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:329@isdm.umontpellier.fr DTSTART;TZID=Europe/Paris:20230928T140000 DTEND;TZID=Europe/Paris:20230928T140000 DTSTAMP:20260828T083151Z URL:https://isdm.umontpellier.fr/events/spotting-expressivity-bottlenecks- and-fixing-them-optimally/ SUMMARY:Spotting Expressivity Bottlenecks and Fixing Them Optimally DESCRIPTION:Camelot Workshop\, St Priest campus\n\nMachine Learning in Mont pellier\, Theory &\; Practice\n\nReject option is a well established te chnique to abstain from prediction when the doubt in the label is too big or because of ressources limits. In this talk I will first present the pro blem of learning with reject option in its more classical use. I will then present two recent applications: the first one focuses on building an eff icient active learning algorithm that uses rejection option learning argum ents to iteratively construct the uncertain region\; the second one deals with reducing energy consumption during inference with a deep neural netwo rk. In the latter application\, we have built networks with early exits\, and it appears that the reject option is particularly useful for calibrati ng thresholds for early existing. 3:30 PM - 4:00 PM *Coffee Break (2nd flo or) 4 PM - 4:45 PM *Mathilde Mougeot\, ENSIIE &\; ENS Paris-Saclay* Tit le: *Leveraging knowledge to design machine learning despite the lack of d ata. Abstract: In recent years\, considerable progress has been made in th e implementation of decision support procedures based on machine learning methods through the exploitation of very large databases and the use of le arning algorithms. In the industrial environment\, the databases available in research and development or in production are rarely so voluminous and the question arises as to whether in this context it is reasonable to use machine learning methods. This talk presents research work around transfe r learning and hybrid models that use knowledge from related application d omains or physics to implement efficient models with an economy of data. S everal achievements in industrial collaborations will be presented that su ccessfully use these learning models to design machine learning for indust rial small data regimes and to develop powerful decision support tools eve n in cases where the initial data volume is limited. References - de Mathe lin\, A.\, Deheeger\, F.\, Mougeot\, M.\, Vayatis\, N. (2023) From Theoret ical to Practical Transfer Learning: the ADAPT library\, Federated and Tra nsfer Learning Springer book. - Nguyen\, Khoa.T.N\, Dairay\, T.\, Meunier\ , R.\, Mougeot\, M. (2023) Fixed-Budget Online Adaptive Learning for Physi cs-Informed Neural Networks. Towards Parameterized Problem Inference\, Lec ture Notes in Computer Science\, Springer volume 14073. 4:45 PM - 5:30 PM *Stephane Chretien\, University of Lyon 2* Title: *Relationship between sa mple size and architecture for the estimation of Sobolev functions using d eep neural networks* Abstract: Beyond the many successes of Deep Learning based techniques in various branches of data analytics\, medicine\, busine ss\, engineering and the human sciences\, a sound understanding of the gen eralisation properties of these techniques is still elusive. Central to th ese successes are the availability of huge datasets and the availability o f huge computational ressources and some of the most recent trends have gi ven paramount importance to the necessity of building huge neural networks with millions of parameters\, and most often\, of several orders of magni tude larger than the size of the training set. This set-up has however led to many surprises and counterintuitive discoveries. Overprametrisation wa s recently shown to favour connectivity in a weak sense of the set of stat ionary points\, hence permitting stochastic gradient type methods to poten tially reach good minimisers in several stages despite the wild nonconvexi ty of the training problem as demonstrated by Kuditipudi et al. Relating g eneralisation to stability\, recent theoretical breakthroughs have been ab le to provide a better understanding of why generalisation cannot even hap pen without overparametrisation as shown by Bubeck et al. Following the id eas developed by Belkin\, a substantial amount of work has also been under taken in order to study the double descent phenomenon\, and the associated benign overfitting property which holds for least norm estimators in line ar and mildly non-linear regression\, as well as in for certain kernel bas ed methods. In the present paper\, we aim at studying the generalisation p roperties of overparametrised deep neural networks using a novel approach based on Neuberger’s theorem.\n\nMachine Learning in Montpellier\, Theor y &\; Practice ATTACH;FMTTYPE=image/jpeg:https://isdm.umontpellier.fr/wp-content/uploads/ 2026/06/ml-mtp-gC78d5.png CATEGORIES:ML MTP LOCATION:Saint Priest Campus\, Camelot Workshop\, 860 rue St Priest\, Montp ellier\, X-APPLE-STRUCTURED-LOCATION;VALUE=URI;X-ADDRESS=860 rue St Priest\, Montpel lier\, ;X-APPLE-RADIUS=100;X-TITLE=Saint Priest Campus\, Camelot Workshop: geo:0,0 END:VEVENT BEGIN:VTIMEZONE TZID:Europe/Paris X-LIC-LOCATION:Europe/Paris BEGIN:DAYLIGHT DTSTART:20230326T030000 TZOFFSETFROM:+0100 TZOFFSETTO:+0200 TZNAME:CEST END:DAYLIGHT END:VTIMEZONE END:VCALENDAR