We tested several unlearning methods and found none of them really erase knowledge from the model - they simply hide it! 🧐
What does this mean? We must tread carefully with unlearning research within diffusion models🚨
Here is what we learned 🧵👇(led by @kevinlu4588)
Excited to share our paper “When Are Concepts Erased from Diffusion Models?” at @NeurIPSConf!
We introduce two conceptual models for erasure mechanisms in diffusion models, and a suite of probes to recover supposedly forgotten concepts.
Project website: https://t.co/pKQmjEASHK
If you are attending #ICML2025, check out our DataWorld workshop on Sat July 19. We have updated the website with more info on speakers & accepted papers! https://t.co/K3U540rqoe
Also happy to chat offline about all things ✨ data ✨
🎤 Giving a keynote talk - "Anomaly Detection Beyond Pretrained Representations" tomorrow at 10:30am at #CVPR, VAND 3.0 workshop (Visual Anomaly and Novelty Detection). Come say hi! 👋
When we "erase" a concept from a diffusion model, is that knowledge truly gone? 🤔
We investigated, and the answer is often 'no'!
Using simple probing techniques, the knowledge traces of the erased concept can be easily resurfaced 🔍
Here is what we learned 🧵👇
📢 Announcing our data-centric workshop at ICML 2025 on unifying data curation frameworks across domains!
📅 Deadline: May 24, AoE
🔗 Website: https://t.co/K3U540rqoe
We have an amazing lineup of speakers + panelists from various institutions and application areas.
Think your latent-noise diffusion watermarking method is robust? Think again!
We show that they are susceptible to adversarial attacks that only require one watermarked example and an off-the-shelf encoder. This attack can forge and remove the watermark with very high accuracy
I'm excited about the recent feature we've been working on at @pika_labs - action editing! 🏃 📹
Video editing methods often focus on spatial edits. Our method can directly edit the *action* of objects, even when the video already contains a distinctive motion (e.g, 👋).
In summary, WIND delivers robust, imperceptible watermarking for both AI-generated and natural images 🛡️🔏
For detailed results and insights, see our paper or code:
https://t.co/C73PCIWHCw
Led by Kasra Arabi in collaboration with @FeuerBenjamin@TealWitter@chegday
(5/5)
In our #ICLR2025 paper, we introduce WIND 🌬️
A method that embeds a distortion watermark directly in the diffusion noise! Our method ensures that the watermark in one image does not reveal information about the watermark in other images 🤫
📝 https://t.co/jmEw4g3scN
(1/5)
And there's more! By leveraging diffusion inpainting, WIND can also protect non-AI-generated images, opening exciting new pathways for copyright protection and authenticity verification across all digital content 🖼️🔑
(4/5)
Existing AI benchmarks are becoming increasingly saturated.
Partnering with @scale_AI, we created Humanity's Last Exam, a new benchmark that tests the capabilities of models at the frontier of human knowledge and reasoning.
The results are out today:
https://t.co/JPC6dkeBKR
Have a question that is challenging for humans and AI?
We (@cais + @scale_AI) are launching Humanity's Last Exam, a massive collaboration to create the world's toughest AI benchmark.
Submit a hard question and become a co-author.
Best questions get part of $500,000 in prizes!
Deadline: Nov 1, 2024
Details: https://t.co/nsOEdIy3L1
Submit here: https://t.co/8lYmlFFUTp