About

I am a PhD student in the Computer Science department at the University of Toronto and Vector Institute. My Supervisors are Sheila McIlraith and Eldan Cohen. I did my undergraduate studies in Software Engineering at Sharif University of Technology.

Research Interests

  • Exact Algorithms
  • Machine Learning
  • Interpretable AI
  • Operations Research
  • Links

    Research Summary

    The recent advancements in artificial intelligence have revolutionized almost every aspect of our lives. Modern deep learning models have achieved superhuman performance in numerous tasks, yet still are mostly black-box models lacking guarantees that classical approaches have provided for decades. Such guarantees can be generally divided into optimality and constraint satisfaction, both heavily studied in the exact optimization literature.

    Most use cases of constraint satisfaction can be gathered under the umbrella term interpretability and include safety-critical applications, privacy and security concerns, or integration of audits by domain experts. With interpretability comes human understandability, the ease of formal reasoning, and more direct control on the properties of the system. Optimality guarantees are also closely related to constraint satisfaction, as they allow finding the boundary of solution quality within the requirements.

    I aim to study how exact optimization methods can be utilized to enhance or complement machine learning algorithms. I believe not only that there is huge potential in integrating classical AI notions such as formal languages, knowledge representation, and constraint satisfaction problems into our modern discussion of AI, but that it is the only way forward to a more powerful and secure future.

    Publications

    Neural Sequence Generation with Constraints via Beam Search with Cuts: A Case Study on VRP

    Pouya Shati, Eldan Cohen, Sheila McIlraith

    SAT-Based Learning of Compact Binary Decision Diagrams for Classification

    Slides

    Pouya Shati, Eldan Cohen, Sheila McIlraith

    SAT-based optimal classification trees for non-binary data

    Pouya Shati, Eldan Cohen, Sheila McIlraith

    Slides

    Code

    SAT-Based Approach for Learning Optimal Decision Trees with Non-Binary Features

    Pouya Shati, Eldan Cohen, Sheila McIlraith

    Slides

    Code

    The Evolution of Indirect Reciprocity Under Action and Assessment Generosity

    Laura Schmid, Pouya Shati, Christian Hilbe, Krishnendu Chatterjee

    Get In Touch

    I would love to hear from you on interesting points of discussion or potential collaborations!

    • Address

      283 D.L. Pratt Building
      6 King's College Road
      Toronto, ON
      M5S 3H5
    • Email

      pouya [at] cs [dot] toronto [dot] edu