PhD Candidat @uni_bamberg_of π | Exploring Robust Deep Learning for Medical Applications π₯ | Coffee Addict βοΈ | Always Curious, Forever Learning π | β½οΈπ₯Ύ
Bamberg, Germany
Joined January 2018
- Tweets28
- Following123
- Followers39
- Likes49
If you are at #MICCAI2026 and work on parameter-efficient fine-tuning, drop by my poster at the EMA4MICCAI workshop.
Tomorrow, Thursday 1 October
Poster session, 9:50 to 10:30
Room Etoile A (Floor U), poster 23
arxiv.org/abs/2607.29462
5/ MoPET (EMA4MICCAI workshop Β· Poster)
Parameter-efficient fine-tuning needs one adapter per task, and merging them risks conflicting gradients. MoPET routes each input through a few low-rank experts in a frozen foundation model.
arxiv.org/abs/2607.29462
If you are at #MICCAI2026 and work on OOD detection, come by @francescodis_'s main-conference poster.
Tomorrow, Monday 28 September
Poster Session 1, 16:00 to 18:00, board M-PM-011
arxiv.org/abs/2606.22649
See you there!
1/ MaRS (main conference Β· Poster)
Reconstruction-based OOD detection in latent space underperforms because of how residuals are scored, not how well they reconstruct. MaRS scores them with a variance-aware Mahalanobis distance.
Led by @francescodis_
arxiv.org/abs/2606.22649
If you are at #MICCAI2026 and work on OOD detection with VLMs, drop by @ShyamNandanRai4's long oral at the UNSURE workshop.
Tomorrow, Sunday 27 September
Long Oral 1, 12:05 to 12:25, Room Boston
Poster 27, 13:10 to 13:55
arxiv.org/abs/2609.08524
4/ Layer Selection in VLMs (UNSURE workshop Β· Oral)
Zero-shot OOD detection with VLMs reads final-layer embeddings only. Intermediate layers carry complementary signal, selected by multi-resolution entropy estimation.
Led by @ShyamNandanRai4
arxiv.org/abs/2609.08524
If you are at #MICCAI2026 and work on polyp detection, drop by my oral at the EndoLINA workshop.
Tomorrow, Sunday 27 September
Oral Session 1, 9:10 to 9:50 (second talk)
Room Bruxelles
Poster of the same work, 10:00 to 10:30
arxiv.org/abs/2608.13711
See you there!
3/ TRUE-Colon (EndoLINA workshop Β· Oral)
Polyp detectors are built on curated clips that omit the long negative stretches of real colonoscopies. On 60 full procedures, clip-trained detectors collapse while procedure-trained ones hold up on both.
arxiv.org/abs/2608.13711
Happy to see five papers I was involved in accepted at #MICCAI2026 in Strasbourg π«π· across the main conference and workshops.
Details and links to all five in the thread below, one post per paper.
π Looking for a novel method to enhance classification robustness under real-world label noise?
Check out this work "An Embedding is Worth a Thousand Noisy Labels" led by @francescodis_ which was published in @TmlrOrg.
π openreview.net/forum?id=X3gSβ¦
π github.com/francescodisalvo0β¦
π Looking for a comprehensive benchmark for the MedMNIST+ database?
Check out our latest paper, "Rethinking model prototyping through the MedMNIST+ dataset collection" now published open-access in @SciReports
π nature.com/articles/s41598-0β¦
π github.com/sdoerrich97/rethiβ¦
If you are at #BMVC2024 in Glasgow π΄σ §σ ’σ ³σ £σ ΄σ Ώ and are interested in robustness for medical image diagnosis, feel free to drop by my oral tomorrow.
ποΈ Wednesday, October 27, 2024
π M3
π 11 AM
πΌοΈ Unsupervised Feature Orthogonalization for Learning Distortion-Invariant Representations
π Excited to introduce: "Unsupervised Feature Orthogonalization for Learning Distortion-Invariant Representations", wich will be presented as an oral at the #RROW Workshop at #BMVC2024 π΄σ §σ ’σ ³σ £σ ΄σ Ώ.
π: arxiv.org/abs/2409.12276
π: github.com/sdoerrich97/unoraβ¦
π₯: @francescodis_, @LedigChr
If you're interested in histopathology and are attending #MICCAI2024 in π²π¦, feel free to drop by my poster tomorrow.
ποΈ Tuesday, 8th of October 2024
π 10:30 - 11:30 AM
πΌοΈ Self-supervised Vision Transformer are Scalable Generative Models for Domain Generalization [T-AM-233]
Another One!π’
Excited to share that our paper "Self-supervised Vision Transformer are Scalable Generative Models for Domain Generalization" is accepted at #MICCAI2024!
Want to know more?
π: arxiv.org/abs/2407.02900
π: github.com/sdoerrich97/vits-β¦
@MICCAI_Society
@xAILabBamberg
π Excited to introduce: "Unsupervised Feature Orthogonalization for Learning Distortion-Invariant Representations", wich will be presented as an oral at the #RROW Workshop at #BMVC2024 π΄σ §σ ’σ ³σ £σ ΄σ Ώ.
π: arxiv.org/abs/2409.12276
π: github.com/sdoerrich97/unoraβ¦
π₯: @francescodis_, @LedigChr
πΊExcited to share yet another work of our #xAILabBamberg!
This work, led by @francescodis_, is for all of you #ComputerVision folks out there interested in #privacy preserving and will be featured at this year's #BMVC conference in Glasgow, UK! π΄σ §σ ’σ ³σ £σ ΄σ Ώ
π: arxiv.org/abs/2408.00639
π£ Our paper "Privacy-preserving datasets by capturing feature distributions with Conditional VAEs," has been accepted at #BMVC2024 π΄σ §σ ’σ ³σ £σ ΄σ Ώ
A huge thank you to all co-authors: David Tafler, @DoerrichS, and @LedigChr.
Paper: arxiv.org/abs/2408.00639
π xAILAB Bamberg goes MICCAI - Part II π
Another work from our group was accepted for #MICCAI2024! This work led by @francescodis_ will be presented at the Workshop on Advancing Data Solutions in Medical Imaging AI #ADSMI.
π arxiv.org/pdf/2406.17536
π github.com/francescodisalvo0β¦
π£ I am happy to share that our paper "MedMNIST-C: Comprehensive benchmark and improved classifier robustness by simulating realistic image corruptions" has been accepted to the #ADSMI workshop at #MICCAI2024!
@MICCAI_Society @DaliWorkshop
Paper: arxiv.org/pdf/2406.17536
Another One!π’
Excited to share that our paper "Self-supervised Vision Transformer are Scalable Generative Models for Domain Generalization" is accepted at #MICCAI2024!
Want to know more?
π: arxiv.org/abs/2407.02900
π: github.com/sdoerrich97/vits-β¦
@MICCAI_Society
@xAILabBamberg