Topic outline

  • General

    Lecturer: PhD Martin Trapp

    Overview: 
    Machine learning systems designed to support decision-making in real-world scenarios need to perform complex reasoning tasks reliably and efficiently under uncertainty. For this, tractable probabilistic models are a very appealing approach as they enable reliable (exact or approx. with guarantees) and efficient reasoning for a wide range of scenarios by design. The spectrum includes models that enable tractable likelihoods (e.g., normalizing flow and autoregressive models), tractable marginals (e.g., mixture models and bounded-treewidth models), and more complex tractable reasoning tasks (e.g., probabilistic circuits).

    In this seminar, we examine the facets of tractability in probabilistic modelling and focus on emerging literature on probabilistic circuits. The seminar will provide a brief introduction to tractable probabilistic models at the beginning of the course followed by presentations of the participants about recent research in the field.

    Participants are expected to critically review the selected literature and actively participate in the discussions.

    Aim: The aim of the course is to provide an introduction to the field of tractable probabilistic models and highlight active research directions in the field (with a focus on probabilistic circuits).

    Prerequisites: Basics of machine learning and statistics, eg. "Machine Learning: Supervised Methods (CS-E4710)" and latent variable models are recommended.

    Target audience: The course is targeted towards MSc and PhD students interested in widening their knowledge of probabilistic machine learning.

    ECTS: 3 credits (pass/fail).

    Exam: no exam

    Grading: Each student presents one paper and acts as an opponent to one selected paper presented by a fellow student. Attendance and active participation in the discussions are mandatory.

    Format: 12 lectures, 6 weeks, every Friday from 10:15 - 12:00 in Room R030/T4 A140.

    Registration: The course is limited to 10 participants selected based on short (max 1 page) motivation letters. The registration period starts on 27.03.2023 and ends on 21.04.2023 and will be via sisu.aalto.fi.

    Motivation letter: Please send the motivation letter via e-mail to martin.trapp@aalto.fi with the subject "CS-E407516 TPM - YOUR NAME"!