Martin Szummer

Martin Szummer is a Researcher at the University of Cambridge, UK, and part of the Spoken Dialogue Systems group. Previously, he worked at Microsoft Research in Cambridge (Machine Learning and Perception Group). He holds a Ph.D. in Machine Learning from MIT (2002) and an M.S. from the MIT Media Lab.

Research Focus: Szummer’s work spans probabilistic models, deep learning, reinforcement learning, and natural language processing. He explores methods to improve interactions with users through dialogue systems and develops techniques to manage complex data via Bayesian inference, semi-supervised learning, and structured prediction. His contributions include probabilistic programming frameworks, big data processing on cloud clusters, and applications in text mining, image recognition, and computational advertising.

Notable Achievements:

• Recipient of the SIGDIAL 2013 Best Paper Award for dialogue system adaptation research.
• Developed the Temporal Texture database, foundational for video analysis and synthesis.

Collaborations & Mentoring: Advised prominent students like Percy Liang (Stanford) and Volodymyr Mnih (DeepMind). His work at Microsoft Research led to innovations in POMDP-based dialogue systems and semi-supervised learning.

Technical Contributions:

• Publications on dialogue systems, probabilistic modeling, and web-scale data mining.
• Authored datasets like the Temporal Texture collection, widely used in computer vision research.