Explainable Machine Learning

A medical LLM can give the right answer for the wrong reasons. Delighted our work is now published in Nature Communications Medicine — evaluating 8 LLMs on 1,269 unstructured seizure descriptions, with a practical guideline to use AI in clinical setup 🧵 doi.org/10.1038/s43856…

Results: FINER-Tuning improves performance on our FINER benchmarks, transfers to other hallucination benchmarks, and even boosts general MLLM ability. On InternVL3.5-14B, we get up to +24.2% on FINER without hurting overall capability.

We also introduce FINER-Tuning: generate fine-grained positive/negative phrases from dense captions, turn them into accept/reject QA pairs, and train MLLMs with DPO so they learn to spot subtle mismatches instead of hallucinating.

Happy to share that we have 3 papers accepted (one oral) at #CVPR2026, see you in Denver! Please see the thread for details👇

Excited to share that our paper “Diamond Maps: Efficient Reward Alignment via Stochastic Flow Maps” won the Best Paper Award at the #ICLR2026 ReALM-GEN Workshop 🎉✨ Check out the paper & code 👇 📕 arxiv.org/abs/2602.05993 💻 github.com/PeterHolderrie…

Accurate reward guidance in diffusion models has been a long-standing challenge. We propose Diamond Maps 💎, through stochastic flow maps we make reward guidance both accurate and scalable.

3/ Post-hoc Probabilistic Vision-Language Models @_antonbaumann, @ruili_pml, Marcus Klasson, Santeri Mentu, @ShyamgopalKart1, @zeynepakata, @arnosolin , Martin Trapp [Paper]: arxiv.org/pdf/2412.06014 🕑 Fri, Apr 24, 2026 • 6:30 AM – 9:00 AM PDT 🚩 Pavilion 3 P3-#313

2/ Are Reasoning LLMS Robust to Interventions on Their Chain-of-thought? @avrecum, Leander Girrbach, @zeynepakata [Paper]: arxiv.org/pdf/2602.07470 🕑 Fri, Apr 24, 2026 • 6:30 AM – 9:00 AM PDT 🚩Pavilion 3 P3-#918

Our EML members are heading for Rio🇧🇷 for #ICLR2026! 🌴We’re excited to share our latest research — three poster presentations. Check out the thread below and come say hi to our authors