Subrata Mitra (@from_subrata) profile picture

Researcher - computer systems, AI/ML efficiency | Senior Research Scientist, Adobe Research All my posts are my personal opinion.

Virtual
Joined May 2010
Photo by Subrata Mitra (@from_subrata)Photo by Subrata Mitra (@from_subrata)Photo by Subrata Mitra (@from_subrata)Photo by Subrata Mitra (@from_subrata)
Subrata Mitra retweeted
This guy literally explained why some people become successful while other stay average. The reason is uncomfortable. Game Theory Watch this:
31
250
949
66,395
Workshop on Synthetic Data and AI (SynthAI 2026), co-located with SIGMOD 2026 on June 5, 2026. 📅 Important Dates (All deadlines are AoE): Submission Deadline: March 20, 2026 (AoE) Notification of Acceptance: April 10, 2026 (AoE) Camera-Ready Submission: April 20, 2026 (AoE)
We are excited to organize a Workshop on Synthetic Data and AI (SynthAI 2026), co-located with SIGMOD 2026 on June 5, 2026. 🔗 Website: synthetic-data-sigmod.github… 📩 Submission Portal (CMT): cmt3.research.microsoft.com/… #SIGMOD #AI #SyntheticData
124
We are excited to organize a Workshop on Synthetic Data and AI (SynthAI 2026), co-located with SIGMOD 2026 on June 5, 2026. 🔗 Website: synthetic-data-sigmod.github… 📩 Submission Portal (CMT): cmt3.research.microsoft.com/… #SIGMOD #AI #SyntheticData
1
1
193
Subrata Mitra retweeted
Dr. Subrata Mitra, Senior Research Scientist, Adobe Research, will be speaking on “Redefining Caching for Generative AI”, on 11 February 2025, at 5:00 PM as part of the CNI Network Seminar Series. cni.iisc.ac.in/seminars/
1
56
Subrata Mitra retweeted
This is interesting as a first large diffusion-based LLM. Most of the LLMs you've been seeing are ~clones as far as the core modeling approach goes. They're all trained "autoregressively", i.e. predicting tokens from left to right. Diffusion is different - it doesn't go left to right, but all at once. You start with noise and gradually denoise into a token stream. Most of the image / video generation AI tools actually work this way and use Diffusion, not Autoregression. It's only text (and sometimes audio!) that have resisted. So it's been a bit of a mystery to me and many others why, for some reason, text prefers Autoregression, but images/videos prefer Diffusion. This turns out to be a fairly deep rabbit hole that has to do with the distribution of information and noise and our own perception of them, in these domains. If you look close enough, a lot of interesting connections emerge between the two as well. All that to say that this model has the potential to be different, and possibly showcase new, unique psychology, or new strengths and weaknesses. I encourage people to try it out!
We are excited to introduce Mercury, the first commercial-grade diffusion large language model (dLLM)! dLLMs push the frontier of intelligence and speed with parallel, coarse-to-fine text generation.
371
1,490
185
11,320
947,024
Check out Adobe Research's new work in Computer Vision at #CVPR2024! research.adobe.com/publicati…
Adobe Research is proud to be one of the sponsors of #CVPR2024! We look forward to the innovative work our researchers will present next week at the conference in Seattle!
7
4
2,503
Subrata Mitra retweeted
12-year-old Parinav has been missing for two days now in Bengaluru’s suburb Whitefield. Some of those searching for him joined me on my show today. Praying he is found very soon: youtu.be/V8KL8qxvhzs?si=bdLZ…
49
674
10
1,016
544,858
Subrata Mitra retweeted
I've known many folks over the years who came from broken marriages, some of my closest friends are children of broken/dysfunctional marriages. Everyone's experiences are different, but I do believe if both parents are there, they have to co-parent no matter their relationship.
Listen, I’m a child of a broken marriage and yeah, it is messed up, but my parents never fought for custody. While Mom brought us up, we met Dad often and stayed over when Mom had to go work. Fathers should meet their children and sometimes they might be the better primary parent. This horrific case in Goa proves that not all mothers are nurturing, some are extremely emotionally disturbed individuals. Judges must take the mental state of both parents into account; default custody judgements should not happen. And if they continue visitation rights have to be there, with strict punishments if ignored.
1
13
972
Subrata Mitra retweeted
Listen, I’m a child of a broken marriage and yeah, it is messed up, but my parents never fought for custody. While Mom brought us up, we met Dad often and stayed over when Mom had to go work. Fathers should meet their children and sometimes they might be the better primary parent. This horrific case in Goa proves that not all mothers are nurturing, some are extremely emotionally disturbed individuals. Judges must take the mental state of both parents into account; default custody judgements should not happen. And if they continue visitation rights have to be there, with strict punishments if ignored.
5
29
4
183
29,707
Subrata Mitra retweeted
The Research and Development Office, @AshokaUniv maintains a public database of grants & fellowships that India based researchers can apply for. Please share widely. cc: @MenonBioPhysics, @AnuragAgrawalMD, @kvijayraghavan docs.google.com/spreadsheets…
14
493
13
1,104
124,800
Our upcoming NSDI-2024 introduces "approximate-caching" + a cache eviction mechanism to provide significant reduction in latency and GPU compute cost for diffusion models.
Super excited to share that our paper "Approximate Caching for Efficiently Serving Diffusion Models" has been accepted at @usenix NSDI '24. Shoutout to Shubham Agarwal for pulling this off. Congrats to all the authors @from_subrata, Srikrishna Karanam, @Koyel_Mukherjee and @wlog.
1
142
We talked to Subrata Mitra, Senior Research Scientist at @AdobeResearch, about how an early visit to a supercomputing center helped shaped his research interests, what it's like to use machine learning to make sense of massive amounts of data, and more! adobe.ly/3OUpm0T
1
19
1,788
Paper: "Reinforced Approximate Exploratory Data Analysis" was accepted at AAAI-2023. #aaai2023 @RealAAAI We consider the impact of sampling in interactive data exploration settings due to approximation errors and propose a reinforcement learning-based solution to mitigate.
2
Subrata Mitra retweeted
Come and explore the theme session on 'Core Research' at #RIISE2022 and get to interact with highly skilled professionals working in industry and academia! Register today at - lnkd.in/diA2tXcT Event Date: Friday, 8th April'22
1
4
Dad questions frisking of 4-yr-old in his absence; @cpronammametro cites rule
6
19
4
52
Subrata Mitra retweeted
Machine Learning Formulas Explained! 👨‍🏫 This is the formula for the Binary Cross Entropy Loss. It is commonly used for binary classification problems. It may look super confusing, but I promise you that it is actually quite simple! Let's go step by step 👇 #RepostFriday
17
325
24
1,599