Shanu Kumar
PhD @ Mohamed bin Zayed University of Artificial Intelligence
Shanu Kumar

About Me

I am a first year PhD student at MBZUAI, advised by Prof. Yova Kementchedjhieva and Prof. Kentaro Inui. I work on vision–language models, with a focus on personalization, alignment, spatial reasoning, and interpretability.
Previously, I was a Data & Applied Scientist at Microsoft (2019–2025). I worked closely with Prof. Monojit Choudhury and Prof. Sandipan Dandapat on multilingual NLP, building and evaluating models that don’t panic when the input switches languages mid-sentence. I also collaborated with Manish Gupta on scaling prompt optimization for industry scale prompts. I completed my B.Tech. in Electrical Engineering at IIT Kanpur, where I was awarded the Silver Medal. I worked there under Prof. Vinay P. Namboodiri, which led to my first full conference paper at CVPR. Outside research, I’m usually trekking or doing photography, occasionally claiming I am collecting data while mostly just admiring the view. More in my CV.

Current Research Focus

Hide

Improving Visual Alignment in Vision–Language Models

I am working on making VLMs align visual and textual representations more reliably and earlier in the network, so vision can actually influence reasoning instead of arriving late to the party.

Temporal Personalization (usable for Jarvis)

I am exploring temporal personalization: modeling preferences and context that change over time, so assistants can be helpful in a way that feels natural.

Recent Updates

Hide
Aug. 2025
Joined Mohamed bin Zayed University of Artificial Intelligence as a PhD student.
Jul. 2025
Left Microsoft India after 6 years, grateful for the people, problems, and plenty of “ship-it” moments.
Apr. 2025
Our work on SCULPT: Systematic Tuning of Long Prompts got accepted to ACL 2025.
Jul. 2019
Awarded the Silver Medal at IIT Kanpur for exceptional undergraduate research.
Mar. 2019
Received research grants from Microsoft Research and the Indian National Academy of Engineering.
Jan. 2019
Received a Research Grant from Microsoft Research for Adversarial Adaptation of Scene Graph Models for Understanding Civic Issues.

Publications

Hide
2025
LITMUS++
LITMUS++: An Agentic System for Predictive Analysis of Low-Resource Languages Across Tasks and Models
Avni Mital Kumar, Shanu Kumar, Sandipan Dandapat, Monojit Choudhury
AACL Demo
SCULPT
SCULPT: Systematic Tuning of Long Prompts
Shanu Kumar, Akhila Yesantarao Venkata, Shubhanshu Khandelwal, Bishal Santra, Parag Agrawal, Manish Gupta
ACL Main
Cultural Kaleidoscope
Navigating the Cultural Kaleidoscope: A Hitchhiker’s Guide to Sensitivity in LLMs
Somnath Banerjee, Sayan Layek, Hari Shrawgi, Rajarshi Mandal, Avik Halder, Shanu Kumar, et al.
NAACL Main
SafeInfer
SafeInfer: Context Adaptive Decoding Time Safety Alignment for LLMs
Somnath Banerjee, Soham Tripathy, Sayan Layek, Shanu Kumar, Animesh Mukherjee, Rima Hazra
AAAI AI Alignment Track
Uncertainty-Guided CoT
Enhancing Zero-shot CoT Prompting via Uncertainty-Guided Strategy Selection
Shanu Kumar, Saish Mendke, Karody Lubna Abdul Rahman, Santosh Kurasa, Parag Agrawal, Sandipan Dandapat
COLING Oral
Socio-Culturally Aware Moderation
Socio-Culturally Aware Evaluation Framework for LLM-Based Content Moderation
Shanu Kumar, Gauri Kholkar, Saish Mendke, Anubhav Sadana, Parag Agrawal, Sandipan Dandapat
COLING SUMEval Workshop
Attributional Safety Failures
Attributional Safety Failures in LLMs under Code-Mixed Perturbations
Somnath Banerjee, Pratyush Chatterjee, Shanu Kumar, Sayan Layek, Parag Agrawal, Rima Hazra, Animesh Mukherjee
arXiv Under review
Towards Safer Pretraining
Towards Safer Pretraining: Analyzing and Filtering Harmful Content in Webscale datasets
Sai Krishna Mendu, Harish Yenala, Aditi Gulati, Shanu Kumar, Parag Agrawal
IJCAI Main
2023
DiTTO
DiTTO: A Feature Representation Imitation Approach for improving Cross-Lingual Transfer
Shanu Kumar, Soujanya Abbaraju, Sunayana Sitaram, Sandipan Dandapat, Monojit Choudhury
EACL Main
READ
READ: Reinforcement-based Adversarial Learning for Text Classification...
Rohit Sharma*, Shanu Kumar*, Avinash Kumar
arXiv Preprint
2022
Diversity and Uncertainty in Moderation
“Diversity and Uncertainty in Moderation” are the Key to Data Selection...
Shanu Kumar, Sandipan Dandapat, Monojit Choudhury
NAACL Findings
MTL Zero-shot Performance Prediction
Multi Task Learning For Zero Shot Performance Prediction of Multilingual Models
Kabir Ahuja*, Shanu Kumar*, Sandipan Dandapat, Monojit Choudhury
ACL Oral
2019
Mitigating Uncertainty of Classifier
Mitigating Uncertainty of Classifier for Unsupervised Domain Adaptation
Shanu Kumar, Vinod Kumar Kurmi, Praphul Singh, Vinay P. Namboodiri
arXiv Preprint
Attending to Discriminative Certainty
Attending to Discriminative Certainty for Domain Adaptation
Vinod Kumar Kurmi*, Shanu Kumar*, Vinay P. Namboodiri
CVPR Poster
Adversarial Adaptation of Scene Graph Models
Adversarial Adaptation of Scene Graph Models for Understanding Civic Issues
Shanu Kumar, Shubham Atreja, Anjali Singh, Mohit Jain
WWW Short