From Optima to Equilibria: Game-theoretic Modelsof Pattern Analysis and Recognition.

#artificial-intelligence
Share

Since its introduction by John von Neumann in the 1940s, game theory has provided a powerful mathematical language for modeling systems of interacting agents with potentially conflicting objectives. Unlike classical optimization, which relies on a single global objective function, game theory shifts the focus to equilibrium as a fundamental organizing principle for complex systems. In this talk, I will present a long-standing research program developed over the past two decades that explores a game-theoretic perspective on machine learning and pattern recognition. The central idea is to interpret data elements—such as points, labels, or features—as interacting agents, whose relationships are encoded through payoff functions, and whose collective behavior is described in terms of equilibrium configurations. This viewpoint leads to principled formulations of clustering, semi-supervised learning, graph matching, and contextual classification, and reveals deep connections between evolutionary dynamics, quadratic optimization, and graph-based learning. Beyond offering a unifying conceptual framework, equilibrium-based modeling provides natural support for soft assignments, robustness to noise, and distributed computational schemes with well-defined stability properties. Applications to computer vision illustrate how equilibrium thinking can complement, and in some cases reshape, traditional optimization-based approaches in modern AI.


  Date and Time

  Location

  Hosts

  Registration


  • Add_To_Calendar_icon Add Event to Calendar
  • Guangzhou, Guangdong
  • China
  • Contact Event Host

  Speakers

Marcello Pelillo

Topic:

From Optima to Equilibria: Game-theoretic Modelsof Pattern Analysis and Recognition.

Since its introduction by John von Neumann in the 1940s, game theory has provided a powerful mathematical language for modeling systems of interacting agents with potentially conflicting objectives. Unlike classical optimization, which relies on a single global objective function, game theory shifts the focus to equilibrium as a fundamental organizing principle for complex systems. In this talk, I will present a long-standing research program developed over the past two decades that explores a game-theoretic perspective on machine learning and pattern recognition. The central idea is to interpret data elements—such as points, labels, or features—as interacting agents, whose relationships are encoded through payoff functions, and whose collective behavior is described in terms of equilibrium configurations. This viewpoint leads to principled formulations of clustering, semi-supervised learning, graph matching, and contextual classification, and reveals deep connections between evolutionary dynamics, quadratic optimization, and graph-based learning. Beyond offering a unifying conceptual framework, equilibrium-based modeling provides natural support for soft assignments, robustness to noise, and distributed computational schemes with well-defined stability properties. Applications to computer vision illustrate how equilibrium thinking can complement, and in some cases reshape, traditional optimization-based approaches in modern AI.

Biography:

Marcello Pelillo is a Professor of Computer Science and Artificial Intelligence at Ca’ Foscari University of Venice, where he leads the Computer Vision and Machine Learning Lab, and a Changjiang Distinguished Professor at Zhejiang Normal University. He has been the Director of the European Centre for Living Technology (ECLT) and the Chair of the Computer Science School at Ca’ Foscari University for two consecutive terms. He has held visiting research positions at Yale University (USA), University College London (UK), McGill University (Canada), University of Vienna (Austria), York University (UK), NICTA (Australia), Wuhan University (China), South China University of Technology (Guangzhou, China). He is also an external affiliate of the Computer Science Department at Drexel University (USA) and of the Italian Institute of Technology. His research interests are in the areas of computer vision, machine learning and pattern recognition where he has published more than 300 technical papers in refereed journals, handbooks, and conference proceedings. He has been General Chair for ICCV 2017, Program Chair for ICPR 2020, and he is regularly an Area Chair for the top conferences of his field. He is the Chief Editor of Frontiers in Computer Science – Computer Vision, and serves (or has served) on the Editorial Boards of several journals, including IEEE Transactions on Pattern Analysis and Machine Intelligence, Pattern Recognition, IET Computer Vision, Visual Intelligence, SN Computer Science, etc. He is also on the Advisory Board of Springer’s International Journal of Machine Learning and Cybernetics. Prof. Pelillo is a Fellow of the IEEE, the IAPR, and the AAIA, and has served as an IEEE Distinguished Lecturer for several terms. In 2023 he has been elected the World’s AI Top Scientist of the International Artificial Intelligence Industry Alliance (AIIA).
 
 




  Media

Meeting Record Photo 308.61 KiB
Meeting Record Photo 174.49 KiB