General and specific goals of SeCVPR project

  • Contribute to advance the state of the art.
  • Create the basis for implementing a future European network of excellence on the project theme.
  • Develop a security model for adversarial classification based on game theory.
  • Develop prototypes of secure recognition systems and validate them on real applications (computer security and biometric recognition).
  • Create a "think tank” on the research theme through collaboration with the Italian and European partners involved in the project.
  • Disseminate the results at the European level and involve some Sardinian companies in the validation of the project results in order to explore their future industrial exploitation.
  • Publish critical reviews on the theoretical foundations of the security of automatic recognition in adversarial environments (papers on secure pattern recognition based on game theory; papers on new methods of design of automatic recognition in adversarial environments based on the paradigm of "security by design";  papers on the methods developed for computer security and biometric identification)
  • Release of "open source" software that implements the secure recognition algorithms developed in the project.

 

These are the main steps of the project:

  • Analysis of foundations of security of pattern recognition systems: this step is aimed at the analysis of possible theoretical foundations of security for pattern recognition systems, analyzing the state of the art of related scientific disciplines that could contribute to these foundations (Game theory and security of pattern recognition systems; Statistical learning theory and security of pattern recognition systems; Computer security)
  • Development of a new security model based on evolutionary game theory: this activity focuses on game theory for the development of a new model for the security of pattern recognition systems in adversarial environments. The models based on game theory proposed so far are based on the classical notion of "Nash equilibrium", which does not take into account the dynamics that lead to these equilibria. This limits their applicability in real‐world settngs, where decision‐making strategies of the "players" are determined by the learning process. This project tries to consider consider instead the "evolutionary game theory" (an evolution of the classic theory), which interprets the concept of equilibrium as the result of a learning process (or evolution) that takes place over time.
  • Design of secure pattern recognition systems: this step is devoted to the development of new methods for the design of pattern recognition systems according to the paradigm of “security by design”. This activity focuses on the development of a new design cycle for pattern classifiers where the designer attempts to proactively anticipate the adversary’s move, that is, he tries to figure out the most relevant threats and attacks that may be worth countering before deploying the classifier.
  • Case studies and development of proof­‐of‐concept systems: this last step is aimed at validation of the proposed methods on three relevant case studies and at the development some proof­‐of‐concept systems.