Shashikant

#

[IIT Bombay @iitbombay]

A proud moment for IIT Bombay!
Former Deputy Director and distinguished academician Prof. Juzer Vasi has been conferred the Padma Shri, in recognition of his extraordinary lifetime contribution to science and engineering. A widely respected name in the field of semiconductors and microelectronics, Prof. Vasi, Department of Electrical Engineering, continues to inspire many as Professor Emeritus.

@PadmaAwards
@IITBAA1

[President of India @rashtrapatibhvn]

President Droupadi Murmu presents Vigyan Yuva- Shanti Swarup Bhatnagar Awards to:
•Shri Ankur Garg in Space Science and Technology.
•Prof. Mohanasankar Sivaprakasam in Technology & Innovation.

[IIT Gandhinagar @iitgn]

IITGN is proud to share that 𝗔𝗻𝗸𝘂𝗿 𝗚𝗮𝗿𝗴 (PhD, Electrical Engineering) has been honoured with the 𝗩𝗶𝗴𝘆𝗮𝗻 𝗬𝘂𝘃𝗮 – 𝗥𝗮𝘀𝗵𝘁𝗿𝗶𝘆𝗮 𝗩𝗶𝗴𝘆𝗮𝗻 𝗣𝘂𝗿𝗮𝘀𝗸𝗮𝗿 𝟮𝟬𝟮𝟱 (𝗦𝗵𝗮𝗻𝘁𝗶 𝗦𝘄𝗮𝗿𝘂𝗽 𝗕𝗵𝗮𝘁𝗻𝗮𝗴𝗮𝗿 𝗔𝘄𝗮𝗿𝗱).

[OpenAI @OpenAI]

By popular request: you can now branch conversations in ChatGPT, letting you more easily explore different directions without losing your original thread.

Available now to logged-in users on web.

[Michael Black @Michael_J_Black]

This applies to science and scientific writing. No amount of experiments will make up for a bad story. Don't go do all the work and then try to tell a mediocre story with the results. Michal Irani once told me that she writes the introductions to her papers long before she has results. This is storyboarding. In my group, we often draw our experimental figures out by hand on a board, photograph them, and insert them as placeholders in the paper. It helps you visualize the full story. There is a huge amount of work in a scientific publication. Laying out the story in advance means that you don't waste time on something that is never going to work or have impact. Of course, sometimes the results come back and don't support your story. In that case, like Pixar, you throw out the story and start again.

[François Fleuret @francoisfleuret]

The AI field is now split into (A) a "traditional" ml/dl domain, and (B) a "psycho-AI" domain where innovation requires an understanding of / intuition about the cognitive capabilities of pretrained models and how to prompt / fine-tune them.

These two fields are IMO separated.

[Shanmuganathan Raman @shanmugaphoto]

Two papers from our team @iitgn on Pano Hair Modeling by Shashikant and Incremental Multi Scene Modeling using NGP by @prajwalsingh_25, Ashish and Gautam are accepted at @BMVCconf

More details soon!

#bmvc2025 #iitgn #bmvc #computervision #ai #graphics

[Shanmuganathan Raman @shanmugaphoto]

@docmilanfar Unfortunate comment… One can say rather AI saved the researchers by allowing them to pump and dump thousands of do-and-see-what-happens papers! Best and elegant works I have come across in Computer Vision do not use any AI/ML techniques and have strong fundamental understanding.

[Shanmuganathan Raman @shanmugaphoto]

Met my present and past PhD students @iitgn during my visit to @iitbombay. Aalok is a Visiting Fellow at @TIFRScience and Prajwal is working with me now. Aalok has done some great works like Knot Art and Hand Shadow Art during his PhD.

#iitgn

[Shanmuganathan Raman @shanmugaphoto]

Six papers will be presented from our team @iitgn at the 2025 IEEE International Conference on Image Processing (ICIP)! Congrats to all the coauthors!!

https://t.co/v4AExue5hR

More details soon!

