Our implementation in Jax can be found on Github at github.com/cnguyen10/pl2d

This research is a part of the PecMan project (sites.google.com/view/pecmanp...) funded by EPSRC - UKRI, led by Professor Gustavo Carneiro from CVSSP - University of Surrey, and in collaboration with Dr Toan Do from @monashuniversity.bsky.social

A workload constraint is also integrated into, allowing the system to distribute workload evenly across all experts (otherwise, the system will not learn and defer all samples to the best team member, causing unfairness and burnout).

This paper addresses the problem of missing annotations made by human experts, meaning that each human expert annotates only a part of the training dataset.

Excited to share that our research in "Probabilistic learning to defer" (openreview.net/forum?id=zl0...) has been accepted as an oral presentation at #ICLR2025.