Shamik Roy

Attending #NAACL2025 this week to present our paper "Constrained Decoding with Speculative Lookaheads". Please feel free to say hi if you see me. :) Paper link: aclanthology.org/2025.naacl-lon…

Come to my keynote tomorrow! 9:10 am at the Trustworthy NLP workshop @trustnlp. I'll be talking about building reliable NLP tools to empower researchers in text-as-data fields #naacl2024 #naacl24 @naaclmeeting

Paper link: arxiv.org/pdf/2403.05766

#NAACL2024 paper alert 🚨: FLAP: Flow-Adhering Planning with Constrained Decoding in LLMs Feel free to stop by our oral presentation session today if you are visiting @naaclmeeting in person. 😃 When: Today (June 17) at 4 PM Mexico City Time Where: DON ALBERTO 2 Track: Dialogue

@DeanCarignan @sailiks @DX89B Thanks @DeanCarignan for appreciating our work! :) We are also keeping a close eye on Guidance - it seems incredibly useful.

A creative use of constrained decoding to improve LLMs' capabilities for multi-step planning. Great to see novel applications like this for decoding techniques. Congrats to @ShamiikRoy , @sailiks , @DX89B , Saab Mansour, Arshit Gupta! #ConstrainedDecoding

Preprint: arxiv.org/pdf/2403.05766…

5/🧵 Our proposed approach alleviates the need for fine-tuning LLMs using domain-specific data and outperforms other decoding and prompting-based baselines in faithful planning.

Delighted to share our latest paper from @AmazonScience, "FLAP: Flow Adhering Planning with Constrained Decoding in LLMs", which just got accepted to #NAACL2024 main! Work done in collaboration with my exceptional teammates @sailiks , @DX89B , Saab, and Arshit. 1/🧵

Paper link: arxiv.org/pdf/2310.07155…

📢 I'll present our poster on socio-semantic representation of text and its application on characterizing online movements at #EMNLP2023 (virtually). Session: PS7 (GT) Dec 10, 11 AM SGT/ Dec 9, 7 PM PT. Come by if around. Looking forward to an engaging session on #NLProc.🙂

@kzintas Thanks, Zinat!

@ShamiikRoy Congrats dada!

Pleased to share that the last part of my PhD thesis has been accepted to #EMNLP2023 Findings! TL;DR: We proposed graph-based structured prediction on rich socio-semantic representation of text for identifying perspectives on social movements. Preprint: arxiv.org/pdf/2310.07155…

@NavedTheNerd Thanks, Naved!

@ShamiikRoy Congratulations man!

@ShamiikRoy That appendix is like a other paper 😂

Happy to share that paper based on my last year's internship at my current team got accepted to @aaclmeeting. Credit goes to my awesome collaborators at AWS AI Labs. 🙂 Check out the preprint to learn about conversation style transfer and its applications: arxiv.org/pdf/2302.08362…

Check out the preprint of our upcoming EMNLP paper arxiv.org/pdf/2109.04535…. In this paper, we introduce morality frames, a knowledge representation framework for organizing moral attitudes directed to entities in political tweets & a deep relational learning model to detect them.

@Tunaz_Islam @ml_pacheco_ Thanks!

@ShamiikRoy @ml_pacheco_ Congratulations :)

Long paper got accepted to EMNLP 2021. Thanks to my awesome collaborators @ml_pacheco_ and Dr. Dan Goldwasser. 🙂 This paper is an extension of our SocialNLP 2021 paper (aclanthology.org/2021.socialnlp…). Arxiv version coming soon.