Tracking in google doc: https://docs.google.com/document/d/1XR3COq9sQGcVATJj7Lx8cYa-7q-VmkOXWpqMar4JJSs/edit?usp=sharing

Related past workshops

• https://federated-learning.org/fl@fm-neurips-2024/
• https://sites.google.com/view/mcdc2025/home

Overleaf link for the proposal: https://www.overleaf.com/9468236156rrssvwqxwggr#fef54c

Topics

• Communication efficient training (DP, PP, CP, etc)
  • architecture modifications
  • compression
• Training over globally distributed compute
  • fault tolerance
  • large-scale training frameworks to utilize volunteer compute (eg., swarm)
• Asynchronous optimization (DP, PP, etc)
  • including model parallel scheduling methods
• Adversarial or model extraction attacks
• Robustness to malicious actors during training
  • Byzantine robustness optimization
  • Verification
• Blockchain + AI
• Efficient decentralized inference
  • test time scaling
  • distillation
• Efficient post-training for decentralized models
• Continual learning of foundational models
  • Progressively growing models
  • training beyond post-training

Advisors

• Phil
• Simon - confirmed
• Steve