[Dottorcomp] Seminari di Matematica Applicata. Mercoledì 19 aprile. Axel Parmentier.

[Stefano Lisini]

Seminari di Matematica Applicata, Dipartimento di Matematica "F. Casorati"
e Istituto del CNR IMATI "E. Magenes" di Pavia.

Mercoledì 19 aprile 2023, alle ore 16.15 precise, presso l'aula C8 di
Ingegneria,

Axel Parmentier (Ecole de Ponts de Paris)

terrà un seminario dal titolo:

Learning with combinatorial optimization layers and applications to dynamic
vehicle routing.

Il seminario verrà anche trasmesso in diretta su zoom.

Link Zoom:
https://us02web.zoom.us/j/83344185446?pwd=RHRGai91RkZQTjg0eEVRMWQ5WXFjZz09

Abstract.

Combinatorial optimization (CO) layers in machine learning (ML) pipelines
are a powerful tool to tackle data-driven decision tasks, but they come
with two main challenges. First, the solution of a CO problem often behaves
as a piecewise constant function of its objective parameters. Given that ML
pipelines are typically trained using stochastic gradient descent, the
absence of slope information is very detrimental. Second, standard ML
losses do not work well in combinatorial settings. A growing body of
research addresses these challenges through diverse methods. Unfortunately,
the lack of well-maintained implementations slows down the adoption of CO
layers. Building upon previous works, we introduce a probabilistic
perspective on CO layers, which lends itself naturally to approximate
differentiation and the construction of structured losses. We recover many
approaches from the literature as special cases, and we also derive new
ones. Based on this unifying perspective, we present InferOpt.jl, an
open-source Julia package that 1) allows turning any CO oracle with a
linear objective into a differentiable layer, and 2) defines adequate
losses to train pipelines containing such layers. Our library works with
arbitrary optimization algorithms, and it is fully compatible with Julia’s
ML ecosystem. In the second part of the talk, we focus on the dynamic
vehicle routing problem of the 2022 EURO-NeurIPS challenge (1). Using a CO
layer in a deep learning pipeline enabled to win the challenge. We focus on
the structure of the pipeline used as a policy, and on the algorithm used
to train it, which are natural applications of the probabilistic
perspective introduced during the first part of the talk.
(1) https://euro-neurips-vrp-2022.challenges.ortec.com/

---------------------
Pagina web dei Seminari di Matematica Applicata
https://matematica.unipv.it/ricerca/cicli-di-seminari/seminari-di-matematica-applicata/
-------------- parte successiva --------------
Un allegato HTML è stato rimosso...
URL: http://ipv01.unipv.it/pipermail/dottorcomp/attachments/20230413/9143dc5a/attachment.htm