Events

Louis Bigo (University of Lille) – Modeling language in musical scores

In this talk, we focus on two estimation problems: i) sparse regression and ii) sparse and smooth signal recovery. The first one is known to be NP-hard; we show that the second one is equivalent to a submodular minimization problem and, hence, it is polynomially solvable. For both problems, we derive a sequence of strong conic relaxations. The proposed rank-one strengthening can be interpreted as a non-separable, non-convex, unbiased sparsity-inducing regularizer, which dynamically adjusts its penalty according to the shape of the estimation error function without inducing bias for the sparse solutions. Computational experiments with benchmark datasets show that the proposed conic formulations are solved fast and result in near-optimal estimators for non-convex L0-problems. Moreover, the resulting estimators also outperform L1 approaches from a statistical perspective, achieving high prediction accuracy and good interpretability.

Zhang Hao (4Paradigm Technology) – Parallel Computing in Machine Learning Database: A Case Study of OpenMLDB

Hengky Wirawan, Cheng Yue (Dyson Singapore) – Dyson Manufacturing Landscape

Algorithmic recourse aims to recommend an informative feedback to overturn an unfavourable machine learning decision. We introduce in this paper the Bayesian recourse, a model-agnostic recourse that minimizes the posterior probability odds ratio. Further, we present its min-max robust counterpart with the goal of hedging against future changes in the machine learning model parameters. The robust counterpart explicitly takes into account possible perturbations of the data in a Gaussian mixture ambiguity set prescribed using the optimal transport (Wasserstein) distance. We show that the resulting worst-case objective function can be decomposed into solving a series of two-dimensional optimization subproblems, and the min-max recourse finding problem is thus amenable to a gradient descent algorithm. Contrary to existing methods for generating the robust recourse, the robust Bayesian recourse does not require a linear approximation step. The numerical experiment demonstrates the effectiveness of our proposed robust Bayesian recourse facing model shifts.

Subhajit Roy (Indian Institute of Technology Kanpur) – Deferred concretization in symbolic execution via fuzzing

Master of Architecture Lecture series 2022 – ‘Design beyond Zero’

Averaging data distributions is a core subroutine throughout data science. Wasserstein barycenters (a.k.a. Optimal Transport barycenters) provide a natural approach for this problem that captures the geometry of the data, and are central to diverse applications in machine learning, statistics, and computer graphics. However, despite considerable attention, it remained unknown whether Wasserstein barycenters can be computed in polynomial time. Our recent work provides a complete answer to this question and uncovers the subtle dependence of the answer on the dimension due to the continuous nature of the problem.

In this talk, Jason will explain these results and how they fit more broadly into the research program on algorithms for “structured” Multimarginal Optimal Transport problems.
Joint work with Enric Boix

Conversations in Design, Technology, and Society with Dr. Adam Drazin and Dr. Pauline Garvey

Great decisions are made with great models and great data. Over the last couple of years, we’ve seen the advantages for companies that can react quickly to the changes in markets, technology, and consumer behaviour. In this session we’ll talk about how Google handles data and how we’re helping companies prepare for their data-driven futures.