Systemic validation of fuzzy rule bases: accounting for data availability and the specific characteristics of fuzzy inference
This PhD topic lies within the field of symbolic artificial intelligence. Unlike approaches based on neural networks, these methods rely on explicit rules, often provided by experts or learned from limited data, making them interpretable but potentially imperfect.
The central problem is therefore the validation of fuzzy rule bases: the goal is to ensure that the rules produce consistent, useful, and reliable results. Existing methods use global metrics (overall system performance) and local metrics (the quality of each rule), but they do not sufficiently account for certain important specificities. For example, interactions between rules can strongly influence the final behavior.
The thesis proposes to develop a comprehensive and systematic approach to validate these rule bases, whether data is available or not. In particular, it aims to design new metrics capable of capturing these interactions, drawing inspiration, for example, from graph-based approaches (such as FinGrams or reputation systems).
The work will include the definition of a methodological framework, the proposal of new validation measures, as well as their implementation and experimental evaluation.
The expected outcomes are more precise tools for detecting problematic rules, and an overall improvement in the performance and reliability of fuzzy inference systems.
Securing Generative AI Model: Detection of Advanced Backdoor Attacks
This PhD aims to investigate and detect backdoor attacks within generative AI model ecosystems, including standalone models, retrieval-augmented generation systems (RAG), and LLM-based agent. The research will focus on developing novel detection and defense mechanisms against stealthy trigger-based attacks, emphasizing real-world deployment scenarios and robust evaluation benchmarks. In addition to developing defense mechanisms and releasing the code as open source, the thesis also aims to provide the scientific community with a comprehensive evaluation framework.
Context: Many users (persons, institutions, NGOs and even industries) are currently not in a position to develop their own AI agents. Thus, they may download open-source genAI models or LLM-based agents that are typically designed to be highly accessible and user-friendly, requiring minimal to no technical expertise. This practice is widespread due to the large number of open-source models and LLM agent implementations available online (e.g. Hugging Face hosts over two million public models). Unfortunately, the behavioral integrity of the downloaded model is never verified, and the model may have been previously backdoored. There is therefore an urgent need to provide defense mechanisms capable of scanning the components of a generative AI system (models and knowledge bases) and identifying those that have been poisoned.
Objective: The research will focus on developing novel detection and defense mechanisms against stealthy trigger-based attacks, emphasizing real-world deployment scenarios and robust evaluation benchmarks. In addition to developing defense mechanisms and releasing the code as open source, the thesis also aims to provide the scientific community with a comprehensive evaluation framework.
Post-training neural architecture optimization for small language models
Generative AI, and particularly language models (LLM), have sparked a new revolution in AI with applications across all domains. However, LLMs are highly resource-intensive and, hence, difficult to implement on autonomous embedded systems. LLMs can be optimized by modifying their architecture to replace heavy Transformer layers with lighter alternatives. Given the difficulty of training LLM "from scratch," this thesis aims to develop post-training neural architecture optimization methods applicable to small LLM (SLM). Additionally, the thesis seeks to propose performance metrics of different layers of an SLM and their alternatives, to guide the replacement, and thus propose a comprehensive methodology for optimizing SLMs while considering hardware constraints. The work will be valorized through publications in major AI conferences and journals, and the developed codes and methods could be integrated into the tools developed at CEA.
Out-of-Distribution Detection with Vision Foundation Models and Post-hoc Methods
The thesis focuses on improving the reliability of deep learning models, particularly in detecting out-of-distribution (OoD) samples, which are data points that differ from the training data and can lead to incorrect predictions. This is especially important in critical fields like healthcare and autonomous vehicles, where errors can have serious consequences. The research leverages vision foundation models (VFMs) like CLIP and DINO, which have revolutionized computer vision by enabling learning from limited data. The proposed work aims to develop methods that maintain the robustness of these models during fine-tuning, ensuring they can still effectively detect OoD samples. Additionally, the thesis will explore solutions for handling changing data distributions over time, a common challenge in real-world applications. The expected results include new techniques for OoD detection and adaptive methods for dynamic environments, ultimately enhancing the safety and reliability of AI systems in practical scenarios.
LLM-Assisted Generation of Functional and Formal Hardware Models
Modern hardware systems, such as RISC-V processors and hardware accelerators, rely on functional simulators and formal verification models to ensure correct, reliable, and secure operation. Today, these models are mostly developed manually from design specifications, which is time-consuming and increasingly difficult as hardware architectures become more complex.
This PhD proposes to explore how Large Language Models (LLMs) can be used to assist the automatic generation of functional and formal hardware models from design specifications. The work will focus on defining a methodology that produces consistent and executable models while increasing confidence in their correctness. To achieve this, the approach will combine LLM-based generation with feedback from simulation and formal verification tools, possibly using reinforcement learning to refine the generation process.
The expected outcomes include a significant reduction in manual modeling effort, improved consistency between functional and formal models, and experimental validation on realistic hardware case studies, particularly RISC-V architectures and hardware accelerators.
