Sahil Sidheekh: “Building expressive and tractable generative models for reliable, human-allied AI”

I work at the intersection of probabilistic generative models and human-allied learning. My primary research focuses on building AI systems that are both expressive and interpretable, i.e models that can reason, explain, and adapt under uncertainty. Much of my work centers on probabilistic circuits (PCs), exploring how structural constraints enable exact inference while still retaining the flexibility of deep models. This includes developing better optimization schemes, learning methods and hybrid deep-probabilistic architectures for PCs. A second thread of my research focuses on making AI systems more reliable and human-aligned. To this end, I work on credibility-aware multimodal fusion, human-allied learning frameworks, and models that incorporate domain constraints or human feedback to improve controllability and trust. My long-term goal is to build principled, tractable, and transparent generative models that can meaningfully collaborate with people.