École Polytechnique Fédérale de Lausanne
Infoscience - École polytechnique fédérale de LausanneNot a member yet
191401 research outputs found
Sort by
Déconstruire l’habitat bourgeois : transformation de deux villas urbaines en logement collectif
No full textRenouveler la ville de l’intérieur. Dans la continuité d’une réflexion féministe exprimée dans l’énoncé théorique Vers une architecture rabat-joie et visant à décortiquer la façon dont l’idéologie patriarcale et capitaliste influence notre conception de l’espace domestique, de la famille et du genre, le projet remet en cause la dimension sociale et spatiale d’un modèle domestique hérité de la bourgeoisie du XIXe siècle : les villas urbaines lausannoises. Il questionne le mythe de la famille traditionnelle en tant que norme sociale où la cellule familiale est utilisée comme instrument de pouvoir divisant les classes, les genres, les races et les âges. Un modèle social et spatial qui continue aujourd’hui d’influencer la manière dont la ville et l’habitat sont (re)produits. Les normes soutiennent des langages et des formes génériques. Lorsque celles-là sont troublées, il devient possible d’imaginer des alternatives qui s’en affranchissent. Le projet reconnaît le potentiel de renouvellement, d’adaptation et de transformation du patrimoine bâti afin de répondre à l’évolution des modes de vie de la société contemporaine. La dé-construction des deux villas urbaines expérimente une domesticité flexible et singulière, délaissant le caractère individuel au profit du collectif, brouillant les limites entre privé et public, permettant une liberté d’usage et d’occupation selon les rythmes et les besoins de chacun·e·x. Il s’agit de faire maison et faire famille autrement.ALICEENAC-SARCote: 2024.092MEM.1/1Groupe de suivi: Dietz, Dieter (dir. pédagogique) ; Pedrazzini, Yves (prof.) ; Valdez Juarez, Ruben (maître EPFL) ; Eliasson, Susanne (expert)Professeur responsable de l'Enoncé: Pedrazzini, Yves (ENAC IA LASUR)Enoncé théorique de master: Vers une architecture rabat-joi
Leveraging Collaborative Digital Platforms for Supervising Project-Based Learning Activities: A Case Study
No full textIn contemporary engineering education, project-based learning (PBL) has emerged as a prominent pedagogical approach to foster transversal skills like collaboration and communication. However, effective supervision and guidance of student groups in PBL present challenges. This paper introduces an innovative approach to support student groups and teaching assistants (TAs) in PBL, utilizing a digital experience platform. The presented pedagogical scenario features a digital collaboration journal used by students to document their progress. Through the implementation of this scenario in an interdisciplinary course, data was collected via surveys and interviews. Results indicate that students perceived the collaboration journal as supportive for poster preparation and discussions with the TAs. While the scenario was well received, challenges persist in fostering online communication. Future iterations aim to enhance the platform's role in facilitating group-TA interactions and providing a comprehensive collaborative environment. This research contributes to the advancement of PBL methodologies by offering practical insights into leveraging digital platforms to enhance collaboration, communication, and feedback processes, ultimately enriching the learning experience for students and TAs alike.SCI-STI-DGAVP-E-LEAR
Large-scale georeferenced neuroimaging and psychometry data link the urban environmental exposome with brain health
