My year of reading in 2025: www.fregu856.com/post/year_of...
I read 113 papers in 2025, complete list: github.com/fregu856/pap...
Top 25 papers that I found particularly interesting and/or well written (in alphabetical order):
My year of reading in 2025: www.fregu856.com/post/year_of...
I read 113 papers in 2025, complete list: github.com/fregu856/pap...
Top 25 papers that I found particularly interesting and/or well written (in alphabetical order):
New preprint, work led together with Erik Thiringer:
Scanner-Induced Domain Shifts Undermine the Robustness of Pathology Foundation Models.
arxiv.org/abs/2601.04163
Isn't there a lot of noise in these decisions, just like for conference papers etc?
31.10.2025 16:02 — 👍 1 🔁 0 💬 1 📌 0Grattis!
10.10.2025 04:29 — 👍 1 🔁 0 💬 0 📌 0
I just reached 500 read papers on the Github repository I use to track and organize my reading:
github.com/fregu856/pap...
Very happy to have joined the group of David Clifton at IBME in Oxford as a postdoc, to work on machine learning for healthcare!
The group is also recruiting multiple new postdocs, please apply before August 18:
eng.ox.ac.uk/jobs/job-det...
The waiting area is also quite dull and gets really crowded, probably the worst part of my entire trip.
16.07.2025 10:08 — 👍 0 🔁 0 💬 1 📌 0I think it was already in the batch of papers I was given to rate, basically no pathology-related papers, for example. ICML was definitely better in this regard.
04.07.2025 04:16 — 👍 1 🔁 0 💬 1 📌 0Not super happy with my assigned NeurIPS papers this year, I found them less interesting/relevant than I usually do. But oh well, still quite solid papers overall, and I do think it's good to be forced to read papers from slightly different areas sometimes.
03.07.2025 11:48 — 👍 0 🔁 0 💬 1 📌 0
Our paper "Taming Diffusion Models for Image Restoration: A Review" has now been published, work led by Ziwei Luo:
royalsocietypublishing.org/doi/10.1098/...
New preprint, work lead by Ziwei Luo:
Forward-only Diffusion Probabilistic Models.
arxiv.org/abs/2505.16733
github.com/Algolzw/FoD
algolzw.github.io/fod/
Looks very useful, thanks for sharing!
02.04.2025 06:57 — 👍 1 🔁 0 💬 0 📌 0Nice, saw this on arxiv and thought it seemed interesting, might read this as well, thanks!
24.03.2025 07:42 — 👍 0 🔁 0 💬 0 📌 0
Nice, thanks!
I actually wrote "The one proper method change that seems to have the biggest effect is probably adding the KoLeo regularization loss term?" in my notes, so would be nice to read more about how that works.
The main thing definitely seems to be that they scale iBOT from ViT-L/16 trained on ImageNet-22k (14 million images) to ViT-g/14 trained on their LVD-142M dataset (142 million images).
Their model distillation approach is also interesting, distilling their ViT-g down to ViT-L and smaller models.
"We revisit existing discriminative self-supervised approaches [...] such as iBOT, and we reconsider some of their design choices under the lens of a larger dataset. Most of our technical contributions are tailored toward stabilizing and accelerating [...] when scaling in model and data sizes"
23.03.2025 15:21 — 👍 1 🔁 0 💬 1 📌 0
iBOT: Image BERT Pre-Training with Online Tokenizer (ICLR 2022)
DINOv2: Learning Robust Visual Features without Supervision (TMLR, 2024)
DINOv2 doesn't really add much methodological difference compared to iBOT, they give a good summary of what they do:
I've been trying to properly understand how/why DINOv2 works, and I think this is a good sequence of papers to read for that:
(BYOL) Bootstrap Your Own Latent: A New Approach to Self-Supervised Learning (NeurIPS 2020)
(DINO) Emerging Properties in Self-Supervised Vision Transformers (ICCV 2021)
I really didn't like the ICML review template though, why do we have to make things overly complicated? Please just give me some variation of Summary, Strenghts, Weaknesses, Questions, Detailed comments and Justification of rating!
14.03.2025 13:20 — 👍 0 🔁 0 💬 0 📌 0102 papers to be exact: www.fregu856.com#service
14.03.2025 13:19 — 👍 0 🔁 0 💬 0 📌 0Finished my 5 #ICML reviews, and realized that I now have passed 100 reviewed papers in total during my career. Actually feels like a pretty cool milestone!
14.03.2025 13:12 — 👍 6 🔁 0 💬 1 📌 0
First time I'm reviewing for MIDL. Quite interetsing papers overall, and I like the review template.
But, 8 pages in this template seems too short. Not enough space to actually do things properly (e.g., explain the method in detail ~and~ have an extensive experimental evaluation).
My year of reading in 2024: www.fregu856.com/post/year_of...
I read 99 papers in 2024. Complete list: github.com/fregu856/pap...
Top 15 favorite papers that I found particularly interesting and/or well-written (in alphabetical order):
Recent preprint: Evaluating Deep Regression Models for WSI-Based Gene-Expression Prediction.
arxiv.org/abs/2410.00945
Recent preprint: Evaluating Computational Pathology Foundation Models for Prostate Cancer Grading under Distribution Shifts
arxiv.org/abs/2410.06723