#icip2025 #iitgn #ai #computervision

[François Fleuret @francoisfleuret]

Years ago I made a list of how to cheat in ML for a course, and it remains surprisingly up-to-date.

In increasing order of dishonesty.

1/6

[Star Wars Holocron @sw_holocron]

Episodes 8 and 9 of ANDOR Season 2 are two of the greatest episodes of television of all time.

[International Conference on 3D Vision (3DV) @3DVconf]

Best Paper Award #3DV2025 goes to "An object is Worth 64x64 Pixels"!

[Yuliang Xiu @yuliangxiu]

There are three concurrent works connecting language with amazing garment code (https://t.co/u7KIlRYlEQ) that have all been accepted by #CVPR2025:

- AIpparel: A Large Multimodal Foundational Model for Digital Garments (https://t.co/y5pTDaldiM)
- Design2GarmentCode: Turning Design Concepts into Tangible Garments Through Program Synthesis (https://t.co/ntHo8srZlv)
- ChatGarment: Garment Estimation, Generation, and Editing via Large Language Models (https://t.co/GPZwcWQfHr)

Trends are something you can only wait for, not chase.

It's like waiting for a train at the station (GarmentCode); if you stay put, a train will eventually come (LLM, VLM, etc.). As for the one that just left, we ordinary folks can't catch up (being scooped) or end up burning out (being asked to compare numerous new baselines by R2).

[Keenan Crane @keenanisalive]

I don't know who needs to hear this but:

"SDF" has, for some, become a catch-all term that actually means "a function that tells you if you're inside/outside a shape, and a number that roughly indicates how far."

A true signed *distance* function really gives you *distance.*

[Andrej Karpathy @karpathy]

New 3h31m video on YouTube:
"Deep Dive into LLMs like ChatGPT"

This is a general audience deep dive into the Large Language Model (LLM) AI technology that powers ChatGPT and related products. It is covers the full training stack of how the models are developed, along with mental models of how to think about their "psychology", and how to get the best use them in practical applications.

We cover all the major stages:
1. pretraining: data, tokenization, Transformer neural network I/O and internals, inference, GPT-2 training example, Llama 3.1 base inference examples
2. supervised finetuning: conversations data, "LLM Psychology": hallucinations, tool use, knowledge/working memory, knowledge of self, models need tokens to think, spelling, jagged intelligence
3. reinforcement learning: practice makes perfect, DeepSeek-R1, AlphaGo, RLHF.

I designed this video for the "general audience" track of my videos, which I believe are accessible to most people, even without technical background. It should give you an intuitive understanding of the full training pipeline of LLMs like ChatGPT, with many examples along the way, and maybe some ways of thinking around current capabilities, where we are, and what's coming.

(Also, I have one "Intro to LLMs" video already from ~year ago, but that is just a re-recording of a random talk, so I wanted to loop around and do a lot more comprehensive version of this topic. They can still be combined, as the talk goes a lot deeper into other topics, e.g. LLM OS and LLM Security)

Hope it's fun & useful!
https://t.co/75mXcUBI8L

[François Fleuret @francoisfleuret]

It is hard to overstate how cool and powerful is flex attention. @cHHillee

https://t.co/IWVpxY9qLn

TL;DR: it is an implementation of the attention operator in @pytorch that allows in particular to efficiently "carve" the attention matrix.

1/3

[Shanmuganathan Raman @shanmugaphoto]

Participated in a Live Panel Discussion @ddgirnarlive today.

https://t.co/P0sGzRhFZv

[Sida Peng @pengsida]

Excited to share our SIGGRAPH Asia 2024 paper GVHMR 🥳.
We can accurately estimates human motions in the world coordinate system, which are readily transferred to the simulator.

Code: https://t.co/DVhs83Yem9
HuggingFace demo: https://t.co/WKvsSqy02w

[Poli-gasm @Poligasm]

@CVPR, do you stand with #Israel or #Gaza? What message are you trying to send by including such a statement, and what precedent are you setting for future conferences?

#RussiaUkraineWar #CVPR #CVPR2025