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:320@isdm.umontpellier.fr
DTSTART;TZID=Europe/Paris:20230915T140000
DTEND;TZID=Europe/Paris:20230915T140000
DTSTAMP:20260828T083010Z
URL:https://isdm.umontpellier.fr/events/how-deep-learning-can-facilitate-t
 he-extraction-of-information-from-remote-sensing-data/
SUMMARY:How Deep learning can facilitate the extraction of information from
  remote sensing data
DESCRIPTION:Maison de la Télédétection\n\nMachine Learning in Montpellie
 r\, Theory &amp\; Practice\n\nRemote sensing images\, produced in large qu
 antities\, contain information that is already being used to monitor clima
 te change\, improve security and understand the environment. However\, thi
 s data is difficult to interpret and often requires manual processing. As 
 the amount of data increases\, interpretation becomes a limiting factor no
 t only in terms of time\, but also in terms of the areas in which these im
 ages can be used. While the data is there\, a wide audience cannot take ad
 vantage of it. In this presentation\, we detail recent work on automatic r
 esponse to visual questions. The aim of this work is to provide an answer 
 in natural language to a question (also formulated in natural language) re
 lating to remote sensing images\, opening up the extraction of information
  from these images to as many people as possible. 10h30 - 11h\, Konstantin
 os Panousis\, INRIA Post-doc Title: Sparse Linear Concept Discovery Models
  Abstract: The recent mass adoption of DNNs\, even in safety-critical scen
 arios\, has shifted the focus of the research community towards the creati
 on of inherently intrepretable models. Concept Bottleneck Models (CBMs) co
 nstitute a popular approach where hidden layers are tied to human understa
 ndable concepts allowing for investigation and correction of the network&a
 pos\;s decisions. However\, CBMs however usually suffer from: (i) performa
 nce degradation and (ii) lower interpretability than intended due to the s
 heer amount of concepts contributing to each decision. In this work\, we p
 ropose a simple yet highly intuitive interpretable framework based on Cont
 rastive Language Image models and a single sparse linear layer. In stark c
 ontrast to related approaches\, the sparsity in our framework is achieved 
 via principled Bayesian arguments by inferring concept presence via a data
 -driven Bernoulli distribution. As we experimentally show\, our framework 
 not only outperforms recent CBM approaches accuracy-wise\, but it also yie
 lds high per example concept sparsity\, facilitating the individual invest
 igation of the emerging concepts. 11h - 11h15 PAUSE 11h15 - 11h45\, Diego 
 Marcos\, INRIA CPJ Title: PDiscoNet: Semantically consistent part discover
 y for fine-grained recognition Abstract: Fine-grained classification often
  requires recognizing specific object parts\, such as beak shape and wing 
 patterns for birds. Encouraging a fine-grained classification model to fir
 st detect such parts and then using them to infer the class could help us 
 gauge whether the model is indeed looking at the right details better than
  with interpretability methods that provide a single attribution map. We p
 ropose PDiscoNet to discover object parts by using only image-level class 
 labels along with priors encouraging the parts to be: discriminative\, com
 pact\, distinct from each other\, equivariant to rigid transforms\, and ac
 tive in at least some of the images. In addition to using the appropriate 
 losses to encode these priors\, we propose to use part-dropout\, where ful
 l part feature vectors are dropped at once to prevent a single part from d
 ominating in the classification\, and part feature vector modulation\, whi
 ch makes the information coming from each part distinct from the perspecti
 ve of the classifier. Our results on CUB\, CelebA\, and PartImageNet show 
 that the proposed method provides substantially better part discovery perf
 ormance than previous methods while not requiring any additional hyper-par
 ameter tuning and without penalizing the classification performance. 11h45
  - 12h15 Ananthu Aniraj - INRIA PhD student Title: Masking Strategies for 
 Background Bias Removal in Computer Vision Models Abstract: Models for fin
 e-grained image classification tasks\, where the difference between some c
 lasses can be extremely subtle and the number of samples per class tends t
 o be low\, are particularly prone to picking up background-related biases 
 and demand robust methods to handle potential examples with out-of-distrib
 ution (OOD) backgrounds. To gain deeper insights into this critical proble
 m\, our research investigates the impact of background-induced bias on fin
 e-grained image classification\, evaluating standard backbone models such 
 as Convolutional Neural Network (CNN) and Vision Transformers (ViT). We ex
 plore two masking strategies to mitigate background-induced bias: Early ma
 sking\, which removes background information at the (input) image level\, 
 and late masking\, which selectively masks high-level spatial features cor
 responding to the background. Extensive experiments assess the behaviour o
 f CNN and ViT models under different masking strategies\, with a focus on 
 their generalization to OOD backgrounds. The obtained findings demonstrate
  that both proposed strategies enhance OOD performance compared to the bas
 eline models\, with early masking consistently exhibiting the best OOD per
 formance. Notably\, a ViT variant employing GAP-Pooled Patch token-based c
 lassification combined with early masking achieves the highest OOD robustn
 ess. Best\, Cássio F. Dantas INRAE\, TETIS [ https://isdm.umontpellier.fr
 /cassiofragadantas.github.io/ | cassiofragadantas.github.io ]\n\nMachine L
 earning in Montpellier\, Theory &amp\; Practice
ATTACH;FMTTYPE=image/jpeg:https://isdm.umontpellier.fr/wp-content/uploads/
 2026/06/ml-mtp-gC78d5.png
CATEGORIES:ML MTP
LOCATION:Maison de la Télédétection\, 500 rue Jean-François Breton\, Mo
 ntpellier\, 
X-APPLE-STRUCTURED-LOCATION;VALUE=URI;X-ADDRESS=500 rue Jean-François Bret
 on\, Montpellier\, ;X-APPLE-RADIUS=100;X-TITLE=Maison de la Télédétecti
 on: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