AI4AI Lab

Assistant Professor

Imperial College London

s.parbhoo@imperial.ac.uk

Welcome to the AI for Actionable Impact (AI4AI) Research Lab at Imperial College London, led by Dr. Sonali Parbhoo.

About

I am an Assistant Professor at Imperial College London, where I lead the AI for Actionable Impact (AI4AI) lab. My research focuses on developing trustworthy and reliable machine learning methods for high-stakes decision-making, particularly in healthcare applications.

My work spans sequential decision-making under uncertainty, causal inference, and interpretable machine learning to improve clinical care and deepen our understanding of human health. I develop methods that combine theoretical rigor with practical impact, ensuring that AI systems can be safely deployed in medical decision-making contexts.

Research Interests

Reinforcement Learning & Safe RL: Off-policy evaluation, decision-focused learning, and safe policy improvement for sequential medical decisions
Causal Inference: Counterfactual reasoning, causal discovery, and causal Bayesian optimization
Interpretability & Explainability: Concept bottleneck models, feature importance, and guarantee regions for local explanations
Bayesian Methods: Uncertainty quantification and robust inference for medical applications
Healthcare AI: Applications in HIV treatment, critical care (ARDS), and clinical decision support

Background

I received my PhD (summa cum laude) in 2019 from the University of Basel, Switzerland, where I built intelligent models for understanding the interplay between host and virus in the fight against HIV. I was previously a postdoctoral fellow at Harvard University and a Swiss National Science Fellow. I was named a Rising Star in AI in 2021.

My work has been published at leading machine learning conferences (NeurIPS, ICML, AISTATS, AAAI) and medical journals (Nature Medicine, Nature Communications, AMIA, PLoS One, JAIDS). I received the IBM Best Paper Award for Machine Learning in Healthcare at NeurIPS ML4H 2016.

Join Us

I am actively seeking PhD students interested in reinforcement learning, causal inference, Bayesian methods, and interpretability with applications to healthcare. If you’re passionate about developing AI systems that can be safely and effectively deployed in high-stakes medical contexts, please reach out!

news

Nov 15, 2025	AI4AI Lab website launched! Explore our research in trustworthy AI for healthcare.
Jan 01, 2025	Two papers accepted to AISTATS 2025! Topics include safe policy improvement and shortcut mitigation.

selected publications

arXiv

The Alignment Auditor: A Bayesian Framework for Verifying and Refining LLM Objectives

Matthieu Bou, Nyal Patel, Arjun Jagota, and 2 more authors

arXiv preprint arXiv:2510.06096, 2025

Bib

@article{Bou2025AlignmentAuditor,
  title = {The Alignment Auditor: A Bayesian Framework for Verifying and Refining LLM Objectives},
  author = {Bou, Matthieu and Patel, Nyal and Jagota, Arjun and Krishna, Satyapriya and Parbhoo, Sonali},
  journal = {arXiv preprint arXiv:2510.06096},
  year = {2025},
}

MLHC

Improving ARDS Diagnosis Through Context-Aware Concept Bottleneck Models

Anish Narain, Ritam Majumdar, Nikita Narayanan, and 2 more authors

In Proceedings of Machine Learning Research, 2025

arXiv Bib

@inproceedings{Narain2025ARDS,
  title = {Improving ARDS Diagnosis Through Context-Aware Concept Bottleneck Models},
  author = {Narain, Anish and Majumdar, Ritam and Narayanan, Nikita and Marshall, Dominic and Parbhoo, Sonali},
  booktitle = {Proceedings of Machine Learning Research},
  volume = {298},
  pages = {1--32},
  year = {2025},
}

RLC

Concept-Based Off-Policy Evaluation

Ritam Majumdar, J Teversham, and Sonali Parbhoo

In Reinforcement Learning Conference, 2025

arXiv Bib

@inproceedings{Majumdar2025ConceptOPE,
  title = {Concept-Based Off-Policy Evaluation},
  author = {Majumdar, Ritam and Teversham, J and Parbhoo, Sonali},
  booktitle = {Reinforcement Learning Conference},
  year = {2025},
}

AISTATS

Decision-point guided safe policy improvement

Abhishek Sharma, Leo Benac, Sonali Parbhoo, and 1 more author

In The 28th International Conference on Artificial Intelligence and Statistics, 2025

arXiv Bib

@inproceedings{Sharma2025DecisionPoint,
  title = {Decision-point guided safe policy improvement},
  author = {Sharma, Abhishek and Benac, Leo and Parbhoo, Sonali and Doshi-Velez, Finale},
  booktitle = {The 28th International Conference on Artificial Intelligence and Statistics},
  year = {2025},
}

