←
Tanmoy Chakraborty1,542 posts
Tanmoy Chakraborty avatar

Tanmoy Chakraborty

@Tanmoy_Chak

Chair Prof in AI, Full Professor @iitdelhi; ACM Distinguished Speaker; Lab @lcs2lab; Previously @IIITDelhi @UofMaryland @iitkgp; #NLP #LLMs

◷ Joined October 2014▧ 194 Photos and videos
Tanmoy Chakraborty avatar
Tanmoy Chakraborty@Tanmoy_Chak·19 Dec 2024

🌟 𝐀 𝐍𝐞𝐰 T𝐞𝐱𝐭𝐛𝐨𝐨𝐤 -- 𝐈𝐧𝐭𝐫𝐨𝐝𝐮𝐜𝐭𝐢𝐨𝐧 𝐭𝐨 𝐋𝐚𝐫𝐠𝐞 𝐋𝐚𝐧𝐠𝐮𝐚𝐠𝐞 𝐌𝐨𝐝𝐞𝐥𝐬 🌟 I am excited to share the release of my new textbook, 𝘐𝘯𝘵𝘳𝘰𝘥𝘶𝘤𝘵𝘪𝘰𝘯 𝘵𝘰 𝘓𝘢𝘳𝘨𝘦 𝘓𝘢𝘯𝘨𝘶𝘢𝘨𝘦 𝘔𝘰𝘥𝘦𝘭𝘴 (#LLMs) -- Perhaps the first textbook on LLMs. Target Audience: 👉 Students/beginners, Looking for a structured starting point to learn LLMs 👉 Teachers, planning to offer a course on LLMs 👉 Industry professional, seeking to deepen their understanding of LLMs Explore the Book: 🔗 Book Website: tanmoychak.com/llmbook/ 📑 Table of Contents: tanmoychak.com/llmbook/toc.p… 🛒 Available on Amazon: amazon.in/dp/936386474X/ Enhance Your Learning Experience: 👉 Slides & Lecture Videos: Chapter-wise resources -- lcs2-iitd.github.io/ELL881-A… 👉 Exercises & Solutions: Practice with detailed chapter exercises (solutions available on request). 👉 Upcoming @nptel_official Course: Starting January 2025! Preview here: onlinecourses.nptel.ac.in/no… Book Endorsement: 📖 Foreword by Prof. Tim Baldwin @eltimster 👏 Endorsements from Prof. Iryna Gurevych @IGurevych and Prof. Pushpak Bhattacharyya #LLMs #Textbook @iitdelhi @WileyIndiaPL @lcs2lab

Post attachment
Tanmoy Chakraborty avatar
Tanmoy Chakraborty@Tanmoy_Chak·Nov 16

AlphaEdit on Llama-3.1-8B: 97% edit efficacy in English → 54% in Urdu. 1/ We split the cross-lingual gap of ROME, MEMIT and AlphaEdit into three exact terms. The dominant one is orthogonal residual mismatch: source-language edits never reach the directions the target language needs. 2/ Fixes that use only source-language residual statistics hit a provable lower bound. Contrastive realignment improves key-space propagation, but value-space mismatch remains. 3/ Tested on Llama-3.1-8B and Gemma-3-4B across EN, DE, HI, UR and ZH. Led by Aastha Verma, @anwoy_ and Sparsh Jain

Tanmoy Chakraborty avatar
Tanmoy Chakraborty@Tanmoy_Chak·Nov 16

You edit a fact in English. Does the "multilingual" model also learn it in Hindi? Mostly, no! Then are these models truly "𝐦𝐮𝐥𝐭𝐢𝐥𝐢𝐧𝐠𝐮𝐚𝐥"? We are the first to explain this bottleneck from a 𝒕𝒉𝒆𝒐𝒓𝒆𝒕𝒊𝒄𝒂𝒍 𝒗𝒊𝒆𝒘𝒑𝒐𝒊𝒏𝒕 of model interpretability. Our new #NeurIPS26 paper argues this isn’t a flaw in the editing algorithm. It’s a geometric bottleneck in the model itself. @NeurIPSConf 📄 arxiv.org/abs/2611.10646 💻 github.com/C-anwoy/cross-lin…

