Offre en lien avec l’Action/le Réseau : – — –/Doctorants

Laboratoire/Entreprise : FEMTO-ST
Durée : 3years
Contact : noura.dridi@ens2m.fr
Date limite de publication : 2023-06-30

Contexte :
Most of the real physical system and everyday situations include
uncertainty. This is the case for medical diagnosis, weather forecasting, evolution of the stock market and so on. In the literature two types of uncertainty are distinguished:
aleatoric uncertainty denotes the one that is inherent to the data, e.g., noise in measurements or natural variability of the inputs, and epistemic uncertainty related to the model and due to lack of knowledge. Measuring the uncertainty is important, so as to support the user in the action to take. For example, when an anomaly is detected, with weak confidence level, another source of information should be added (image, human intervention, etc.) before planning intervention actions. More generally, quantification of the prediction uncertainty allows to trust or not predictions. In fact, incorrect overconfident predictions can be harmful and lead to erroneous decision.

Sujet :
Goal of the thesis: The goal of this thesis is to develop a robust method to evaluate uncertainty for machine and deep learning algorithm predictions. Major of works focused on improving the algorithm performance, few works deal with measuring the uncertainty
related to the predictions. In particular in this thesis we want to relax some hypothesis in the existing approach related to the distribution of the data and symmetry of the algorithm. This subject is challenging with many theoretical and applicatives difficulties. It is multidisciplinary including competences in probability, statistic and data processing. The
two principal goal are:
-First, we aim to measure the impact of uncertainty miss evaluation on the decision.
-The second part is focused on developing new method to quantify uncertainty, that can be applied to different type of data and without restrictive constraint on distribution or the exchangeability.
The third part, includes generalization of the proposed method when we have noisy and/or missing data.
The second part include study of the theoretical aspects: proof of convergence, complexity issue. In addition to practical aspects: independence from the chosen algorithm, architecture of the NN, implementation… Finally, a validation criterion is defined to attest
the performance of the uncertainty measure.

Profil du candidat :
Master in applied mathematics (or equivalent). Probability, statistic.
Good skills in Python programming. Experience in machine learning/deep learning

Formation et compétences requises :
Master in applied mathematics (or equivalent: engineering school diploma)

Adresse d’emploi :
FEMTO-ST
15B avenue des Montboucons
25030 Besançon cedex France

Document attaché : 202304261402_ThesisOfferFEMTO.pdf