Chirag Nagpal

I am a core member of Meta's GenAI Safety Alignment and Post-training teams where I build recipes for taming Llamas with Reinforcement Learning.

I am a PhD graduate of the School of Computer Science at Carnegie Mellon University. During my PhD I worked on approximate inference techniques for semi-parametric graphical models to estimate censored time-to-event outcomes and counterfactuals. I am was part of the Auton Lab and advised by Prof. Artur Dubrawski.

I am also interested in challenges of applying Data Science and Machine Learning in health.

I have previously worked as a Research Scientist at Google's Brain and Responsible AI Teams.

During my PhD, I have worked as a Summer Associate in Prof. Manuela Veloso's new AI Research Team within the Corporate and Investment Banking division of J.P. Morgan. I also spent a summer working with Kush R. Varshney and others at IBM Research's Thomas J. Watson Research Centre as a Science for Social Good Fellow.

I try to be a good citizen of the School of Computer Science by serving on the Dean's PhD Students Advisory Committee and co-chairing the SCS Dec/5.

Preprints

Bias in Language Models: Beyond Trick Tests and Toward RUTEd Evaluation
Kristian Lum, Jacy Reese Anthis, Chirag Nagpal, and Alex D’Amour
[pdf]
Risk-Aware Framework Development for Disruption Prediction: Alcator C-Mod and DIII-D Survival Analysis
Zander Keith, Chirag Nagpal, Cristina Rea, and R. Alex Tinguely
[pdf]
Transforming and Combining Rewards for Aligning Large Language Models
Zihao Wang, Chirag Nagpal, Jonathan Berant, Jacob Eisenstein, Alex D'Amour, Sanmi Koyejo and Victor Veitch
[pdf]
Helping or herding? Reward Model Ensembles Mitigate but do not eliminate Reward Hacking
Jacob Eisenstein, Chirag Nagpal, Alekh Agarwal, Jonathan Berant and others
[pdf]
Recovering Sparse and Interpretable Subgroups with Heterogeneous Treatment Effects with Censored Time-to-Event Outcomes
Chirag Nagpal, Vedant Sanil and Artur Dubrawski.
[pdf]
Broadening the Time Horizon: Adaptive Risk Scores for Time-to-Event Prediction
Chirag Nagpal, Artur Dubrawski and Berk Ustun.
[pdf]

Publications

Participatory Systems for Personalized Prediction
Hailey Joren, Chirag Nagpal, Katherine Heller and Berk Ustun.
NeurIPS - Neural Information Processing Systems Conference 2023
[pdf]
Counterfactual Phenotyping with Censored Time-to-Events
Chirag Nagpal, Mononito Goswami, Keith Dufendach and Artur Dubrawski.
KDD - ACM Conference on Knowledge Discovery and Data Mining 2022
[pdf] [code]
Deep Cox Mixtures for Survival Regression.
Chirag Nagpal, Steve Yadlowsky, Negar Rostamzadeh and Katherine Heller.
MLHC - Machine Learning for Healthcare Conference 2021
[pdf] [code]
Interpretable Subgroup Discovery in Treatment Effect Estimation with Application to Opioid Prescribing Guidelines.
Chirag Nagpal, Dennis Wei, Bhanukiran Vinzamuri, M. Shekhar, S. E. Berger, S. Das and Kush Varshney.
CHIL - ACM Conference on Health, Inference and Learning 2020 (Spotlight Presentation)
[pdf] [code] [notebook]
Deep Survival Machines: Fully Parametric Survival Regression and Representation Learning for Censored Data with Competing Risks.
Chirag Nagpal, Xinyu Li, and Artur Dubrawski.
JBHI - IEEE Journal of Biomedical and Health Informatics
(and) NeurIPS Machine Learning for Health Workshop 2019 (Spotlight Presentation, top 3% of over 300 papers)
[pdf] [talk] [code]
Dynamically Personalized Detection of Hemorrhage.
Chirag Nagpal, Xinyu Li, Michael Pinsky, and Artur Dubrawski.
MLHC - Machine Learning for Healthcare Conference 2019
[pdf]

Workshop Presentations

Deep Parametric Time-to-Event Regression with Time-Varying Covariates
Chirag Nagpal, Vincent Jeanselme and Artur Dubrawski.
AAAI Spring Symposium on Survival Prediction 2021
[pdf] [code]
Bayesian Consensus: Consensus Estimates from Miscalibrated Instruments under Heteroscedastic Noise.
Chirag Nagpal, Robert E. Tillman, Prashant P. Reddy, and Manuela M. Veloso.
NeurIPS Robust AI in Financial Services Workshop 2019
[pdf]
An Entity Resolution Approach to Isolate Instances of Human Trafficking Online.
Chirag Nagpal, Kyle Miller, Benedikt Boecking, and Artur Dubrawski.
Bloomberg Data for Good Exchange 2017
[pdf]

Professional Service

Organizer

2023	AAAI Fall Symposium - Survival Prediction - Algorithms, Challenges and Applications

Area Chair

2024	Machine Learning for Healthcare Conference (MLHC)

Reviewer

2020	NeurIPS Machine Learning for Health Workshop (ML4H)
2020	Neural Information Processing Systems (NeuRIPS)
2020	ACL Student Research Workshop (ACL-SRW)
2020	Machine Learning for Healthcare Conference (MLHC)
2020	ICLR ML in Real-Life Workshop (ML-IRL)
2020	ACM Conference on Health, Inference and Learning (CHIL)
2019	NeurIPS Machine Learning for Health Workshop (ML4H)

Chirag Nagpal

Preprints

Bias in Language Models: Beyond Trick Tests and Toward RUTEd Evaluation

Risk-Aware Framework Development for Disruption Prediction: Alcator C-Mod and DIII-D Survival Analysis

Transforming and Combining Rewards for Aligning Large Language Models

Helping or herding? Reward Model Ensembles Mitigate but do not eliminate Reward Hacking

Recovering Sparse and Interpretable Subgroups with Heterogeneous Treatment Effects with Censored Time-to-Event Outcomes

Broadening the Time Horizon: Adaptive Risk Scores for Time-to-Event Prediction

Publications

Participatory Systems for Personalized Prediction

Counterfactual Phenotyping with Censored Time-to-Events

Deep Cox Mixtures for Survival Regression.

Interpretable Subgroup Discovery in Treatment Effect Estimation with Application to Opioid Prescribing Guidelines.

Deep Survival Machines: Fully Parametric Survival Regression and Representation Learning for Censored Data with Competing Risks.

Dynamically Personalized Detection of Hemorrhage.

Software

auton-survival: an open-source package for Regression, Counterfactual Estimation, Evaluation and Phenotyping with Censored Time-to-Event Data

Workshop Presentations

Deep Parametric Time-to-Event Regression with Time-Varying Covariates

Bayesian Consensus: Consensus Estimates from Miscalibrated Instruments under Heteroscedastic Noise.

An Entity Resolution Approach to Isolate Instances of Human Trafficking Online.

Clinical Abstracts

Interpretable Treatment Prioritization Rule Identifies Diabetic Patients Who Can Benefit from Prompt Coronary Revascularization.

Novel Machine Learning Technique Defines Patients Who Benefit from Off-Pump CABG

Phenogrouping of hemorrhagic trauma patients using latent variable machine learning.

Accuracy of identifying venous thromboembolism by administrative coding compared to manual review.

Professional Service

Organizer

Area Chair

Reviewer

Teaching

10-708 Probabilistic Graphical Models, Fall 2020

11-761/661 Language and Statistics, Fall 2019

Personal