pr\u00e9sentent<\/p>\n\n\n\n
Autoris\u00e9 \u00e0 pr\u00e9senter ses travaux en vue de l\u2019obtention du Doctorat de l’Institut Polytechnique de Paris, pr\u00e9par\u00e9 \u00e0 l’Institut Polytechnique de Paris T\u00e9l\u00e9com SudParis en :<\/p>\n\n\n\n
le MARDI 26 MAI 2026 \u00e0 14h00<\/p>\n\n\n\n
\u00e0<\/p>\n\n\n\n
Amphith\u00e9\u00e2tre 2
Telecom SudParis 19 place Marguerite Perey 91120 Palaiseau<\/p>\n\n\n\n
Membres du jury :<\/strong><\/p>\n\n\n\n M. Hossam AFIFI<\/strong>, Full professor, Institut Polytechnique de Paris T\u00e9l\u00e9com SudParis, FRANCE – Directeur de these R\u00e9sum\u00e9 :<\/strong><\/p>\n\n\n\n La prolif\u00e9ration des capteurs g\u00e9olocalis\u00e9s et donn\u00e9es mobiles rend l’apprentissage profond spatiotemporel central pour comprendre les dynamiques en r\u00e9seau. Toutefois, extraire des informations exploitables exige de surmonter des d\u00e9fis en d\u00e9couverte de motifs, mod\u00e9lisation structurelle et robustesse statistique. Cette th\u00e8se pr\u00e9sente un cadre coh\u00e9rent faisant progresser l’\u00e9tat de l’art dans trois domaines critiques : clustering de graphes spatiotemporels, pr\u00e9diction topologique d’ordre sup\u00e9rieur et att\u00e9nuation du d\u00e9calage de distribution. Premi\u00e8rement, nous abordons le clustering de graphes spatiotemporels et l’analyse de donn\u00e9es mobiles. Les m\u00e9thodes traditionnelles traitent les s\u00e9ries temporelles ind\u00e9pendamment, ignorant les relations spatiales et produisant des r\u00e9sultats fragment\u00e9s. Nous proposons des m\u00e9thodes d\u00e9couvrant les structures urbaines latentes au-del\u00e0 de l’analyse purement temporelle : 1) l’identification de signatures sociales via le clustering spatiotemporel de consommation mobile, r\u00e9v\u00e9lant diverses fonctions urbaines (r\u00e9sidentiel, affaires, loisirs), et 2) un clustering urbain \u00e0 grille fine mod\u00e9lisant conjointement les d\u00e9pendances spatiotemporelles profondes via des r\u00e9seaux convolutifs et r\u00e9currents. Ces contributions aident urbanistes et op\u00e9rateurs \u00e0 optimiser les ressources selon les dynamiques r\u00e9elles. La validation sur des m\u00e9tadonn\u00e9es mobiles m\u00e9tropolitaines prouve que notre approche capture des motifs \u00e0 de multiples \u00e9chelles, des vastes r\u00e9gions aux micro-dynamiques invisibles aux r\u00e9solutions classiques. Deuxi\u00e8mement, nous traitons les limites structurelles des mod\u00e8les de graphes en pr\u00e9vision et imputation spatiotemporelles. Les m\u00e9thodes actuelles encodent les relations spatiales via des graphes, limit\u00e9s aux connexions par paires. Cette perspective est restrictive, de nombreux ph\u00e9nom\u00e8nes urbains impliquant des interactions collectives d’ordre sup\u00e9rieur, ind\u00e9composables en simples paires. Pour y rem\u00e9dier, cette th\u00e8se introduit les complexes simpliciaux comme cadre topologique repr\u00e9sentant ces relations complexes. Nous proposons un r\u00e9seau de neurones spatiotemporel simplicial conscient de la structure, \u00e9tendant le passage de messages aux simplexes de dimensions arbitraires (sommets, ar\u00eates, triangles). Cette architecture capture comportements synchronis\u00e9s et motifs r\u00e9gionaux via des m\u00e9canismes d’attention et un flux d’information bidirectionnel entre dimensions, am\u00e9liorant l’expressivit\u00e9 et la pr\u00e9cision pr\u00e9dictive. Les exp\u00e9riences sur des donn\u00e9es (trafic, environnement, t\u00e9l\u00e9communications) montrent des am\u00e9liorations substantielles face aux graphes classiques, particuli\u00e8rement pour