Rohit Gandikota
@rohitgandikota
Followers
935
Following
178
Media
70
Statuses
260
Ph.D. AI @ Northeastern University. Understanding, mapping, and editing knowledge in large generative models. Ex-Scientist Indian Space Research Organization
Boston, MA
Joined June 2013
Can you ask a Diffusion Model to break down a concept? 👀. SliderSpace 🚀 reveals maps of the visual knowledge naturally encoded within diffusion models. It works by decomposing the model's capabilities into intuitive, composable sliders. Here's how 🧵👇
10
62
399
RT @rohitgandikota: Erasing Concepts from FLUX.1 models 🔥. ESD now supports FLUX models, and we've recently added SDXL support as well. Ch….
github.com
Erasing Concepts from Diffusion Models . Contribute to rohitgandikota/erasing development by creating an account on GitHub.
0
7
0
Erasing Concepts from FLUX.1 models 🔥. ESD now supports FLUX models, and we've recently added SDXL support as well. Check out the code and share your experiments with us:. (This is a long-awaited request from our users - thank you for your patience!).
github.com
Erasing Concepts from Diffusion Models . Contribute to rohitgandikota/erasing development by creating an account on GitHub.
🚨UPDATE: Erasing Concepts from Diffusion Models (ESD). ESD now supports diffusers @huggingface. 🚀 New Feature: You can now surgically remove an attribute from a concept. We made SDv1.4 forget that cowboys wear hats🤠 → 👤. Code:
2
7
58
Yes indeed! . Here are 3 pieces of advice by @davidbau for PhD students interested in Mech Interp
@davidbau's talk on "What is AI interpretability for?" is about to start! . If you know David or his talks in the past - you know it is going to be special🔥. A fan or skeptic of mech Interp? come ask David some questions. Location: Grand C1 Hall or Zoom
0
0
19
RT @miv_cvpr2025: Mechanistic Interpretability for Vision Workshop has officially begun @CVPR ! 🚀. Join us at Grand C1 Hall for insightful….
0
2
0
RT @materzynska: Come join us at the @miv_cvpr2025 workshop today at the @CVPR in room C1! We have an amazing lineup of speakers 🔊 includi….
0
4
0
RT @nickhjiang: Vision transformers have high-norm outliers that hurt performance and distort attention. While prior work removed them by r….
0
134
0
RT @_AmilDravid: Artifacts in your attention maps? Forgot to train with registers? Use 𝙩𝙚𝙨𝙩-𝙩𝙞𝙢𝙚 𝙧𝙚𝙜𝙞𝙨𝙩𝙚𝙧𝙨! We find a sparse set of activat….
0
63
0
RT @_AmilDravid: Tired of looking at pixels and want to look at some neurons? Come join us @miv_cvpr2025 this Thursday @CVPR in room "Grand….
0
1
0
Super excited about @miv_cvpr2025! . What an incredible lineup of invited speakers! 👀 You instantly know that the workshop is going to be 🔥. Come to Grand C1 hall on June 12th.
It might be a rainy @CVPR this time, but we at MIV workshop have you covered!. Come to Grand C1 hall and listen to our great speakers talk about why mechanistic interpretability is important for vision models. Date: June 12th, 9AM.Location: Grand C1 hall. More info?👇
0
3
10
RAG for Wikipedia! . Seems like a good way to streamline evaluations when editing knowledge in LLMs.
RAGs are extremely useful, and yet there isn't an opensource RAG system for wikipedia (or I couldn't find it). So I built WikiRAG, a simple open-source github + hugging repo✨. Spin up your own RAG server for wikipedia in a single line. 🚗💨
0
0
2
Is the knowledge of a concept **really** removed when we erase a concept from a diffusion model? . @kevinlu4588 found that the answer is often - NO! . Checkout these clever yet simple techniques to search for the traces of knowledge in your erased models.
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 🧵👇
0
0
10
RT @rohitgandikota: @OrgadHadas @boknilev @materzynska @davidbau 1. Enhance a concept. You can enhance a concept inside a model's knowledge….
0
1
0
RT @sheridan_feucht: I used to think formal reasoning was central to language and intelligence, but now I’m not so sure. Wrote a short post….
0
4
0
RT @boknilev: @amuuueller presenting sparse feature circuits at #ICLR2025 .Also Aaron is starting as faculty at BU in the fall so reach out….
0
5
0
RT @OrgadHadas: An exciting and up-to-date implementation for text-to-image model editing of our TIME method!. Great to see these ideas evo….
0
3
0
RT @davidbau: @rohitgandikota @OrgadHadas @materzynska AI systems are sold as black boxes, so you might think it is impossible to understan….
0
1
0
RT @davidbau: Realize what @rohitgandikota and @OrgadHadas and @materzynska have done here. It goes beyond making an AI that can edit some….
0
2
0
Try the code yourself - it doesn't take much time and doesn't require a big GPU too!!. Paper: Project: Code: work w/ @OrgadHadas @boknilev @materzynska @davidbau.
github.com
Unified Concept Editing in Diffusion Models. Contribute to rohitgandikota/unified-concept-editing development by creating an account on GitHub.
1
1
32
How does UCE edit a model so fast with such low memory requirements ?. Under the hood, we are editing the cross attention weights of the diffusion models using a closed form solution. No gradients required!. Checkout this thread for an in-depth explainer.
How can we eliminate bias, copyright issues, and offensive content in text-to-image models? Introducing UCE, our model-editing approach that addresses all these challenges simultaneously. Explainer on the unified approach and why addressing these issues is so important 👇(1/12)
1
0
10