Unsupervised domain adaptation with application to urban (...) - Pérez - Workshop 3 - CEB T1 2019
Patrick Pérez (Valeo) / 04.04.2019
nsupervised domain adaptation with application to urban scene analysis.
In numerous real world applications, no matter how much energy is devoted to build real and/or synthetic training datasets, there remains a large distribution gap between these data and those met at run-time. This gap results in severe, possibly catastrophic, performance loss. This problem is especially acute for automated and autonomous driving systems, where generalizing well to diverse testing environments remains a major challenge. One promising tool to mitigate this issue it unsupervised domain adaptation (UDA), which assumes that un-annotated data from the "test domain" are available at training time, along with the annotated data from the "source domain". We will discuss different ways to approach UDA, with application to semantic segmentation and object detection in urban scenes. We will introduce a new approach, called AdvEnt, that relies on combining adversarial training with minimization of decision entropy (seen as a proxy for uncertainty).
----------------------------------
Vous pouvez nous rejoindre sur les réseaux sociaux pour suivre nos actualités.
Facebook : https://www.facebook.com/InstitutHenriPoincare/
Twitter : https://twitter.com/InHenriPoincare
Instagram : https://www.instagram.com/instituthenripoincare/
*************************************
Langue : Anglais; Date : 04.04.2019; Conférencier : Pérez, Patrick; Évenement : Workshop 3 - CEB T1 2019; Lieu : IHP; Mots Clés :
Видео Unsupervised domain adaptation with application to urban (...) - Pérez - Workshop 3 - CEB T1 2019 канала Institut Henri Poincaré
nsupervised domain adaptation with application to urban scene analysis.
In numerous real world applications, no matter how much energy is devoted to build real and/or synthetic training datasets, there remains a large distribution gap between these data and those met at run-time. This gap results in severe, possibly catastrophic, performance loss. This problem is especially acute for automated and autonomous driving systems, where generalizing well to diverse testing environments remains a major challenge. One promising tool to mitigate this issue it unsupervised domain adaptation (UDA), which assumes that un-annotated data from the "test domain" are available at training time, along with the annotated data from the "source domain". We will discuss different ways to approach UDA, with application to semantic segmentation and object detection in urban scenes. We will introduce a new approach, called AdvEnt, that relies on combining adversarial training with minimization of decision entropy (seen as a proxy for uncertainty).
----------------------------------
Vous pouvez nous rejoindre sur les réseaux sociaux pour suivre nos actualités.
Facebook : https://www.facebook.com/InstitutHenriPoincare/
Twitter : https://twitter.com/InHenriPoincare
Instagram : https://www.instagram.com/instituthenripoincare/
*************************************
Langue : Anglais; Date : 04.04.2019; Conférencier : Pérez, Patrick; Évenement : Workshop 3 - CEB T1 2019; Lieu : IHP; Mots Clés :
Видео Unsupervised domain adaptation with application to urban (...) - Pérez - Workshop 3 - CEB T1 2019 канала Institut Henri Poincaré
Показать
Комментарии отсутствуют
Информация о видео
Другие видео канала
Tropical motivic integrationMaking sense of cross-scale dynamics... - Ringsmuth - Workshop 3 - CEB T3 2019Un petit aperçu des matrices de WignerPartitions, Modular Forms and Moduli Spaces - 4/8SHM 18/10/2019 - Comment une approche émique des textes mathématiques... - ProustWistan Marchadour - Réseaux de neurones et application en OncologieWhat is macroscopic quantum information?Jam Maths & SantéIsoperimetric inequalities in high dimensional convex sets (Lecture 1 - Part 1)Promises and challenges of Deep Learning in Cosmology - Lanusse - Workshop 2 - CEB T3 2018Planetary environmental boundaries and their modelling - Gerten - Workshop 3 - CEB T3 2019Nonlinear and Stochastic methods in climate and GFD- Takao - Workshop 1 - CEB T3 2019SHM - 16/01/15 - Constructivismes en mathématiques - Frédéric BrechenmacherEstimating with continuous quantum measurements20/10/2015 - Sergiu Klainerman - Lecture 2 - On The Mathematical Theory of Black HolesSeize the Moments: Enhancing Moment Estimation for Subdiffraction Incoherent ImagingAsymptotic and Probabilistic Aspects of Representation Theory - 1/8Journées Hénon - 19/21 - Pierre HénonThe vacuum-gap transmon qubit...Sochastic process models for precipitation processes... - Neelin - Workshop 1 - CEB T3 2019La recherche mathématique se prend au jeu, du 24 au 27 mai 2018, Place Saint Sulpice