New work from Guiqiu Liao & PCASO Lab accepted to #MICCAI2025
We bring object-centric learning to the OR—predicting future slot configs to discover tools & anatomy, opening ways for interpretable surgical scene understanding with low computational demands
#SurgicalAI#SurgicalDataScience
Thrilled to see our TinyRNN paper in @nature! We show how tiny RNNs predict choices of individual subjects accurately while staying fully interpretable. This approach can transform how we model cognitive processes in both healthy and disordered decisions. doi.org/10.1038/s415...
🚨 New Challenge Alert! 🚨
The Open Surgical Suturing Skills Challenge is back — now with brand new data and a tracking task! 🧵🛠️
Think you can assess skill or track hands & tools in surgical videos? Come join us!
🔗 More info: www.synapse.org/Synapse:syn6...
📣 Follow updates: @hoffoh.bsky.social
LastCall MICCAI #MedicalImaging Computing in Resource Constrained Settings Workshop & KI (MIRASOL) event.fourwaves.com/mirasol/pages TravelGrants will be available #miccai
You still have time to submit your work to the FAIMI Workshop at MICCAI 2025 focused on fairness, bias, and equity in AI for medical imaging.
Check out the important dates and details:
👉 faimi-workshop.github.io/2025-miccai-...
#FAIMI #MICCAI2025 #MedicalImaging #FairAI #HealthcareAI #AIethics
Guiqiu Liao, Matjaz Jogan, Eric Eaton, Daniel A. Hashimoto: FORLA:Federated Object-centric Representation Learning with Slot Attention https://arxiv.org/abs/2506.02964 https://arxiv.org/pdf/2506.02964 https://arxiv.org/html/2506.02964
Object-centric federated learning across heterogeneous datasets, new work from PCASO lab
Glad to see that our recent paper on digital twins for biomedical systems aligns with this view: simple correlational data cannot be used for digital twins and, accordingly, for building decisions.
www.nature.com/articles/s41...
A contour enhanced funnel plot demonstrating publication bias
The 'datasauraus' plot series, demonstrating the need to visualise data
A slide demonstrating that bar plots can hide important information, such as outliers and skewness
A description of raincloud plots for illustrating mean differences
Today I'm delivering an open science lecture for our master students. These kind of lectures are one of the most enjoyable parts of my job
PCASO’s Guiqiu Liao gave a great presentation on VDST-Net at WACV25. Learning segmentation masks from extremely weak labels reduces cost of annotating objects in multi-dimensional data (videos, volumetric data).
Video: video.computer.org/WACV-Posters...
#WACV2025 #surgicaldatascience
For the new science refugees here, I invite you to slip into the warm embrace of 20+ hours of free soothing lectures on scientific methods, a stats course for people who hate statistics, would rather be doing research or planning revolutions. Try the 1st vid - it's not what you expect. #stats 🧪
How do we judge the viscosity of materials? It turns out it depends not only on how they move, but also on how they look. Our latest work with the incredible Jeffrey Martin now out in RSOS royalsocietypublishing.org/doi/10.1098/...
Do large language models develop "emergent" models of the world? My latest Substack posts explore this claim and more generally the nature of "world models":
LLMs and World Models, Part 1: aiguide.substack.com/p/llms-and-w...
LLMs and World Models, Part 2: aiguide.substack.com/p/llms-and-w...
Can LLMs be used to discover interpretable models of human and animal behavior?🤔
Turns out: yes!
Thrilled to share our latest preprint where we used FunSearch to automatically discover symbolic cognitive models of behavior.
1/12
Modern-Day Oracles or Bullshit Machines?
Jevin West (@jevinwest.bsky.social) and I have spent the last eight months developing the course on large language models (LLMs) that we think every college freshman needs to take.
thebullshitmachines.com
"One thing is certain: The changes we make ourselves will be healthier than the ones our adversaries demand."
New work for @undark.org:
undark.org/2025/02/06/o...
"Imagine a world in which an A.I. can analyze your reading patterns and alert you that you’re about to buy a book where there’s only a 10 percent chance you’ll get past Page 6..."
Fuck off. That's not empowerment, that's outsourcing your soul. That's outsouling.
www.nytimes.com/2025/01/25/o...
New work led by PCASO Lab’s stellar Guiqiu Liao: Slot-BERT introduces long-range unsupervised learning of object-centric surgical video representations. By combining slot attention with masked prediction it robustly scales to long videos across multiple surgical domains arxiv.org/abs/2501.12477
I wrote about the concept of agency (both human and artificial) in the year 2025. gracewlindsay.com/2025/01/24/2...
Deficient Executive Control in Transformer Attention www.biorxiv.org/content/10.1101/2025.01....
Our new Harvard/Stanford study out today in Nature Medicine! Our results show language models, while strong at medical exams, stumble when they need to gather patient information through natural conversation - a key finding as healthcare explores AI integration.
How to solve? Enter CRAFT-MD.
Amazing collective problem solving in ants:
Bar chart of different barriers to healthcare
How do disparities in healthcare access affect ML models? 💰📉🧐 We found that low access to care -> worse EHR data quality -> worse ML performance in a dataset of 134k patients. Work with Anna Zink (on the faculty job market rn!) + Hongzhou Luan, presented at #ML4H2024
A short list of tips for keeping a clean, organized ML codebase for new researchers: eugenevinitsky.com/posts/quick-...
When we enter a new environment, we use visual input to rapidly build an internal model of the local spatial environment. How does our brain do this? We review past literature and suggest some new ways forward in our new review in @currentbiology.bsky.social: authors.elsevier.com/sd/article/S...
Yusuke Hosoya, Masanori Suganuma, Takayuki Okatani
Rethinking Annotation for Object Detection: Is Annotating Small-size Instances Worth Its Cost?
https://arxiv.org/abs/2412.05611
Tried out Rebiber (github.com/yuchenlin/re...) recently—a tool for cleaning up and standardizing bib entries. Extremely helpful for dealing with citation inconsistencies and ghost arXiv entries. Worth checking out if you’re managing a lot of references. #ResearchTools #BibTex
Good news everyone! A new version of graph-tool is just out! @graph-tool.skewed.de
graph-tool.skewed.de
Graph-tool is a comprehensive and efficient Python library to work with networks, including structural, dynamical, and statistical algorithms, as well as visualization. 1/N
#networkscience
A great overview of probabilistic segmentation including modeling of observer variability in healtcare use cases
