AlgoPerf Workshop

How can we train neural networks faster with better algorithms?

February 11th & 12th, 2025, MPK 11, Hacker Way, Menlo Park

 

Workshop Description

The AlgoPerf Workshop focuses on training algorithms for deep neural networks and the question, “How can we train neural networks faster?” Training contemporary neural nets is a lengthy and costly process often involving extensive “babysitting” and brittle training recipes. The workshop will provide an open forum to discuss novel algorithms and best practices. Drawing from the insights of the recently concluded first iteration of the AlgoPerf: Training Algorithms benchmark competition, the workshop will feature spotlight talks from teams behind the winning submissions in both competition categories. In addition to spotlight and invited talks, the workshop will include multiple open roundtables and panel discussions offering attendees ample opportunities to share experiences, insights, and challenges related to training deep neural networks (see our Schedule). The goal is to bring together academic and industry researchers and practitioners in machine learning and optimization to discuss advancements and the future of training algorithms.

The workshop will take place on February 11th and 12th, 2025, at MPK 11, Hacker Way, Menlo Park, California, USA, hosted by Meta AI.

 

Interest & Registration

If you are interested in attending the AlgoPerf Workshop, please register your interest via the AlgoPerf Workshop Interest Form. Due to limited capacity, registration is limited. We will consider all registered interests and send out invitations according to our capacity.

 

News

• October 22, 2024:
    🖥️ The workshop website is live!

 

Thanks

Special thanks to Meta and MLCommons for supporting and sponsoring this workshop.