Tanmoy Chakraborty avatar
Tanmoy Chakraborty@Tanmoy_Chak·Nov 11

Delighted to announce TRUSTMAPS -- an @ICMRDELHI -funded National Centre led by @iitdelhi with AIIMS Delhi, IIT Kharagpur, Ashoka University & Tata Medical Centre. As a PI, I will work with 18 Co-PIs across 5 institutes. Our mission: Build India's first national certification ecosystem for Medical AI. More details here: lnkd.in/p/g9gi7yNY 📢 Follow the official TRUSTMAPS LinkedIn page for research updates, collaborations, and upcoming initiatives: linkedin.com/company/trustma… @AshokaUniv @aiims_newdelhi @IITKgp

Post attachment
LCS2 Lab avatar
⟳ Tanmoy Chakraborty retweeted
LCS2 Lab@lcs2lab·Nov 1

Congratulations to Tamal Majumder, a PhD student at our lab, on getting the #BestPosterAward at the #IITDelhi #OpenHouse2026! 🏆 We are delighted to see his work being recognized and wish him continued success. Congratulations, Tamal! 👏

Post attachment
LCS2 Lab avatar
⟳ Tanmoy Chakraborty retweeted
LCS2 Lab@lcs2lab·Oct 27

📰 Media Feature Alert 📢 Glad to see our perspectives on efficient AI and Small Language Models featured in Lokmat Times! The future of AI is not just about building larger models, but about making them smaller, more efficient, and accessible. @Tanmoy_Chak #EfficientAI #SLMs

Post attachment
Tanmoy Chakraborty avatar
Tanmoy Chakraborty@Tanmoy_Chak·Oct 22

Heading to Budapest for #EMNLP2026! Happy to catch up and chat about: a) Exciting research happening at our lab @lcs2lab b) ARR reviewing challenges and how we can improve the process (as an EiC of @ReviewAcl) c) Faculty opportunities at @iitdelhi @emnlpmeeting

I am delighted to announce the acceptance of 7 papers to EMNLP 2026 (Main: 4 and Findings: 3). Congratulations to the entire lab for the hard work!! The preprints of most of the papers are already available. The remaining will be uploaded soon. @lcs2lab @YardiScAI @iitdelhi @emnlpmeeting #EMNLP2026

Post attachment
ayan sengupta avatar
⟳ Tanmoy Chakraborty retweeted
ayan sengupta@ayans007·Oct 11

🚨 New article: From Scaling Laws to Downscaling Laws x.com/ayans007/status/210927… In 2025 we argued at ICML that the field needs downscaling laws. In 2026 we deliver one at EMNLP. A 10% lower loss costs 4–8× the carbon emission. Pruning 43% of a model keeps 70% of its performance, with zero retraining. Scaling up is expensive. Scaling down is predictable. @Tanmoy_Chak @lcs2lab #downscaling #efficientllms #modelcompression

ayan sengupta @ayans007 Oct 11 Article From Scaling Laws to Downscaling Laws: What Pruning Ten LLMs Taught Us Scaling laws tell you what a bigger model buys. Nobody had a law for what a smaller one costs. In our ICML 2025 position paper we argued the field needs downscaling laws. Our EMNLP 2026 paper delivers

Tanmoy Chakraborty avatar
Tanmoy Chakraborty@Tanmoy_Chak·Oct 6

Honoured to host the legendary Prof. @liuhuan and Prof. Longbing Cao at our lab @lcs2lab @iitdelhi today! Our entire lab had a wonderful opportunity to interact with them and receive their invaluable feedback. And Prof. Huan’s sense of humour is simply out of this world! ☺️

Post attachmentPost attachment
Tanmoy Chakraborty avatar
Tanmoy Chakraborty@Tanmoy_Chak·Oct 2

In knowledge distillation (KD), teachers hallucinate, and standard KD forces the student to copy them anyway, spending probability mass on the teacher's noisy, low-confidence predictions and overwriting priors the student already had right. Our fix, CaRE-KD (accepted at #NeurIPS2026), makes the student decide when to trust the teacher. At the token level, a confidence gate switches between mean-seeking and mode-seeking divergence depending on how sure the teacher is. At the sequence level, a rejection mechanism drops updates where the teacher is epistemically less reliable than the student. Preprint: arxiv.org/pdf/2609.36734 @NeurIPSConf @lcs2lab @ayans007 #SLMs

Post attachment