@deeptigp

Faculty @ Boston University; I work on all things pixels

 Under a rock
deeptigp.github.io/
Joined August 2008
  • Tweets5.3K
  • Following1.2K
  • Followers1.4K
208Media
[1/4] The human eye doesn't process every single pixel of a video continuously—it focuses on what changes. So why are our video AI models wasting compute on redundant frames? Introducing Swift Sampling: a test-time technique inspired by the human visual system. 🧠👇
4
21
3
193
21,057
Thanks for hosting me today @YuanqiD !
Yuanqi Du
@YuanqiDOct 19
Next Tuesday, our MSR NE Generative Modeling and Sampling Seminar will feature Deepti Ghadiyaram speaking (in person) about "Improving Efficiency and Robustness of Generative Models"! Information to join below👇
1
12
413
Happening now! Come say hi to @xavierohan
Deepti Ghadiyaram@deeptigpOct 5
Do VLMs truly reason or memorize data? We show ❗even transformations like rotation & scaling still throw them off ❗finetuning fails to generalize across domains tinyurl.com/colmbu26 📅 Stop by our poster on Wed @ 4:30 PM and chat with @xavierohan #COLM2026 @COLM_conf
17
455
Deepti Ghadiyaram retweeted
Arash Vahdat
@ArashVahdatOct 5
Please help spread the word! We’re hiring Research Scientists and Research Interns at NVIDIA Research to work on some of the most exciting problems in generative AI: 🌎 World Models & Video Diffusion 🧠 Diffusion Language Models 🧬 Biomolecule Design If your research aligns with any of these areas, please apply directly through the links below. Open positions + application links: research.nvidia.com/labs/gen…
4
58
9
487
67,275
[New post] A Tuesday medium.com/@deeptigp/a-tuesd…
1
10
329
Do VLMs truly reason or memorize data? We show ❗even transformations like rotation & scaling still throw them off ❗finetuning fails to generalize across domains tinyurl.com/colmbu26 📅 Stop by our poster on Wed @ 4:30 PM and chat with @xavierohan #COLM2026 @COLM_conf
2
1
23
857
Thank you @GoodfireAI for supporting my research group's research on interpretability of multi-modal LLMs. Excited to use Silico! 🚀
Goodfire
@GoodfireAIAug 20
We’re giving out $1M in grants of free Silico usage for academic and nonprofit researchers focused on AI interpretability and alignment. We feel extreme urgency about advancing interpretability for alignment, and we want to help more researchers push it forward. 🧵
2
11
433
Please join us in congratulating Deepti Ghadiyaram on her new role as the co-director of our Artificial Intelligence Research Initiative (AIR)! Deepti has been a core faculty member of AIR since summer of 2024 and will co-direct with Kate Saenko. Read more: spr.ly/6010BGJeSO
1
4
206
The scale and the effect of the destruction that was caused in just a few minutes during the floods is heart wrenching.
Sadhguru
@SadhguruJVAug 27
🙏🏽
1
2
330
One hundred days by Alice Pung was a hard but a great book to get through. I did the equivalent of Joey (from Friends) putting the book he is scared to read further in the freezer. I admired how Karuna learns to stand her ground and stops the generational trauma with her.
4
238
Thank you to @deeptigp for the amazing lecture and for the rich interactions with all the students at ACVSS 26 Ghana.
Deepti Ghadiyaram@deeptigpAug 2
Had a heartfelt experience teaching and interacting with a group of 30 outstanding African students as part of @ACVSS_AI acvss.ai/. I was beyond impressed with the depth of their questions, their interest in pursuing research, and their kindness and generosity!
3
14
1,142
Had a heartfelt experience teaching and interacting with a group of 30 outstanding African students as part of @ACVSS_AI acvss.ai/. I was beyond impressed with the depth of their questions, their interest in pursuing research, and their kindness and generosity!
2
1
1
16
1,643
VLMs lack geometric invariance: xthomasbu.github.io/visual_i… - Rotate / scale an object: perf. breaks down. - Models perform best on real photos & worse on sketches (data memorization) ❌Scale, prompting does not fix. By @xavierohan , Jason, Zach #COLM2026
8
579
Deepti Ghadiyaram retweeted
Ismini Lourentzou
@Ismini_LJun 25
It's been an incredible year, with wonderful collaborators and a CAREER award I'm grateful for. I got a lot of feedback on mine, so I want to do the same for others. Opening office hours for faculty writing/thinking about #NSFCAREER proposals. Bring a draft, questions, or just chat. calendly.com/lourent2-illino… Happy to share tips and give honest feedback.

NSF CAREER chat - Lourentzou, Ismini

calendly.com
3
2
2
21
4,280
Deepti Ghadiyaram retweeted
WiCV@WiCVworkshopJun 22
📣 Call for Reviewers 📣 WiCV at #ECCV2026 is looking for reviewers to help evaluate workshop submissions. Interested in supporting the community and contributing to WiCV? Submit your nomination below: docs.google.com/forms/d/e/1F… Please share/retweet 💙!

Reviewer Self-Nomination WiCV@ECCV 2026

Call for potential reviewers at the WiCV workshop at ECCV 2026 in Malmö, Sweden. Junior reviewers are welcome and will be assigned a lower review load.

docs.google.com
2
2
729
We show how easy it is to break omni-MLLMs with very subtle audio and multimodal attacks. 1. Audio only attacks effect visual only questions 2. Audio attacks can fool content moderation systems. Project page: cskyl.github.io/MLLM-Typogra… Kudos to @TianleChen14 👏
Tianle Chen@TianleChen14Jun 17
Excited to share our new paper: "A Systematic Study of Cross-Modal Typographic Attacks on Audio-Visual Reasoning." Multimodal LLMs increasingly listen as well as see - so we asked: can you attack them through speech? It turns out you can, easily.
4
485
A great event to learn about the latest Mech Interpretability research. Save the date and attend!
Gabriel Franco@gvsfrancoJun 10
🧠🤖 The 2026 New England Mechanistic Interpretability (NEMI) Workshop will be Aug. 14 at Boston University! Help spread the word and join the New England mech interp community! Registration and submission info in thread:👇
5
997
I just published a story: "Something blue" medium.com/p/something-blue-…
5
275
Our group has a strong presence with several talks and posters at workshops and the main conference at #CVPR2026. Please stop by and talk to @dahye_kim @TianleChen14 @xavierohan Youngsun Lim
1
2
19
819
[3/4] Swift Sampling 🏎️ retains "temporal surprises." ⚡ Test-time technique (no retraining needed) 📉 Adds < 0.02x overhead to your pipeline 🏆 Outperforms complex, computationally intensive query-based algorithms across 10+ vision tasks.
1
5
1,040
[4/4] Swift Sampling is a game-changer for long video reasoning with strict frame budgets -- we saw accuracy jumps of up to +12.5 points! 📈 Project page: kim-dahye.github.io/swift-sa… Incredible work by @dahye_kim , @SachdevaBhuvan @karan_uppal3 👏
2
13
1,082
Load more