ML+Coffee: LLM Paper Discussion and ML Project Discussions

Hosted by the ML+X community, the monthly ML+Coffee social brings together machine learning (ML) practitioners across campus so that we can connect with one another, discuss and work on ML projects, and enjoy some caffeinated refreshments ☕. Attendees are encouraged to bring their laptops and/or any questions about ML.

ML+Coffee will take place from 9-11am on 10/9, in room 1145 of the Discovery Building. Kindly register to help us plan our catering needs.

1. 9-10am (Paper Discussion): We will begin with a 15-20 minute overview of the paper "Scaling LLM Test-Time Compute Optimally Can Be More Effective than Scaling Model Parameters" by Charlie Snell et al. The paper discusses how using more compute during inference—such as running multiple iterations or exploring various answers—can enhance model performance without increasing its size. By leveraging additional compute resources at test time, rather than during training, this approach can significantly improve accuracy and efficiency. After the summary, we will follow up with communal discussions on the practical implications of these techniques for ML projects, and how this method could enhance model deployment, particularly in resource-constrained scenarios.
2. 10-11am (Social, Discuss Ongoing ML Projects): ML+Coffee offers a casual and social atmosphere where ML practitioners can problem-solve with one another. Beginners and experts, alike, are invited to join the discussions and network with the community.