Summer Deep Reading on Deep Learning (Transformer Circuits)

Matthew O'Meara, PhD and his lab are hosting what we're calling Summer Deep Reading on Deep Learning. This year it will be a multi-part journal club on Transformer Circuits. We're going to start with Grokking and touch on lazy/rich training regimes, thermodynamics and phase transitions. The aim is to make the math accessible while getting into advanced topics. 

All are welcome, please pass this announcement to folks in your lab or anyone else you think may be interesting.

Over the course of the series, we are going to explore the use of generative and agentic AI and how we can use it to engage with new ideas and learn from each other. So each week we'll suggest exercises covering prompting strategies, ClaudeCode, Skills/MCPs, agentic workflows, and rigor and reproducibility. As a finale, for those that can attend, we'll hold a multi-day hackathon where we can work together to integrate what we've learned into resource for others.

Schedule and Logistics

We'll meet every-other week on Fridays at 10-11am location 4B700, where the coordinates may vary depending on availability.

• Friday 5/22
• Friday 6/5
• Friday 6/19
• Friday 7/3
• Hackathon 7/15-17th

* Each session we'll cover a paper journal as a journal club, and share what we learned about and through the AI tools.

 * For coordination and up-to-date information, please join the M-Community Group and this will add you to the #summer-deep-reading-deep-learning channel on slack. Note: It may take up to 24 hours for Mcommunity to sync with Slack. 

Expectations:

• Read the paper and be curious