JMLR

Rethinking discount regularization: New interpretations, unintended consequences, and solutions for regularization in reinforcement learning

Sarah Rathnam, Sonali Parbhoo, Siddharth Swaroop, and 3 more authors

Journal of Machine Learning Research, 2024

arXiv Bib

@article{Rathnam2024Discount,
  title = {Rethinking discount regularization: New interpretations, unintended consequences, and solutions for regularization in reinforcement learning},
  author = {Rathnam, Sarah and Parbhoo, Sonali and Swaroop, Siddharth and Pan, Weiwei and Murphy, Susan A and Doshi-Velez, Finale},
  journal = {Journal of Machine Learning Research},
  volume = {25},
  number = {255},
  pages = {1--48},
  year = {2024},
}

MLHC

Decision-Focused Model-based Reinforcement Learning for Reward Transfer

Abhishek Sharma, Sonali Parbhoo, Omer Gottesman, and 1 more author

In Proceedings of Machine Learning Research, 2024

arXiv Bib

@inproceedings{Sharma2024DecisionFocused,
  title = {Decision-Focused Model-based Reinforcement Learning for Reward Transfer},
  author = {Sharma, Abhishek and Parbhoo, Sonali and Gottesman, Omer and Doshi-Velez, Finale},
  booktitle = {Proceedings of Machine Learning Research},
  volume = {252},
  pages = {1--39},
  year = {2024},
}

TMLR

Risk sensitive dead-end identification in safety-critical offline reinforcement learning

TW Killian, Sonali Parbhoo, and M Ghassemi

Transactions on Machine Learning Research, 2023

arXiv Bib

@article{Killian2023Risk,
  title = {Risk sensitive dead-end identification in safety-critical offline reinforcement learning},
  author = {Killian, TW and Parbhoo, Sonali and Ghassemi, M},
  journal = {Transactions on Machine Learning Research},
  year = {2023},
}

NeurIPS

Addressing leakage in concept bottleneck models

Marton Havasi, Sonali Parbhoo, and Finale Doshi-Velez

In Advances in Neural Information Processing Systems, 2022

Bib

@inproceedings{Havasi2022Leakage,
  title = {Addressing leakage in concept bottleneck models},
  author = {Havasi, Marton and Parbhoo, Sonali and Doshi-Velez, Finale},
  booktitle = {Advances in Neural Information Processing Systems},
  volume = {35},
  pages = {23386--23397},
  year = {2022},
}

TMLR

Learning-to-defer for sequential medical decision-making under uncertainty

S Joshi, Sonali Parbhoo, and Finale Doshi-Velez

Transactions on Machine Learning Research, 2021

arXiv Bib

@article{Joshi2021LearningToDefer,
  title = {Learning-to-defer for sequential medical decision-making under uncertainty},
  author = {Joshi, S and Parbhoo, Sonali and Doshi-Velez, Finale},
  journal = {Transactions on Machine Learning Research},
  year = {2021},
}

Nat Commun

Determinants of HIV-1 reservoir size and long-term dynamics during suppressive ART

N Bachmann, C Von Siebenthal, V Vongrad, and 3 more authors

Nature Communications, 2019

Bib

@article{Bachmann2019Determinants,
  title = {Determinants of HIV-1 reservoir size and long-term dynamics during suppressive ART},
  author = {Bachmann, N and Von Siebenthal, C and Vongrad, V and Turk, T and Neumann, K and others},
  journal = {Nature Communications},
  volume = {10},
  number = {1},
  pages = {3193},
  year = {2019},
}

PLOS ONE

Improving counterfactual reasoning with kernelised dynamic mixing models

Sonali Parbhoo, Omer Gottesman, AS Ross, and 4 more authors

PloS One, 2018

Bib

@article{Parbhoo2018Counterfactual,
  title = {Improving counterfactual reasoning with kernelised dynamic mixing models},
  author = {Parbhoo, Sonali and Gottesman, Omer and Ross, AS and Komorowski, M and Faisal, A and Bon, I and others},
  journal = {PloS One},
  volume = {13},
  number = {11},
  pages = {e0205839},
  year = {2018},
}