les pr\u00e9visions \u00e0 long terme et donn\u00e9es manquantes. Troisi\u00e8mement, nous affrontons la fragilit\u00e9 statistique due au d\u00e9calage de distribution spatiotemporel, qui couple d\u00e9rive temporelle et h\u00e9t\u00e9rog\u00e9n\u00e9it\u00e9 spatiale. Les normalisations existantes ignorent la topologie spatiale, traitant chaque n\u0153ud ind\u00e9pendamment et \u00e9chouant en environnement dynamique. Nous d\u00e9veloppons un cadre bas\u00e9 sur des r\u00e9seaux de normalisation r\u00e9versibles pour d\u00e9tecter, caract\u00e9riser et s’adapter aux changements de distribution. Notre approche offre des transformations rigoureusement r\u00e9versibles et spatialement conscientes, garantissant la restauration des statistiques originales tout en assurant des pr\u00e9dictions robustes en milieu dynamique. En r\u00e9sum\u00e9, cette th\u00e8se pr\u00e9sente un corpus unifi\u00e9 allant de l’analyse fondamentale des donn\u00e9es mobiles \u00e0 la pr\u00e9vision robuste de r\u00e9seaux. En combinant des m\u00e9thodes innovantes (clustering, mod\u00e9lisation topologique, adaptation aux d\u00e9calages), nos travaux offrent de solides garanties th\u00e9oriques et des algorithmes pratiques acc\u00e9l\u00e9rant les applications des syst\u00e8mes en r\u00e9seau complexes. The proliferation of location-aware sensor networks and mobile data has made spatiotemporal deep learning central to understanding networked dynamics. However, extracting actionable insights requires overcoming fundamental challenges in pattern discovery, structural modeling, and statistical robustness. This thesis presents a cohesive framework advancing the state-of-the-art across three critical areas: Spatiotemporal Graph Clustering, Higher-Order Topological Prediction, and Distribution Shift Mitigation. First, we address Spatiotemporal Graph Clustering and Mobile Data Analytics. Traditional methods treat time series independently, ignoring spatial relationships and producing fragmented results. We propose novel methods to discover latent urban structures beyond temporal-only analysis. This includes: 1) identifying social signatures via spatiotemporal clustering of mobile service consumption to reveal distinct urban functions (e.g., residential, business, and entertainment zones), and 2) performing fine-grained urban grid clustering by jointly modeling deep spatiotemporal dependencies using convolutional and recurrent neural networks. These contributions enable urban planners and telecom operators to delineate functional zones and optimize resources based on actual user dynamics. Validation on metropolitan mobile metadata demonstrates our approach captures meaningful patterns at multiple scales, from coarse regions to micro-scale dynamics invisible at traditional resolutions. Second, we tackle the structural limitations of existing graph-based models in Spatiotemporal Forecasting and Imputation. Current methods encode spatial relationships through graphs, capturing only pairwise connections. This node-centric perspective is restrictive, as many urban phenomena involve collective, multi-way interactions that cannot be decomposed into pairwise components. To overcome this, this thesis introduces simplicial complexes as a topological framework for representing higher-order spatiotemporal relationships. We propose the Structure-Aware Simplicial Spatiotemporal Neural Network, which extends message-passing to simplices of arbitrary dimensions (vertices, edges, triangles, etc.). This architecture captures synchronized behaviors and regional patterns via specialized attention mechanisms and bidirectional