DATA Talk :``Shape Constraints Meet Kernel Machines''

Zoltan Szabo (London School of Economics) -
Data Science

Date: November 4th 2021
Location: Eurecom - Eurecom

Abstract: Shape constraints (such as non-negativity, monotonicity, convexity, or supermodularity) provide a principled way to encode prior information in predictive models with numerous successful applications in econometrics, finance, biology, reinforcement learning, and game theory. Incorporating this side information in a hard way (for instance at all point of an interval) however is an extremely challenging problem. In this talk I am going to present a unified and modular convex optimization framework to encode hard affine constraints on function values and derivatives into the flexible class of reproducing kernel Hilbert spaces. The efficiency of the technique is illustrated in the context of joint quantile regression (analysis of aircraft departures), convoy localization and safety-critical control (piloting an underwater vehicle while avoiding obstacles). [This is joint work with Pierre-Cyril Aubin-Frankowski.] References: 1) Pierre-Cyril Aubin-Frankowski, Zoltan Szabo. Hard Shape-Constrained Kernel Machines. NeurIPS-2020. Link: https://proceedings.neurips.cc/paper/2020/hash/03fa2f7502f5f6b9169e67d17cbf51bb-Abstract.html [real-valued output] 2) Pierre-Cyril Aubin-Frankowski, Zoltan Szabo. Handling Hard Affine SDP Shape Constraints in RKHSs. TR (under submission to JMLR). Link: http://arxiv.org/abs/2101.01519 [vector-valued output] Bio: Zoltan Szabo is a Professor of Data Science at the Department of Statistics, London School of Economics. His research interest is statistical machine learning with focus on kernel methods, information theory (https://bitbucket.org/szzoli/ite-in-python/), scalable computation, and their applications. Zoltan enjoys helping and interacting with the machine learning (ML) and statistics community at various levels. He serves/served as (i) an Area Chair of ICML, NeurIPS, COLT, AISTATS, UAI, IJCAI, ICLR, (ii) the moderator of statistical machine learning (stat.ML) on arXiv, (iii) the founder and Program Chair of the Data Science Summer School (DS^3, 2017-2020), (iv) an editorial board member of JMLR and associate editor of the journal Mathematical Foundations of Computing, (v) a reviewer of various journals in ML and statistics, (vi) a reviewer of European (ERC), Israeli (ISF) and Swiss (SNSF) grant applications, (vii) a mentor of newcomers (NeurIPS, ICML). For further details, please see his website (https://zoltansz.github.io/). EURECOM Data Science Seminars: https://ds.eurecom.fr/seminars/ https://mediaserver.eurecom.fr/channels/#data-science-seminars (internal)


Permalink: https://www.eurecom/seminar/104618