CEA Tech
PhD position - Knowledge-driven universal multimodal representation
CEA Tech
CEA Tech is building on CEA Leti’s successful track record innovating for industry.
Visit employer page
Published: 8 days ago
Application deadline: Oct 31
Location: Grenoble, France
You need to sign in or create an account to save this job
Please mention that you found this job on Academic Positions when applying.

PhD position - Knowledge-driven universal multimodal representation



Computer science and software


It is now common knowledge that the domain of artificial intelligence has seen a significant renewal these last years. It is mainly due to the impressive results obtained, for many tasks, by deep neural networks that raise many opportunities. In particular, these approaches have shown very impressive results for several classical visual recognition tasks, natural language and speech processing and recognition tasks but they also enable considerable advances in higher level tasks such as playing at the Go Game or answering to visual questions. However, despite the superior performance of these deep neural networks approaches, it remains challenging to understand their inner workings and explain their output predictions. As a consequence they are often considered as block-boxes which is a strong limitation for their applications in various domains in which transparency and explainability are of strong importance. As a consequence, despite the recent success, one of the major challenge of AI is to be able to designed AI-enabled systems, able to predict but also to explain and to justify the results of their predictions. Although, various initiatives have been launched recently both in the artificial intelligence community (XAI : explainable artificial intelligence) and in the machine learning one (Interpretable machine learning models). In particular it is common to classify the approaches on this field into three families: (1) Approaches that try to build interpretable and explainable models by design, either by using models that are by essence interpretable such that compact decision trees or rule-based systems or by capturing the interpretability into the objective function; (2) Approaches that try to interpret and understand the internal representation of black-boxes models and (3) Approaches that try to interpret the predictions resulting from black-boxes models.In this context and taking part to these initiatives, this thesis aims at combining and unifying various approaches coming from different sub-fields of Artificial Intelligence (knowledge representation and reasoning, machine and deep-neural learning, computer vision and multimedia processing) to propose new approaches to learn more universal and more interpretable multimodal representations. A possible task to test and evaluate the contributions is the Visual Question Answering (VQA) problem, defined as the problem of automatically answering questions about images. Indeed, it is known as a excellent way to test the capabilities of a system to reason about entities and their relationships and it is often taken as the AI-complete task in both the context of explainable AI and the related context of universal representations.


Département Intelligence Ambiante et Systèmes Interactifs (LIST)

Vision & Ingénierie des Contenus (SAC)






CEA Saclay - Nano-INNOVBat 861 - PC 173 - F91191 Gif Sur Yvette CedexFrance

Phone number: +33 (0)1 69 08 0152

Email: herve.le-borgne@cea.fr


Ecole Centrale Paris

Interfaces: Approches interdisciplinaires / fondements; applications et innovation





Start date on 01-09-2018




Laboratoire MICS

Centrale-SupélecCampus de Châtenay-Malabry

Phone number:


« The age limit is 26 years for PhD offers and 30 years old for post-doc offers. »

Apply now
You need to sign in or create an account to save this job


Personal details
Upload your CV and attachments
The file format must be a .doc, .pdf, or .rtf, and the file cannot exceed 2 MB in size.
The file format must be a .doc, .pdf, or .rtf, and the file cannot exceed 2 MB in size.

By applying for a job listed on Academic Positions you agree to our terms and conditions and privacy policy.


CEA Tech
CEA Tech
Location: Grenoble, France | Closing on Oct 31
PhD position - Fabrication of transistors in semimetal nanowires
SL-DRT-18-0787 RESEARCH FIELD Solid state physics, surfaces and interfaces ABSTRACT Bismuth is a semimetal. In the bulk state it is a semimetal, or, more precisely, a semiconductor with a negative bandgap (-80 meV). In the nanowire state, however, the...
CEA Tech
CEA Tech
Location: Grenoble, France
Post Doc - Mechanical modelling of a PEM fuel cell stack for automotive application
PsD-DRT-18-0069 RESEARCH FIELD Mechanics, energetics, process engineering ABSTRACT The proton exchange membrane fuel cell (PEMFC) is a promising candidate for many applications, whether for stationary applications or transport. Concerning the automotive...
CEA Tech
CEA Tech
Location: Grenoble, France | Closing on Oct 31
PhD position - New diamond-based polymer composite materials toward high thermal and mechanical performance applications
SL-DRT-18-0637 RESEARCH FIELD Ultra-divided matter, Physical sciences for materials ABSTRACT Diamond is well known for its outstanding properties such as its hardness, high thermal conductivity or high electrical resistivity. Mostly synthesized by CVD...
CEA Tech
CEA Tech
Location: Grenoble, France | Closing on Oct 31
PhD position - Participatory localization of connected objects through deep learning techniques
SL-DRT-18-0806 RESEARCH FIELD Mathematics - Numerical analysis - Simulation ABSTRACT Conventional localization methods based on low-cost and low-complexity wireless communication standards (e.g., Long Range - IoT Lora or Sigfox systems, WiFi/BT-LE) can...