No full textIn face of cumulating evidence about the impact of human-induced environmental changes on mental health and behavior, our understanding of the main effects and interactions between environmental factors – i.e., the exposome and the brain – is still limited. We seek to fill this knowledge gap by leveraging georeferenced large-scale brain imaging and psychometry data from the adult community-dwelling population (n = 2672; mean age 63 ± 10 years). For monitoring brain anatomy, we extract morphometry features from a nested subset of the cohort (n = 944) with magnetic resonance imaging. Using an iterative analytical strategy testing the moderator role of geospatially encoded exposome factors on the association between brain anatomy and psychometry, we demonstrate that individuals' anxiety state and psychosocial functioning are among the mental health characteristics showing associations with the urban exposome. The clusters of higher anxiety state and lower current psychosocial functioning coincide spatially with a lower vegetation density and higher air pollution. The univariate multiscale geographically weighted regression identifies the spatial scale of associations between individuals’ levels of anxiety state, psychosocial functioning, and overall cognition with vegetation density, air pollution and structures of the limbic network. Moreover, the multiscale geographically weighted regression interaction model reveals spatially confined exposome features with moderating effect on the brain-psychometry/cognitive performance relationships. Our original findings testing the role of exposome factors on brain and behavior at the individual level, underscore the role of environmental and spatial context in moderating brain-behavior dynamics across the adult lifespan.LG
Computational complexity of deep learning: fundamental limitations and empirical phenomena
No full textThis manuscript is the lecture notes of B. Barak’s course in the Les Houches ‘Statistical Physics and Machine Learning’ summer school in 2022. It surveys various proxies for computational hardness in random planted problems, from the low-degree likelihood ratio to statistical query complexity and the Franz–Parisi criterion, as well as the various relationships between those criteria. We also present a few aspects of the study of deep learning, from both a theoretical and empirical point of view.PCSLIDEPHICS
Ensuring Solution Uniqueness in Fixed-Point-Based Harmonic Power Flow Analysis with Converter-Interfaced Resources: Ex-post Conditions
No full textRecently, the authors of this paper proposed a method for the Harmonic Power-Flow (HPF) calculus in polyphase grids with widespread deployment of Converter-Interfaced Distributed Energy Resources (CIDERs). The HPF problem was formulated by integrating the hybrid nodal equations of the grid with a detailed representation of the CIDERs hardware, sensing, and controls as Linear Time-Periodic (LTP) systems, and solving the resulting mismatch equations using the Newton-Raphson (NR) method. This work introduces a novel problem formulation based on the fixed-point algorithm that, combined with the contraction property of the HPF problem, provides insights into the uniqueness of its solution. Notably, the effectiveness of the fixed-point formulation and the uniqueness of the solution are evaluated through numerical analyses conducted on a modified version of the CIGR É low-voltage benchmark microgrid.DES
Knowledge-Aware Text–Image Retrieval for Remote Sensing Images
No full textImage-based retrieval in large Earth observation archives is challenging because one needs to navigate across thousands of candidate matches only with the query image as a guide. By using text as information supporting the visual query, the retrieval system gains in usability, but at the same time faces difficulties due to the diversity of visual signals that cannot be summarized by a short caption only. For this reason, as a matching-based task, cross-modal text–image retrieval often suffers from information asymmetry between text and images. To address this challenge, we propose a Knowledge-aware Text–Image Retrieval (KTIR) method for remote sensing images. By mining relevant information from an external knowledge graph, KTIR enriches the text scope available in the search query and alleviates the information gaps between text and images for better matching. Moreover, by integrating domain-specific knowledge, KTIR also enhances the adaptation of pretrained vision–language models to remote sensing applications. Experimental results on three commonly used remote sensing text–image retrieval benchmarks show that the proposed knowledge-aware method leads to varied and consistent retrievals, outperforming state-of-the-art retrieval methods.ECE
What can be learnt with wide convolutional neural networks?
