Attending ICML? Check out our work on confounder-free continual learning ♾️ Our R-MDN layer can be applied to most modern architectures, including ViTs, to remove spurious correlations on-the-fly

One of the core challenges continual learning addresses is data privacy protection. Deep learning models don’t just learn from data—they also store a surprising amount of it. That’s why unlearning is critical for safe, real-world deployment. Join our seminar tomorrow to dive deeper into this topic!

Join us this Thursday 8 May for:

U n l e a r n i n g
~ in continual scenarios ~

together with:
Zhenyi Wang, Sijia Liu, and Kairan Zhao

6 PM CET / 9 AM PT / 12 PM ET

Hope to see you there!!!

Registration here: stanford.zoom.us/meeting/regi...

I had a fantastic time last week attending @khipu-ai.bsky.social in Santiago 🇨🇱 where I had the opportunity to present my recent work on 𝐷𝑦𝑛𝑎𝑚𝑖𝑐 𝑇𝑟𝑎𝑖𝑛𝑖𝑛𝑔 𝑎𝑛𝑑 𝑀𝑜𝑛𝑖𝑡𝑜𝑟𝑖𝑛𝑔 𝑜𝑓 𝐴𝐼-𝑠𝑢𝑝𝑝𝑜𝑟𝑡𝑒𝑑 𝑀𝑒𝑑𝑖𝑐𝑎𝑙 𝐷𝑒𝑣𝑖𝑐𝑒𝑠 and participate on a panel discussion on the future of Computer Vision ❤️

Join us for our March ContinualAI seminar on 𝐸𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑡 𝑓𝑖𝑛𝑒-𝑡𝑢𝑛𝑖𝑛𝑔 𝑡𝑒𝑐ℎ𝑛𝑖𝑞𝑢𝑒𝑠 (𝑎𝑑𝑎𝑝𝑡𝑒𝑟𝑠, 𝐿𝑜𝑅𝐴, 𝑎𝑛𝑑 𝑏𝑒𝑦𝑜𝑛𝑑) with Eric Coleman and Wentao Zhang! ♾️
Thursday, March 6th 🕘 9 AM PST | 🕛 12 PM EST | 🕕 6 PM CET
Register here 👉 stanford.zoom.us/meeting/regi...

Great talk today by @adriandalca.bsky.social at the @stanfordmedicine.bsky.social IBIIS & AIMI seminar, covering efficient interactive segmentation from scribbles, zero-shot segmentation to new tasks and an agent-driven vision-language framework that solves complex medical imaging tasks 🤖

Join our first ContinualAI seminar of 2025 on 𝑇𝑎𝑠𝑘-𝐴𝑔𝑛𝑜𝑠𝑡𝑖𝑐 𝐶𝑜𝑛𝑡𝑖𝑛𝑢𝑎𝑙 𝐿𝑒𝑎𝑟𝑛𝑖𝑛𝑔 𝑓𝑜𝑟 𝑡ℎ𝑒 𝑂𝑝𝑒𝑛 𝑊𝑜𝑟𝑙𝑑 with Megan Baker and Gusseppe Bravo
𝐓𝐡𝐢𝐬 𝐓𝐡𝐮𝐫𝐬𝐝𝐚𝐲, 𝐅𝐞𝐛 𝟔𝐭𝐡 🕕 6 PM CET | 🕘 9 AM PST | 🕛 12 PM EST
Register here 👉 stanford.zoom.us/meeting/regi...

👁️ Last but not least, he reiterated the importance of the PCCP for updating AI-enabled medical devices, which we break down in our latest pre-print: arxiv.org/pdf/2412.20498

👉 Medical text can be easily de-identified by open-source tools for simplifying data sharing (e.g. the Stanford AIMI De-identifier by Pierre Chambon et al.)

👉 Vision-language foundation models (e.g. CheXagent 🩻 by Zhihong Chen et al.) can bring significant time savings in radiology report drafting and interpretation

👉 Synthetically generated images can augment training sets to increase model accuracy

👉 Patient-friendly explanations of radiology reports improve patient understanding and are rarely harmful

Great inaugural seminar for the Stanford AIMI Grand Rounds 🥳 @curtlanglotz.bsky.social addressed many current topics of AI for medical imaging, including: