All Years Seminars
[INGI] 2025-08-26 (11:00) : Robustness for Tabular Machine Learning: A Back-and-Forth Journey between Research and Industry
At Shannon room
Speaker :
Maxime Cordy (SnT Université du Luxembourg)
Abstract : Adversarial attacks are widely recognized as a critical security threat for machine learning, but most were originally designed in the context of image recognition, where arbitrary pixel-level perturbations are applied. Such attacks often fail to capture the realities of domains governed by strict constraints on valid inputs. This is particularly the case in tabular machine learning, where only feasible feature-level perturbations can occur. This talk explores the gap between classical adversarial attack formulations and real-world applicability. I will review our recent research on constrained feature-space attacks, which aim to generate realistic adversarial examples under domain-specific restrictions. Drawing on our own experience in applying these methods to industrial use cases, I will highlight the challenges of evaluating robustness in practice and discuss opportunities for new research in this area.
2025-08-20 (14h) : Efficient Distance-Adaptive Subgradient Methods
At Euler building (room A.002)
Speaker :
Anton Rodomanov (CISPA, Germany)
Abstract :
Subgradient methods based on the idea of gradient normalization are an appealing class of algorithms for convex optimization because they automatically adapt to the local growth rate of the objective and require no problem-specific parameters---except for a reasonably accurate estimate of the initial distance to the solution. Overestimating or underestimating this distance can, however, substantially slow convergence. We address this limitation by incorporating into normalization-based methods a recently proposed technique that dynamically estimates the distance via the displacement of iterates from the starting point. This removes the need for any problem-specific information, while preserving the adaptability of the base methods and ensuring rigorous convergence guarantees. Illustrative examples include nonsmooth Lipschitz functions, Lipschitz- or Hölder-smooth functions, functions with high-order Lipschitz derivatives, quasi-self-concordant functions, and $(L_0, L_1)$-smooth functions. We further extend the approach to problems with functional constraints using a simple switching strategy.
2025-06-17 (14h) : Matrix Factorization and Approximation of Nonnegative Rank Two
At Euler building (room A.002)
Speaker :
Van Dooren, Paul
Abstract :
We consider the problem of finding the best nonnegative rank two approxi-
mation of an arbitrary nonnegative matrix. We first provide a parametrization
of all possible nonnegative factorizations of a nonnegative matrix of rank 2. We
then use this result to construct a suboptimal, but cheaply computable, solution
of the nonnegative rank 2 approximation problem for an arbitrary nonnegative
matrix input; this can then be used as a starting point for the Alternating Least
Squares method, resulting in both an improved computational complexity and
an enhanced output quality. We provide numerical experiments to support these
claims. We also look at some variants of the problem, including symmetry con-
straints and three-way factorizations.
This is joint work with Etna Lindy and Vanni Noferini (Aalto University).
2025-05-20 (14h) : Model-Based and Data-Driven Interventions in Network Games
At Euler building (room A.002)
Speaker :
Nima Monshizadeh (University of Groningen)
Abstract :
In modern cyber-physical-human systems like power grids and traffic networks, self-interested decisions by users or firms often lead to inefficiencies such as congestion, blackouts, and systemic risks. These challenges underscore the need for effective intervention mechanisms to coordinate self-interested behavior. However, designing effective interventions with guaranteed performance is challenging, as planners typically lack detailed knowledge of users’ private preferences or behaviors. This informational gap and privacy considerations complicate the prediction of user responses and hinder the development of suitable control strategies. In this talk, we examine the problem of intervention design in network games, where agents’ cost or utility functions are interdependent; as exemplified by applications such as multi-commodity markets. The central question is how the type of information available to the planner — ranging from full to partial knowledge of utility functions, network structure, or desired target profiles — shapes the design of effective interventions. We also give special attention to scenarios in which the planner has no a priori knowledge of agent cost functions and must rely instead on historical observations of agent actions and past interventions. Given these diverse informational settings, we discuss a range of static, dynamic, adaptive, and data-driven intervention strategies aimed at steering the system toward socially desirable outcomes. The results illustrate how combining control-theoretic, game-theoretic, and data-driven insights enables the design of interventions that are effective under various informational constraints.
2025-05-06 (14h) : Axisymmetric magnetic control in ST40
At Euler building (room A.002)
Speaker :
Benjamin Vincent (UCLouvain)
Abstract :
Tokamaks magnetically confine plasmas (i.e. ionised gas) up to temperatures where the nuclear fusion reaction is sustainable. The real-time operation of a tokamak relies on a Plasma Control System, which is responsible for data acquisition, pulse supervision and feedback control. The plasma current, position, and shape can be actuated by the poloidal field coils.
The talk will introduce the concept of tokamaks and magnetically confined plasmas. The axisymmetric magnetic control problems will be presented and illustrated in the context of the spherical tokamak ST40.
Seminars
INMA Contact Info
Mathematical Engineering (INMA)
L4.05.01
Avenue Georges Lemaître, 4
+32 10 47 80 36
secretaire-inma@uclouvain.be
Mon – Fri 9:00A.M. – 5:00P.M.
JOIN US