Marin Vlastelica

Max Planck Institute for Intelligent Systems

Papers

Total Citations

H-Index

About

Marin Vlastelica is a researcher working at the intersection of machine learning, combinatorial optimization, and robotics, with a particular focus on bridging deep learning with structured algorithmic reasoning and developing agile, adaptive control systems. His most influential contribution, "Differentiation of Blackbox Combinatorial Solvers" (2019, 53 citations), introduced a groundbreaking approach for integrating combinatorial algorithms as differentiable building blocks within neural networks, enabling end-to-end training of architectures that tackle discrete optimization problems — a significant step toward more capable AI systems. This work has become a foundational reference for researchers exploring the fusion of deep learning with classical combinatorial methods. Beyond optimization, Vlastelica has made notable strides in robot learning, particularly in imitation learning and skill acquisition. His work on adversarial imitation from partial or unlabeled demonstrations demonstrates a commitment to making robot learning more practical and data-efficient, reducing reliance on carefully curated expert datasets. His research on diverse skill learning further advances robotic systems capable of navigating complex, multi-constraint environments. Collectively, his publications reflect a coherent research vision: enabling intelligent systems — whether reasoning over combinatorial structures or controlling physical robots — to learn flexibly and effectively from limited, imperfect information.