Research
My research is centered on a fundamental challenge in artificial intelligence: enabling machines to reason under uncertainty in ways that are both powerful and reliable.
To achieve this goal, I focus on designing probabilistic generative models that can capture complex data distributions while retaining the ability to perform exact inference, which I believe is an essential property for building interpretable and dependable systems. I also try to bridge learning and reasoning through neuro-symbolic approaches, where structured domain knowledge is embedded into probabilistic models to enhance their robustness and alignment with human understanding. Finally, I am also interested in meta-learning, with the goal of enabling systems to generalize quickly from minimal data / limited supervision.
Across these directions, my aim and hope is to advance foundations of trustworthy and reliable AI systems that express calibrated uncertainty (i.e. they know what they don't know), adapt gracefully to unfamiliar settings, and can be meaningfully understood and guided by humans.
Probabilistic Generative Models
Expressive generative models that retain tractable exact inference — spanning Probabilistic Circuits, Normalizing Flows, and hybrid architectures like Probabilistic Flow Circuits.
Tractable Sharpness-Aware Learning of Probabilistic Circuits
AAAI '26Sahil Sidheekh, Hrithik Suresh, Vishnu Shreeram, Sriraam Natarajan, Narayanan C. Krishnan
The 40th Annual AAAI Conference on Artificial Intelligence (AAAI), 2026
Tractable Representation Learning with Probabilistic Circuits
TMLR '25Steven Braun, Sahil Sidheekh, Antonio Vergari, Marius Mundt, Sriraam Natarajan, Kristian Kersting
Transactions on Machine Learning Research (TMLR), 2025
Building Expressive and Tractable Probabilistic Generative Models: A Review
IJCAI '24Sahil Sidheekh, Sriraam Natarajan
The 33rd International Joint Conference on Artificial Intelligence (IJCAI), 2024
VQ-Flows: Vector Quantized Local Normalizing Flows
UAI '22Sahil Sidheekh, Chris B. Dock, Tushar Jain, Radu Balan, Maneesh K. Singh
The 38th Conference on Uncertainty in Artificial Intelligence (UAI), 2022· 3 citations
On Characterizing GAN Convergence Through Proximal Duality Gap
ICML '21Sahil Sidheekh, Aroof Aimen, Narayanan C. Krishnan
Proceedings of the 38th International Conference on Machine Learning (ICML), PMLR 139, 9660–9670, 2021· 68 citations
Neuro-Symbolic AI
Embedding structured domain knowledge and human feedback into probabilistic models for robust, human-aligned decision-making and reliable multi-modal fusion.
Human-Allied Relational Reinforcement Learning
ACS '25FG Darvishvand, H Shindo, Sahil Sidheekh, Kristian Kersting, Sriraam Natarajan
The Twelfth Annual Conference on Advances in Cognitive Systems (ACS), 2025
A Unified Framework for Human-Allied Learning of Probabilistic Circuits
AAAI '25Sahil Sidheekh, Athresh Karanam, Saurabh Mathur, Sriraam Natarajan
The 39th Annual AAAI Conference on Artificial Intelligence (AAAI), 2025
Credibility-Aware Multi-Modal Fusion Using Probabilistic Circuits
AISTATS '25Sahil Sidheekh, Pranuthi Tenali, Saurabh Mathur, Erik Blasch, Kristian Kersting, Sriraam Natarajan
The 28th International Conference on Artificial Intelligence and Statistics (AISTATS), 2025
Human-in-the-loop or AI-in-the-loop? Automate or Collaborate?
AAAI '25Sriraam Natarajan, Sahil Sidheekh, Saurabh Mathur, Wolfgang Stammer, Kristian Kersting
The 39th Annual AAAI Conference on Artificial Intelligence (AAAI), 2025
Credibility-Aware Reliable Multi-Modal Fusion Using Probabilistic Circuits
AAAI '24Sahil Sidheekh, Pranuthi Tenali, Saurabh Mathur, Erik Blasch, Kristian Kersting, Sriraam Natarajan
The 2nd Workshop on Deployable AI (DAI), co-located at AAAI, 2024
On the Robustness and Reliability of Late Multi-Modal Fusion using Probabilistic Circuits
FUSION '24Sahil Sidheekh, Pranuthi Tenali, Saurabh Mathur, Erik Blasch, Sriraam Natarajan
The 27th International Conference on Information Fusion (FUSION), 2024
Bayesian Learning of Probabilistic Circuits with Domain Constraints
UAI '23Sahil Sidheekh, Athresh Karanam, Saurabh Mathur, Sriraam Natarajan
The 6th Workshop on Tractable Probabilistic Modeling (TPM), co-located at UAI, 2023
Meta-Learning & Few-Shot Learning
Fast generalization from limited data — stress-testing meta-learning under distribution shift and understanding when adaptation helps or hurts.
Leveraging Task Variability in Meta-Learning
SN Comp Sci '23Aroof Aimen, Bharat Ladrecha, Sahil Sidheekh, Narayanan C. Krishnan
SN Computer Science, 4(5), 539. Springer, 2023· 1 citations
Adaptation: Blessing or Curse for Higher Way Meta-Learning
IEEE TAI '23Aroof Aimen, Sahil Sidheekh, Bharat Ladrecha, Hansin Ahuja, Narayanan C. Krishnan
IEEE Transactions on Artificial Intelligence, 5(4), 1844–1856, 2023· 5 citations
