. Some of the TAs for this class originally took this class over 10+y ago and are constantly practicing the basics to improve.

Agenda:
cs131 Computer Vision
Deep learning + RL

Project Ideas. There are tremendous opportunities in CV and Vision LLMs. As an example the benchmarks on image segmentation pipelines need updating. They require additional work to establish performance/cost if the traditional algorithms are replaced with agent/LLMs.

GPU TIme: Free from AMD given a project proposal.
Implssible to make progress on OS code or projects wo substantial GPU time.

Computer Vision
https://stanford-cs131.github.io/winter2025/syllabus.html

multimodal foundation models.
https://www.youtube.com/@jbhuang0604/playlists

Ask Professor Huang
https://jbhuang0604.github.io/#open-office-hour
https://github.com/jbhuang0604/awesome-tips/blob/main/working-with-mentor.md
https://github.com/jbhuang0604/awesome-tips/blob/main/cold-emails.md

Basic CV exercises. Use the 2022 github to access hwx.ipynb
https://stanford-cs131.github.io/winter2025/

2022 public ipynb. The answers are provided automatically by cursor. Disable ai to do on your own
https://github.com/StanfordVL/CS131_release/tree/fall_2022/fall_2022

cs231n videos
https://www.youtube.com/playlist?list=PLoROMvodv4rOmsNzYBMe0gJY2XS8AQg16