
Learning from multimodal data for healthcare and beyond | postdoc @UNC under Tianlong Chen
Durham, NC
Joined August 2016
- Tweets238
- Following673
- Followers471
- Likes739
Pinned Tweet
I asked Claude (Opus + Fable 5) to work on unsolved problems in math, and it found the new lowest-ever counterexample to Borsuk's conjecture, N=63. The last one found was in 2014 (N=64). I posted Claude’s writeup and verification code on my website: nickk124.github.io/borsuk (1/2)
UPDATE: @maaxgri notified me that they actually already found an N=63 counterexample assisted by GPT-5.5 3 months prior to this, with the same construction, different derivation! Please check out their work here: github.com/maaxgrin/borsuk-6…
So far, Claude verified the counterexample in four independent ways: an exact rational/algebraic-arithmetic certificate, two independently-written max-clique algorithms, an ILP solver, and CP-SAT. Actively working to get it verified by a proper mathematician!
I’m thrilled to announce that I will be joining the UNITES Lab under @TianlongChen4 as a postdoc in the UNC Chapel Hill CS Department starting in early 2026. My research will focus on multimodal and agentic AI for healthcare and AI for science (protein LMs and genomic LMs)!
FID has many limitations for medical images (e.g., downstream task misalignment). Introducing the Fréchet Radiomic Distance (FRD): a metric designed from the ground up for better characterizing medical images, led by me and @1richardobinna! Code/paper: github.com/RichardObi/frd-sc…
As part of this paper, we also present the most extensive framework for the meta-evaluation of medical image similarity metrics to date, available at github.com/mazurowski-lab/me…!
1
52
Thanks to our wonderful collaborators! Preeti Verma, Yuwen Chen, Hanxue Gu, Haoyu Dong, Yaqian Chen, Andrew Marshall, @lidiagarrucho, @KaisarKushibar, Daniel Lang, Gene Kim, @Dr_Lars_Grimm, John Lewin, James Duncan, @ja_schnabel, @ooliverdiaz, @KarimLekadir, and @MazurowskiPhD.
1
91
Introducing SLM-SAM 2, led by Yuwen Chen: a new video object segmentation method for medical imaging that speeds up annotation by accurately propagating labels from a single slice across the whole volume! [Code: github.com/mazurowski-lab/SL…, paper: arxiv.org/abs/2505.01854 ]
1
114
With a novel dynamic "short-long" memory, it outperforms SAM 2 and other models by >7% DSC on average, in segmenting various organs, bones, and muscles across modalities! Crucially, it exhibits better robustness to the problem of "over-propagation" of annotations through slices.
1
45
Thanks for reading! Co-authors: Zafer Yildiz, Qihang Li, Yaqian Chen, @HaoyuDong_ , @Hanxue99893888 and @MazurowskiPhD
37
Excited to share my NeurIPS @scifordl paper on how hidden repr. intrinsic dimension evolves through model depth! openreview.net/forum?id=trcU…
A surprising finding: the peak intrinsic dim. is ~a constant fraction of the input data’s intrinsic dim., across diverse datasets + models! 1/3
I also found that the representation intrinsic dim. peaks consistently earlier in medical image models compared to natural image models, pointing to a difference in the abstractness of task-relevant features between these domains. 2/3
1
2
69
Nick Konz retweeted
We (@JanWitowski and I) co-founded a startup! We want to personalize cancer treatment for every patient using AI.
It's a fascinating research journey. I am very proud of what we achieved as a team. There is still so much to be done. Join us! jobs.ashbyhq.com/ataraxis-ai
Today, we are launching Ataraxis AI to fulfill the promise of precision medicine.
We are building AI-native, multi-modal tools that help physicians personalize treatment by predicting patient outcomes and treatment response.
businesswire.com/news/home/2…
Want to generate medical images with better anatomical control? Or just want an easy codebase to train diffusion models? Come check out my poster num. 24 at #MICCAI2024 any time today!
Nick Konz retweeted
Join us at our poster T64 at #MICCAI2024 to talk about DinoBloom 🦖🌺 the first feature extractor in hematology for bloodsmears🩸
All models are available at github.com/marrlab/DinoBloom
Developed w @ValentinKoch_, supervised by @p_tingying @MarrCarsten @HelmholtzMunich @TU_Muenchen












