I work at CEA-List, that is associated with Université Paris-Saclay.

My research is in the area of computer vision with a particular interest on relating visual and textual content and visual generative models.

News

  • [oct 26] One paper accepted at WACV 2027, which is the result of a work with Alexandre-Fournier Montgieux, Akram Elbouanani and their thesis supervisors about auditing face-related biases in VLMs. We proposed an original method to circumvent legitimate restrictions on the use of personal data, such as portrait photos. The code and data are available.
  • [jul 26] Oral presentation of Katia Benali at ICPR 2026 for her work on Interpretable image recognition with variable number of prototypes. I personaly went there to present the poster of Paul Grimal on Unique Step Refinement for Transformer-based Generative Models
  • [jun 26] gold reviewer (top 25%) at ICML 2026 and outstanding reviewer (top 5%) at CVPR 2026.
  • [feb 26] Clément Cornet has his first paper accepted at ICLR 2026. He proposed an unsupervised method for extracting interpretable concepts from neural networks, an alternation to spare autoencoders that he named the “Deleuzian representation”, since its method matches somehow the Deleuze’s definition of a concept. Congratulation to him, it is very promising!

Main public projects

  • the STUDIES project (2023-2027) aims at automatically conducting die studies at a large scale to better understand Ancient Numismatics.
  • the SHARP project (PEPR IA 2023 - 2029) aims to design, analyze and deploy intrinsically frugal models (neural or not) able to achieve the versatility and performance of the best models while requiring only a vanishing fraction of the resources currently needed.
  • the Vanguard project (EU Horizon 2023-2026) aims to strengthen the fight against trafficking in human beings through the provision of improved intelligence picture, delivery of advanced and trustworthy suite of tools, and realisation of innovative training activities.
  • the Deepgreen project (France2030 ANR 2024-2027) that aims to develop an embedded artificial intelligence model that combines sovereignty, trust and frugality, by developping the Eclipse AIDGE open source software platform.

Teaching

I give courses and tutorials of deep learning at CentraleSupelec (since 2017).

Before, I have given courses of computer vision at Centrale Paris (2016), tutorials of NLP at INSTN (2007-2009), courses of video compression at Dublin City University during my post-doc (2005-2006) and various tutorials at Université Joseph Fourier and INP Grenoble during my thesis (2001-2003).

Contact

I am physicaly located at Nano Innov. To get here, I’ll need to give your name to reception first (the appointment must be planned in advance). On the day of your visit, please bring your passport or identity card.

Mailing address: CEA Saclay - Nano-INNOV, DRT/LIST/DIASI/SIALV/LASTI, Bat 861 - PC 184 - F91191 Gif-sur-Yvette Cedex, France.

Short bio

Hervé Le Borgne is a researcher at the CEA-List since 2006, carrying out research on computer vision and focusing on the links between visual and textual content. Previously, he received his Ph.D. from the INP Grenoble in 2004 and worked as a postdoc at Dublin City university from 2004 to 2006 and received an habilitation to supervise thesis (HDR) in 2024 from Université Paris Saclay. He published more than 75 articles in international conferences and journals dealing with computer vision, multimedia retrieval, text-to-image models and more generally multimedia mining in order to extract semantic. He is also the co-inventor of 10 patents in these fields. Recently, his interest focused on visual generative models, fairness and explainability. He has served as a reviewer for several international conferences and journals, including CVPR, ICCV, ECCV, NeurIPS, ICLR, ICML, ACM Multimedia, IEEE T. PAMI and IEEE T. Multimedia. He has been a project manager since 2006, both for public funded (ANR, H2020, HE, PEPR) projects and industrial contracts. He supervised 22 master students co-advised nine completed Ph.D and seven postdocs, currently advising four PhD and two postdocs. Since 2016, he is an external teacher of deep learning at CentraleSupelec.

Positions and Education