Γενικές Ανακοινώσεις

New Year's Corelab Seminar
Friday, January 5, 2024
Multimedia Amphitheater of the National Technical University of Athens (basement of NTUA's Central Library).


15:00 - 15:50     Vasilis Syrganis, Stanford University
                           Causal Machine Learning for Trustworthy Data-Driven Decisions

Abstract: The ease of use of modern machine learning techniques has led to a new era of machine learning mis-use, where data scientists often draw causal conclusions based on ML predictive models. Data-driven decision making necessitates the use of causal reasoning. Causal inference addresses exactly the statistical questions that underpin decision problems, but has been classically developed in idealized scenarios. New application domains, data modalities and richer datasets necessitate the merging of causal inference with machine learning, leading to the new area of Causal Machine Learning. I will go over two causal machine learning methodologies that we developed in the process, which advance causal inference techniques in the presence of high-dimensionality and in the dynamic treatment regime. I will conclude on practical challenges in the field that still remain not satisfactorily addressed if we want to make causal machine learning as easily accessible as predictive machine learning tools.

Bio: Vasilis Syrgkanis ( https://profiles.stanford.edu/vasilis-syrgkanis?tab=bio ) is an Assistant Professor in Management Science and Engineering and (by courtesy) in Computer Science and Electrical Engineering, in the School of Engineering at Stanford University and a member of the Institute for Computational and Mathematical Engineering. His research interests lie in the areas of machine learning, causal inference, econometrics, online and reinforcement learning, game theory/mechanism design and algorithm design. Until August 2022, he was a Principal Researcher at Microsoft Research, New England, where he was a member of the EconCS and StatsML groups. During his time at Microsoft, he co-led the project on Automated Learning and Intelligence for Causation and Economics (ALICE) and was a co-founder of EconML, an open-source python package for causal machine learning. He received his Ph.D. in Computer Science from Cornell University. His research has received best paper awards in COLT, EC, and NeurIPS. He is the recipient of a 2022 Amazon Research Award and of the 2023 Bodossaki Distinguished Young Scientist award.


15:50 - 16:30    Ioannis Panageas, University of California Irvine
                          Learning Dynamics for Nash and Coarse Correlated Equilibria in Bimatrix Games

Abstract: In this talk, we will be focusing on learning in two-player games. We will provide a brief introduction on the possible behaviors of learning algorithms and will mention various techniques that have been heavily used and guarantee convergence to Nash equilibria in zero-sum games. Finally we will show how these techniques can be used to learn Nash equilibria in rank-1 games and what implications these techniques can have for general-sum games.

Bio: Ioannis Panageas ( https://panageas.github.io/ ) is an Assistant Professor of Computer Science at UC Irvine. He is interested in the theory of computation, machine learning and its interface with non-convex optimization, dynamical systems, learning in games, statistics and multi-agent reinforcement learning. Before joining UCI, he was an Assistant Professor at Singapore University of Technology and Design. Prior to that he was a MIT postdoctoral fellow. He received his PhD in Algorithms, Combinatorics and Optimization from Georgia Tech in 2016, a Diploma in EECS from National Technical University of Athens, and a MS in Mathematics from Georgia Tech. He is the recipient of the 2019 NRF fellowship for AI.


16:30 - 17:10    Giannis Fikioris, Cornell University
                           Liquid Welfare guarantees for No-Regret Learning in Sequential Budgeted Auctions

Abstract: In this talk, I'm going to talk about welfare in sequential auctions when the buyers are budget-limited. In practice, these buyers are using complicated algorithms that adaptively tune their bids to maximize their utility while not spending their budget too early or too late. These algorithms are often proprietary and unknown to the general public, especially when the buyer is a company (e.g., Google, Microsoft). However, at the very least, these algorithms satisfy the no-regret assumption: in retrospect using a fixed action would not have led to much more total utility. Our contribution is to show that under this minimal assumption for every buyer, the welfare is not much lower than the maximum one.

Specifically, we use the generalization of the no-regret assumption, that the resulting utility of each buyer is within a \gamma  factor (where \gamma \geq 1) of the utility achievable by shading her value with the same factor at each iteration. Under this assumption, we show a \gamma+1/2+O(1/\gamma) price of anarchy for welfare in first-price auctions, assuming buyers have additive valuations. This positive result is in stark contrast to sequential second-price auctions, where even with \gamma=1, the resulting liquid welfare can be arbitrarily smaller than the maximum welfare. In first-price auctions, we prove a lower bound of \gamma on the liquid welfare loss under the above assumption, making our bound asymptotically tight. For the case when \gamma=1 our theorem implies a price of anarchy upper bound of about 2.41; we show a lower bound of 2 for that case.

Βio: Giannis Fikioris ( https://giannisfikioris.org/ ) is a 4th year PhD student in the theory group at Cornell University, advised by Éva Tardos and supported by the NDSEG fellowship. He works on Algorithmic Game Theory and Learning Theory, with his most recent work on Mechanism Design and/or Online Learning with global constraints. During Fall 2023, he was a Student Researcher in the Market Algorithms team at Google Research. Before coming to Cornell, he received a BSE from the department of Electrical and Computer Engineering at the National Technical University of Athens.

============================================================================================