Shubhashis Roy Dipta

X-MLR @AmazonScience, @scale_AI || PhD @umbc || Reasoning || Tool-calling LLM || 🏠 || 📝

[Shubhashis Roy Dipta @iamdipta007]

@cryptexcode Yes 🥲🥲

[Shubhashis Roy Dipta @iamdipta007]

🎉 Big news: my research received a $𝟐𝟎,𝟎𝟎𝟎 𝐆𝐞𝐦𝐢𝐧𝐢 𝐀𝐜𝐚𝐝𝐞𝐦𝐢𝐜 𝐏𝐫𝐨𝐠𝐫𝐚𝐦 𝐀𝐰𝐚𝐫𝐝 (@googlecloud credits) to scale up my agentic LLM Research.

Thank you to the @Google and @GeminiApp for believing in this direction.

#agenticLLM #rl #NLPproc #ML #LLM

[Shubhashis Roy Dipta @iamdipta007]

AgentCollabBench is accepted at the @FAGENWorkshop at #ICML2026 🎉

Multi-agent AI can hand you a perfectly correct-looking answer while quietly breaking its own reasoning. We built 900 tasks to catch it.

📄 https://t.co/ob0QSjSIeQ

[Artificial Intelligence Papers @SciFi]

TRIAGE: Evaluating Prospective Metacognitive Control in LLMs under Resource Constraints

Zabir Al Nazi, Shubhashis Roy Dipta
https://t.co/aGKpmJ5p5B [𝚌𝚜.𝙰𝙸]

[Natural Language Processing Papers @HEI]

Cornerstones or Stumbling Blocks? Deciphering the Rock Tokens in On-Policy Distillation

Yuxuan Jiang, Runchao Li, Shubhashis Roy Dipta, Dawei Li, Zhao Yang
https://t.co/zCGqdvhkFI [𝚌𝚜.𝙲𝙻 𝚌𝚜.𝙰𝙸]

[Natural Language Processing Papers @HEI]

AgentCollabBench: Diagnosing When Good Agents Make Bad Collaborators

Aritra Mazumder, Shubhashis Roy Dipta, Nusrat Jahan Lia, Tanzila Khan, Kainat Raisa Hossain, Nehaa Shri, Shubhrangshu Debsarkar, Humayra Tasnim, …
https://t.co/kd1KoIg5Oo [𝚌𝚜.𝙲𝙻 𝚌𝚜.𝙰𝙸 𝚌𝚜.𝙻𝙶]

[Machine Learning (ML) Papers @Memoirs]

Omni-Modal Dissonance Benchmark: Systematically Breaking Modality Consensus to Probe Robustness and Calibrated Abstention

Zabir Al Nazi, Shubhashis Roy Dipta, Md Rizwan Parvez
https://t.co/Cnwh9MuhAA [𝚌𝚜.𝙻𝙶]

[Shubhashis Roy Dipta @iamdipta007]

🎉 Excited to share that I’ve successfully 𝗰𝗼𝗺𝗽𝗹𝗲𝘁𝗲𝗱 𝗺𝘆 𝗣𝗵𝗗 𝗽𝗿𝗼𝗽𝗼𝘀𝗮𝗹 and 𝗮𝗱𝘃𝗮𝗻𝗰𝗲𝗱 𝘁𝗼 𝗣𝗵𝗗 𝗰𝗮𝗻𝗱𝗶𝗱𝗮𝗰𝘆.

Grateful for the guidance and support from my advisor and committee throughout this journey.

[Shubhashis Roy Dipta @iamdipta007]

[Findings]
Multimodal Unlearning Across Vision, Language, Video, and Audio: Survey of Methods, Datasets, and Benchmarks
Preprint: https://t.co/cSQf3Lfwhd

[Shubhashis Roy Dipta @iamdipta007]

3 papers accepted at ACL 2026 (1 Main, 2 Findings)! Sharing below 👇 #ACL2026

[Shubhashis Roy Dipta @iamdipta007]

[Findings]
GanitLLM: Difficulty-Aware Bengali Mathematical Reasoning through Curriculum-GRPO
Preprint: https://t.co/R5WvMENrln

[Shubhashis Roy Dipta @iamdipta007]

@michelleefang Merry Christmas and yes yes I need it 🧑‍💻🧑‍💻

[Shubhashis Roy Dipta @iamdipta007]

@gargi_balasu @NeurIPSConf Would love to meet. Sent an email.

[Saeejith Nair @sighjith]

I built https://t.co/VJ7oHdvf7s, a new interface for arXiv

As we enter an era of accelerated scientific discovery, we need better tools that augment human cognition to help us keep up.

Try it: visit papiers ai or swap arxiv -> papiers on any paper URL

[Shubhashis Roy Dipta @iamdipta007]

🎉 Poster presentation at #EMNLP2025

🌐 https://t.co/pY1YUi4VWS

Question: "Which Bollywood movie won the Oscar in 1928?"
Spoiler: none - Oscars began in 1929 😅

That’s exactly the point of our research - teaching LLM to 𝘁𝗵𝗶𝗻𝗸 𝗯𝗲𝗳𝗼𝗿𝗲 𝘁𝗵𝗲𝘆 𝗮𝘀𝘀𝘂𝗺𝗲.

[Shubhashis Roy Dipta @iamdipta007]

📘 Resources:
🌐 Project Page → https://t.co/M92gkaP7ub
📄 Pre-print → https://t.co/bYC0rX4fU0 (camera-ready soon)
💻 Code → https://t.co/z1uSVey8YC

(4/4)

[Shubhashis Roy Dipta @iamdipta007]

🎉 Thrilled to share — our paper "Q2E: Query-to-Event Decomposition for Zero-Shot Multilingual Text-to-Video Retrieval" is accepted to #AACL2025 (Main)!

Excited to present at AACL 2025 🇮🇳
See you all there! 🙌

Details 👇

#AACL2025 #LLM #VideoRetrieval #MultimodalAI #ML #NLP

[Shubhashis Roy Dipta @iamdipta007]

✨ Key Takeaways:
• Intent-aware retrieval → LLM-driven decomposition helps capture real user intent
• Multimodal grounding → inverse-entropy based fusion beats simple averaging
• Zero-shot gains → +4 NDCG (MSR-VTT), +8 NDCG (MultiVENT) 🚀 (3/4)