No full textUnderstanding how convolutional neural networks (CNNs) can efficiently learn high-dimensional functions remains a fundamental challenge. A popular belief is that these models harness the local and hierarchical structure of natural data such as images. Yet, we lack a quantitative understanding of how such structure affects performance, for example the rate of decay of the generalisation error with the number of training samples. In this paper, we study infinitely wide deep CNNs in the kernel regime. First, we show that the spectrum of the corresponding kernel inherits the hierarchical structure of the network, and we characterise its asymptotics. Then, we use this result together with generalisation bounds to prove that deep CNNs adapt to the spatial scale of the target function. In particular, we find that if the target function depends on low-dimensional subsets of adjacent input variables then the decay of the error is controlled by the effective dimensionality of these subsets. Conversely, if the target function depends on the full set of input variables then the error decay is controlled by the input dimension. We conclude by computing the generalisation error of a deep CNN trained on the output of another deep CNN with randomly initialised parameters. Interestingly, we find that, despite their hierarchical structure, the functions generated by infinitely wide deep CNNs are too rich to be efficiently learnable in high dimensions.PCS
Anion Exchange Membrane Water Electrolysis at 10 A ⋅ cm<sup>−2</sup> Over 800 Hours
No full textAnion exchange membrane water electrolyzer (AEMWE) is a potentially cost-effective technology for green hydrogen production. Although the normal current densities of AEMWEs are below 3 A ⋅ cm−2, operating them at higher current densities represents an efficient, but little-explored approach to decrease the total cost of hydrogen production. We show here that a benchmark AEMWE has an operational lifetime of only seconds at an ultrahigh current density of 10 A ⋅ cm−2. By using a more conductive and robust AEM, and judicious choices of ionomers, catalyst, and porous transport layer, we have developed AEMWEs that stably operate at 10 A ⋅ cm−2 with extended lifetimes. The optimized AEMWE has an operational lifetime of more than 800 hours, a 5-order magnetite improvement over the current benchmark. The cell voltage is only 2.3 V at 10 A ⋅ cm−2, comparable to those of the state-of-the-art devices operating at current densities lower than 3 A ⋅ cm−2. This work demonstrates the potential of ultrahigh current density AEMWEs.LSC
The Digital Twin of the Panorama of the Battle of Murten: Notes on the Creation of the World’s Largest Image307
No full textThe Panorama of the Battle of Murten (Louis Braun, 1893/1894) is the only Swiss historical panorama not accessible to the public. As part of the celebration of the 550th anniversary of the Burgundian Wars in 2026, the Foundation for the Panorama of the Battle of Murten and the Laboratory for Experimental Museology at the Swiss Federal Institute of Technology have undertaken a project to digitize and valorize the original painting. The digitization phase has now been completed, producing the largest digital image of an artwork (1,600 gigapixels). This report details this achievement from a technical point of view, and highlights the new strategies for valorization offered by the digital twin. The project is part of an exploratory approach, going beyond standards in the cultural industries and enabling advances in data science research, but also in new museology. The new possibilities offered by ultra-highresolution images concern the fields of conservation, but above all the new paradigms of access to cultural heritage dealing with scientific imaging, large-scale visualization, humanities and public history.EMPLU
Réapprendre le 10-10 : son territoire, son école, ses habitants
No full textLe « 1010 » est un quartier populaire au nord-est de la ville de Lausanne. Construit dans les années 1950 avec l’avènement de l’automobile, ce territoire est divisé par l’autoroute et la route de Berne, fragmentant l’espace de la vie quotidienne. À l’ouest de cette banlieue se trouve l’établissement primaire et secondaire de Grand-Vennes, construit en 1974 sous le système de construction CROCS. Il a été conçu comme un centre social du quartier avec un programme utile pour l’ensemble de ses habitant·e·s. Dès 1988, l’espace du collège a souffert d’un manque de salles de classe. En réponse, les collectivités publiques ont remplacé les infrastructures communes par de nouvelles salles et ont ajouté des containers préfabriqués. En septembre 2024, de nouveaux containers temporaires viendront s’ajouter aux précédents dans le préau. Ils sont accompagnés d’une promesse de la municipalité : elle prévoit un concours d’architecture et 47 millions de CHF pour un projet d’ «agrandissement des surfaces d’enseignement, gymniques et d’accueil parascolaire». C’est en collectif que nous choisissons de sortir de notre école pour aller à la rencontre de ce quartier, de ce collège et apprendre de leurs habitant·e·s. C’est par un processus participatif que nous narrons de nouveaux possibles pour le « 1010 ». Le projet vise à la décentralisation des lieux d’enseignement, à la requalification des communs existants et à la réalisation d’un nouvel axe de mobilité piétonne, afin de repenser le collège comme un repère collectif pour Grand-Vennes. En collaboration avec AZGRAM.ALICEENAC-SARCote: 2024.019MEM.1/2 , MEM.2/2Groupe de suivi: Dietz, Dieter (dir. pédagogique) ; Fivet, Corentin (prof.) ; Valdez Juarez, Ruben (maître EPFL) ; Zahnd, Marion (expert)Professeur responsable de l'Enoncé: Fivet, Corentin (ENAC IA SXL)Enoncé théorique de master: Faire, Défaire, Refaire [Titre énoncé A. Clairac] Produire l’architecte EPFL: continuité, ruptures, (trans)formations [Titre énoncé A. Riand