BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//wp-events-plugin.com//7.4.0.1//EN
TZID:Europe/Paris
X-WR-TIMEZONE:Europe/Paris
BEGIN:VEVENT
UID:194@isdm.umontpellier.fr
DTSTART;TZID=Europe/Paris:20240321T140000
DTEND;TZID=Europe/Paris:20240321T140000
DTSTAMP:20260825T131416Z
URL:https://isdm.umontpellier.fr/events/deep-learning-under-lipschitz-cons
 traints/
SUMMARY:Deep learning under Lipschitz constraints
DESCRIPTION:Room 03.124\, Building 5\, St Priest campus\n\nMachine Learning
  in Montpellier\, Theory &amp\; Practice\n\nThis talk is about Lipschitz c
 onstraints in the context of deep learning. Lipschitz constrained neural n
 etworks are of huge interest in the context of optimal transport (through 
 Kantorovich-Rubinstein duality) and certifiability against adversarial att
 acks. First\, I will explain how to implement these constraints in practic
 e by performing a brief literature reviews of the dominant aproaches and m
 ethods. I will also present the DEEL-LIP library that contains some of the
 se implementations in a Python package. Then\, I will present some of my c
 ontributions to this field. I will show that these class of functions can 
 be used to solve any classification task\, with robustness guarantees\, ge
 neralization guarantees (some of which being architecture-independant)\, a
 nd I will emphasize the importance of entropic regularization to chose an 
 accuracy/robustness tradeoff on a Pareto front. After\, I will present the
  explainability properties of these networks when they are trained with th
 e Hinge-Kantorovich-Rubinstein (HKR) loss\, inspired by optimal transport.
  Their saliency maps are more aligned with humans perception than any othe
 r network\, including on challenging datasets like ImageNet. The HKR loss 
 and Lipschitz neural networks can also be used to compute Signed Distance 
 Function (SDF) in an unsupervised manner\, relieving the need for a ground
  truth based on nearest neighbor. The resulting algorithm can be seen as a
  parametric K-nearest neighbor\, or a robust one-class classifier. Further
 more\, the SDF is also of huge interest in the computer graphics community
  since it allow for implicit surface parametrization with formal guarantee
 s during raytracing. Finally\, I will show how Lipschitz constraints w.r.t
  the input can be converted easily into Lipschitz constraints w.r.t the pa
 rameters\, using a "backpropagation for bounds" algorithm\, which opens pa
 th for deep learning with privacy guarantees without the need for the expe
 nsive gradient clipping operation of DP-SGD.\n\nMachine Learning in Montpe
 llier\, Theory &amp\; Practice
ATTACH;FMTTYPE=image/jpeg:https://isdm.umontpellier.fr/wp-content/uploads/
 2026/06/ml-mtp-gC78d5.png
CATEGORIES:ML MTP
LOCATION:St Priest campus - Building 5 - Room 03.124\, 860 rue St Priest\, 
 Montpellier\, 
X-APPLE-STRUCTURED-LOCATION;VALUE=URI;X-ADDRESS=860 rue St Priest\, Montpel
 lier\, ;X-APPLE-RADIUS=100;X-TITLE=St Priest campus - Building 5 - Room 03
 .124:geo:0,0
END:VEVENT
BEGIN:VTIMEZONE
TZID:Europe/Paris
X-LIC-LOCATION:Europe/Paris
BEGIN:STANDARD
DTSTART:20231029T020000
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
END:STANDARD
END:VTIMEZONE
END:VCALENDAR