AE-PSL
Align first. Compress every training step
Research & open source
Research in decentralized learning and efficient foundation models.
Towards agentic AI at the edge.
14 projects
Enforce unlearning while keeping the request private
Encrypt fewer values. Keep federated updates protected
Many local tasks. One shared task vector
Multimodal fine-tuning with lightweight edge clients
Adapt foundation models by communicating sparse mask changes
Exchange codebooks to make federated communication smaller
Compress both directions of federated training
Learning together when labels cannot be trusted
Understanding invisible backdoors in collaborative learning
Try another term or clear the filters.