I am a final year Ph.D. candidate at Mila & Université de Montréal, advised by Ioannis Mitliagkas.
Currently, I am a visiting researcher at Meta FAIR working on pretraining large language models with the memory and generalization team.
My research interests can be outlined broadly as follows.
- Learning algorithms for efficient adaptation under distribution shifts and robustness to spurious correlations.
- Disentangled (causal) representation learning for sample efficient generalization and interpretability.
- In-context learning and prior-fitted networks for probabilistic (causal) inference.
For further details, please refer to my thesis proposal (report, slides).
My research is supported by the FRQNT doctoral fellowship, and I am deeply grateful for the amazing collaborations that have enrinched my Ph.D. journey.
I am advised by Kartik Ahuja and Pascal Vincent
under the Meta AIM Program,
where I work on compositional generalization and multi-token prediction for large language models.
I also did a summer internship at Microsoft Research Cambridge,
where I worked on amortized inference and in-context learning with
Cheng Zhang and Meyer Scetbon.
Further, in the initial years of my Ph.D. I worked with
Vasilis Syrgkanis at Stanford on causal inference.
Prior to starting my Ph.D., I was a research fellow at Microsoft Research India, where I worked with Amit Sharma on trustworthy machine Learning from the lens of causality, specifically on out-of-distribution generalization, privacy, and explainable machine learning.
Earlier, I completed my undergraduate in Mathematics and Computer Science from the Indian Institute of Technology, Kanpur.
Select Publications & Preprints
-
Amortized Inference of Causal Models via Conditional Fixed-Point Iterations
Divyat Mahajan*, Jannes Gladrow, Agrin Hilmkil, Cheng Zhang, Meyer Scetbon*
Preprint. Under Review.
[arxiv]
[code]
-
Compositional Risk Minimization
Divyat Mahajan, Mohammad Pezeshki, Charles Arnal, Ioannis Mitliagkas, Kartik Ahuja, Pascal Vincent
ICML 2025
[arxiv]
[code]
[presentation]
[poster]
[twitter]
-
Empirical Analysis of Model Selection for Heterogeneous Causal Effect Estimation
Divyat Mahajan, Ioannis Mitliagkas, Brady Neal, Vasilis Syrgkanis
ICLR 2024 (Spotlight)
[arxiv]
[code]
[presentation]
[poster]
[twitter]
-
Additive Decoders for Latent Variables Identification and Cartesian-Product Extrapolation
Sébastien Lachapelle*, Divyat Mahajan*, Ioannis Mitliagkas, Simon Lacoste-Julien
NeurIPS 2023 (Oral)
[arxiv]
[code]
[blog]
[talk(conference)]
[talk(reading group)]
[presentation]
[poster]
-
Interventional Causal Representation Learning
Kartik Ahuja, Divyat Mahajan, Yixin Wang, Yoshua Bengio
ICML 2023 (Oral)
[arxiv]
[code]
[talk]
[presentation]
[poster]
-
Towards efficient representation identification in supervised learning
Kartik Ahuja*, Divyat Mahajan*, Vasilis Syrgkanis, Ioannis Mitliagkas
CleaR 2022
[arxiv]
[code]
[talk]
[presentation]
[poster]
-
Domain Generalization using Causal Matching
Divyat Mahajan, Shruti Tople, Amit Sharma
ICML 2021 (Oral)
[arxiv]
[code]
[talk]
[presentation]
[poster]
-
Preserving Causal Constraints in Counterfactual Explanations for Machine Learning Classifiers
Divyat Mahajan, Chenhao Tan, Amit Sharma
CausalML@NeurIPS 2019 (Oral)
[arxiv]
[code]
[talk]
[presentation]
[poster]
Select Awards & Honours
Software
-
RobustDG
Toolkit for Building Robust ML models that generalize to unseen domains | Github | Microsoft