Neelabh Madan

I am a second year Computer Science Ph.D. student at NYU, Courant. I am advised by Lakshmi Subramanian .

Previously, I worked as a Research Fellow at Microsoft Research, India (MSRI), where I was advised by Dr. Manik Varma and Dr. Amit Sharma. Here I closely worked with the Ads Recommendation Team under Wenhao Lu and Ahskay Soni.

I graduated from Indian Insitute of Technology (IIT) Delhi with a Bachelor's in Mechanical Engineering and a minor degree in Computer Science. During my undergraduate, I worked with Prof. Chetan Arora.

Email  /  CV  /  Google Scholar  /  Twitter  /  Github

profile photo
News
Summer 2025
Applied Scientist Intern @ Amazon
Working on pretraining foundational large language models
Sept 2024
Joined NYU Courant
Started Ph.D. in Computer Science
2022 – 2024
Research Fellow @ Microsoft Research India
Worked on extreme classification and recommendation systems
Research Interests

My research focuses on understanding and improving how large language models reason and behave. I'm particularly interested in agentic memory systems that enable persistent, context-aware reasoning, rule-following behavior to ensure models reliably adhere to instructions and constraints, and how knowledge is intrinsically represented within model weights. I also explore architectural innovations that can make these capabilities more robust and scalable.

Ongoing Work
In-Context Reasoning & Alignment

Analyzing rule-following capabilities and updating knowledge priors in weights during inference time

Verified Reasoning

Developing methods for verifiable and trustworthy reasoning in language models

Agentic AI

Safety, aggregation evaluation, long context using memory architectures, and programmable agents

Under Review
* denotes equal contribution, + denotes significant contribution


Domain Faithfulness Through Counterfactually Robust Learning
Ananth Balashankar, Ankit Bhawdwaj, Neelabh Madan+, Thomas Wies, Lakshmi Subramanian ,
Under Review: CLeaR 2026
Publications


In-Context Alignment at Scale: When More is Less
Neelabh Madan*, Lakshminarayanan Subramanian
ICML Workshop MoFA 2025 [Poster]
Paper
Enhancing Tail Performance in Extreme Classifiers by Label Variance Reduction
Anirudh Buvanesh*, Rahul Chand*, Jatin Prakash*, Bhawna Paliwal, Mudit Dhawan, Neelabh Madan, Deepesh Hada, Yashoteja Prabhu, Manik Varma,
ICLR 2024
Paper / Code
A Stitch in Time Saves Nine: A Train-Time Regularizing Loss for Improved Neural Network Calibration
Neelabh Madan*, Ramya Hebbalaguppe*, Jatin Prakash*, Chetan Arora
CVPR 2022 Oral (4.2% acceptance rate)
Paper / Code