Shashwat Singh (@shashwat_s19) 's Twitter Profile
Shashwat Singh

@shashwat_s19

AI researcher. Interested in the science of pretraining and interpretability.

ID: 1582018723799367680

calendar_today17-10-2022 14:40:41

17 Tweet

49 Followers

85 Following

Shauli Ravfogel (@ravfogel) 's Twitter Profile Photo

Happy to announce our latest research “Representation Surgery: Theory and Practice of Affine Steering", accepted at ICML! A joint work with roeeaharoni Jonathan Herzig @ryandcotterell Ponnurangam Kumaraguru “PK” and Shashwat Singh, done during an internship at Google AI.(1/8) arxiv.org/pdf/2402.09631

Happy to announce our latest research “Representation Surgery: Theory and Practice of Affine Steering", accepted at ICML! A joint work with <a href="/roeeaharoni/">roeeaharoni</a> <a href="/jonherzig/">Jonathan Herzig</a> @ryandcotterell <a href="/ponguru/">Ponnurangam Kumaraguru “PK”</a> and <a href="/shashwat_s19/">Shashwat Singh</a>, done during an internship at <a href="/GoogleAI/">Google AI</a>.(1/8) arxiv.org/pdf/2402.09631
Manu Gaur (@gaur_manu) 's Twitter Profile Photo

🚨 Introducing Detect, Describe, Discriminate: Moving Beyond VQA for MLLM Evaluation. Given an image pair, it is easier for an MLLM to identify fine-grained visual differences during VQA evaluation than to independently detect and describe such differences 🧵(1/n):

🚨 Introducing Detect, Describe, Discriminate: Moving Beyond VQA for MLLM Evaluation.

Given an image pair, it is easier for an MLLM to identify fine-grained visual differences during VQA evaluation than to independently detect and describe such differences 🧵(1/n):
Shashwat Goel (@shashwatgoel7) 's Twitter Profile Photo

Did you think retraining without the deletion set is always enough (even if inefficient) for unlearning? Our paper Corrective Machine Unlearning showed a practical scenario where it is NOT, and is now accepted at TMLR! Work with Ameya P. @ ICML 2025, Amartya Sanyal, Oxford Torr Vision Group, Ponnurangam Kumaraguru “PK” 🧵👇

Did you think retraining without the deletion set is always enough (even if inefficient) for unlearning?

Our paper Corrective Machine Unlearning showed a practical scenario where it is NOT, and is now accepted at TMLR! Work with <a href="/AmyPrb/">Ameya P. @ ICML 2025</a>, <a href="/AmartyaSanyal/">Amartya Sanyal</a>, <a href="/OxfordTVG/">Oxford Torr Vision Group</a>, <a href="/ponguru/">Ponnurangam Kumaraguru “PK”</a> 🧵👇