information flow across simplex dimensions, enhancing expressive power and prediction accuracy. Experiments on traffic, environmental, and telecom datasets show substantial improvements over graph neural network baselines, particularly in long-term forecasting and extensive missing data scenarios. Third, we tackle statistical fragility caused by spatiotemporal distribution shifts, which couple temporal drift and spatial heterogeneity. Existing normalization methods process nodes independently, ignoring graph topology, and fail to ensure reliability in dynamic environments. We develop a comprehensive framework using reversible normalization networks to detect, characterize, and adapt to distribution changes across spatial and temporal scales. Our approach achieves rigorously reversible, spatially-aware transformations that restore original statistical properties while enabling robust predictions in dynamically shifting environments. In summary, this thesis presents a unified body of work progressing from foundational mobile data analytics to robust network forecasting and imputation. By combining innovative methods for graph clustering, topological modeling, and distribution shift adaptation, our contributions provide strong theoretical guarantees and practical algorithms to accelerate applications in complex networked systems.<\/p>\n","protected":false},"excerpt":{"rendered":" L’Ecole doctorale : Ecole Doctorale de l’Institut Polytechnique de Paris et le Laboratoire de recherche SAMOVAR – Services r\u00e9partis, Architectures, Mod\u00e9lisation, Validation, Administration des R\u00e9seaux pr\u00e9sentent l\u2019AVIS DE SOUTENANCE de Monsieur Zhaobo HU Autoris\u00e9 \u00e0 pr\u00e9senter ses travaux en vue de l\u2019obtention du Doctorat de l’Institut Polytechnique de Paris, pr\u00e9par\u00e9 \u00e0 l’Institut Polytechnique de Paris […]<\/p>\n","protected":false},"author":4,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"ocean_post_layout":"","ocean_both_sidebars_style":"","ocean_both_sidebars_content_width":0,"ocean_both_sidebars_sidebars_width":0,"ocean_sidebar":"","ocean_second_sidebar":"","ocean_disable_margins":"enable","ocean_add_body_class":"","ocean_shortcode_before_top_bar":"","ocean_shortcode_after_top_bar":"","ocean_shortcode_before_header":"","ocean_shortcode_after_header":"","ocean_has_shortcode":"","ocean_shortcode_after_title":"","ocean_shortcode_before_footer_widgets":"","ocean_shortcode_after_footer_widgets":"","ocean_shortcode_before_footer_bottom":"","ocean_shortcode_after_footer_bottom":"","ocean_display_top_bar":"default","ocean_display_header":"default","ocean_header_style":"","ocean_center_header_left_menu":"","ocean_custom_header_template":"","ocean_custom_logo":0,"ocean_custom_retina_logo":0,"ocean_custom_logo_max_width":0,"ocean_custom_logo_tablet_max_width":0,"ocean_custom_logo_mobile_max_width":0,"ocean_custom_logo_max_height":0,"ocean_custom_logo_tablet_max_height":0,"ocean_custom_logo_mobile_max_height":0,"ocean_header_custom_menu":"","ocean_menu_typo_font_family":"","ocean_menu_typo_font_subset":"","ocean_menu_typo_font_size":0,"ocean_menu_typo_font_size_tablet":0,"ocean_menu_typo_font_size_mobile":0,"ocean_menu_typo_font_size_unit":"px","ocean_menu_typo_font_weight":"","ocean_menu_typo_font_weight_tablet":"","ocean_menu_typo_font_weight_mobile":"","ocean_menu_typo_transform":"","ocean_menu_typo_transform_tablet":"","ocean_menu_typo_transform_mobile":"","ocean_menu_typo_line_height":0,"ocean_menu_typo_line_height_tablet":0,"ocean_menu_typo_line_height_mobile":0,"ocean_menu_typo_line_height_unit":"","ocean_menu_typo_spacing":0,"ocean_menu_typo_spacing_tablet":0,"ocean_menu_typo_spacing_mobile":0,"ocean_menu_typo_spacing_unit":"","ocean_menu_link_color":"","ocean_menu_link_color_hover":"","ocean_menu_link_color_active":"","ocean_menu_link_background":"","ocean_menu_link_hover_background":"","ocean_menu_link_active_background":"","ocean_menu_social_links_bg":"","ocean_menu_social_hover_links_bg":"","ocean_menu_social_links_color":"","ocean_menu_social_hover_links_color":"","ocean_disable_title":"default","ocean_disable_heading":"default","ocean_post_title":"","ocean_post_subheading":"","ocean_post_title_style":"","ocean_post_title_background_color":"","ocean_post_title_background":0,"ocean_post_title_bg_image_position":"","ocean_post_title_bg_image_attachment":"","ocean_post_title_bg_image_repeat":"","ocean_post_title_bg_image_size":"","ocean_post_title_height":0,"ocean_post_title_bg_overlay":0.5,"ocean_post_title_bg_overlay_color":"","ocean_disable_breadcrumbs":"default","ocean_breadcrumbs_color":"","ocean_breadcrumbs_separator_color":"","ocean_breadcrumbs_links_color":"","ocean_breadcrumbs_links_hover_color":"","ocean_display_footer_widgets":"default","ocean_display_footer_bottom":"default","ocean_custom_footer_template":"","ocean_post_oembed":"","ocean_post_self_hosted_media":"","ocean_post_video_embed":"","ocean_link_format":"","ocean_link_format_target":"self","ocean_quote_format":"","ocean_quote_format_link":"post","ocean_gallery_link_images":"on","ocean_gallery_id":[],"footnotes":""},"categories":[286,402],"tags":[],"class_list":["post-7619","post","type-post","status-publish","format-standard","hentry","category-fractualites-ennews-fr","category-seminaires-ness-2013-fr","entry"],"_links":{"self":[{"href":"https:\/\/samovar.telecom-sudparis.eu\/index.php\/wp-json\/wp\/v2\/posts\/7619","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/samovar.telecom-sudparis.eu\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/samovar.telecom-sudparis.eu\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/samovar.telecom-sudparis.eu\/index.php\/wp-json\/wp\/v2\/users\/4"}],"replies":[{"embeddable":true,"href":"https:\/\/samovar.telecom-sudparis.eu\/index.php\/wp-json\/wp\/v2\/comments?post=7619"}],"version-history":[{"count":1,"href":"https:\/\/samovar.telecom-sudparis.eu\/index.php\/wp-json\/wp\/v2\/posts\/7619\/revisions"}],"predecessor-version":[{"id":7620,"href":"https:\/\/samovar.telecom-sudparis.eu\/index.php\/wp-json\/wp\/v2\/posts\/7619\/revisions\/7620"}],"wp:attachment":[{"href":"https:\/\/samovar.telecom-sudparis.eu\/index.php\/wp-json\/wp\/v2\/media?parent=7619"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/samovar.telecom-sudparis.eu\/index.php\/wp-json\/wp\/v2\/categories?post=7619"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/samovar.telecom-sudparis.eu\/index.php\/wp-json\/wp\/v2\/tags?post=7619"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
M. Ken CHEN<\/strong>, Full professor, Universit\u00e9 Sorbonne Paris Nord, FRANCE – Rapporteur
M. Nadjib ACHIR<\/strong>, Associate Professor, INRIA Saclay, FRANCE – Rapporteur
M. Anastasios GIOVANIDIS<\/strong>, Charg\u00e9 de recherche, Ericsson MASSY, FRANCE – Examinateur
M. Vincent GAUTHIER<\/strong>, Associate Professor, Institut Polytechnique de Paris T\u00e9l\u00e9com SudParis, FRANCE – Co-encadrant de these
M. Hassine MOUNGLA<\/strong>, Full professor, Universit\u00e9 Paris Cit\u00e9, FRANCE – Examinateur<\/p>\n\n\n\n\u00ab Apprentissage profond spatiotemporel : de la d\u00e9couverte de motifs \u00e0 la pr\u00e9vision robuste \u00bb<\/h2>\n\n\n\n
pr\u00e9sent\u00e9 par Monsieur Zhaobo HU<\/h2>\n\n\n\n
Abstract :<\/strong><\/p>\n\n\n\n