Distributed multimodal learning for cooperative acoustic source localization and classification
In many complex environments, such as industrial sites, disaster-stricken buildings, or public spaces, it is necessary to automatically detect and localize sound events (falls, alarms, voices, mechanical failures). Mobile platforms equipped with cameras and microphones represent a promising solution, but a single platform remains limited: its microphone array provides an approximate direction towards the source but not a precise position in space, and its camera may be obstructed. This thesis proposes to study how a network of mobile platform, each carrying a calibrated audio-visual unit, can collaborate to localize and classify such events in 3D. Each platform analyses its own audio-visual observations and shares an estimate of the source direction with its neighbours; the network then combines these estimates to reconstruct the position of the event and identify it. The expected outcomes are a cooperative localization system that is robust to occlusions and partial platform failures.
Hydrodynamic simulations of porous materials for ductile damage
The mechanical behavior of metallic materials under highly dynamical loading (schock) and especially their damage behavior is a topic of interest for the CEA-DAM. For tantalum, damage is ductile : by nucleation, growth and coalescence of voids within the material. Usual ductile damage models have been developed using the simplifying assumption that voids are isolated in the materials. However, recent studies by direct simulations explicitly describing a void population in the material (and experimental observations after failure) have shown the importance of void interaction for predicting ductile damage. Yet, the microscopical mechanisms of this interaction remain little known.
The objective of the PhD is to study the growth and coalescence phases of ductile damage through direct numerical simulations of a porous material undergoing dynamic loading. Hydrodynamic simulations, in which voids are explicitly meshed within a continuous matrix, will be used to study relevant scales of length and time. Monitoring the void population throughout the simulation will provide valuable information on the influence of void interaction during ductile damage. Firstly, the bulk behavior will be compared to the one predicted by usual models of isolated voids, showing the macroscopic effect of void interaction. Secondly, the evolution of the size distribution in the void population will be monitored. The last objective will be to understand microscopic void-to-void interaction. In order to take advantage of the wealth of simulation results, approaches based on artificial intelligence (neural networks on the graph associated with the pore population) will be used to learn the link between a void's neighborhood and its growth.
The doctoral student will have the opportunity to develop their skills in shock physics and mechanics, numerical simulations (with access to CEA-DAM supercomputers), and data science.
Point Spread Function Modelling for Space Telescopes with a Differentiable Optical Model
Context
Weak gravitational lensing [1] is a powerful probe of the Large Scale Structure of our Universe. Cosmologists use weak lensing to study the nature of dark matter and its spatial distribution. Weak lensing missions require highly accurate shape measurements of galaxy images. The instrumental response of the telescope, called the point spread function (PSF), produces a deformation of the observed images. This deformation can be mistaken for the effects of weak lensing in the galaxy images, thus being one of the primary sources of systematic error when doing weak lensing science. Therefore, estimating a reliable and accurate PSF model is crucial for the success of any weak lensing mission [2]. The PSF field can be interpreted as a convolutional kernel that affects each of our observations of interest, which varies spatially, spectrally, and temporally. The PSF model needs to be able to cope with each of these variations. We use specific stars considered point sources in the field of view to constrain our PSF model. These stars, which are unresolved objects, provide us with degraded samples of the PSF field. The observations go through different degradations depending on the properties of the telescope. These degradations include undersampling, integration over the instrument passband, and additive noise. We finally build the PSF model using these degraded observations and then use the model to infer the PSF at the position of galaxies. This procedure constitutes the ill-posed inverse problem of PSF modelling. See [3] for a recent review on PSF modelling.
The recently launched Euclid survey represents one of the most complex challenges for PSF modelling. Because of the very broad passband of Euclid’s visible imager (VIS) ranging from 550nm to 900nm, PSF models need to capture not only the PSF field spatial variations but also its chromatic variations. Each star observation is integrated with the object’s spectral energy distribution (SED) over the whole VIS passband. As the observations are undersampled, a super-resolution step is also required. A recent model coined WaveDiff [4] was proposed to tackle the PSF modelling problem for Euclid and is based on a differentiable optical model. WaveDiff achieved state-of-the-art performance and is currently being tested with recent observations from the Euclid survey.
The James Webb Space Telescope (JWST) was recently launched and is producing outstanding observations. The COSMOS-Web collaboration [5] is a wide-field JWST treasury program that maps a contiguous 0.6 deg2 field. The COSMOS-Web observations are available and provide a unique opportunity to test and develop a precise PSF model for JWST. In this context, several science cases, on top of weak gravitational lensing studies, can vastly profit from a precise PSF model. For example, strong gravitational lensing [6], where the PSF plays a crucial role in reconstruction, and exoplanet imaging [7], where the PSF speckles can mimic the appearance of exoplanets, therefore subtracting an accurate and precise PSF model is essential to improve the imaging and detection of exoplanets.
PhD project
The candidate will aim to develop more accurate and performant PSF models for space-based telescopes exploiting a differentiable optical framework and focus the effort on Euclid and JWST.
The WaveDiff model is based on the wavefront space and does not consider pixel-based or detector-level effects. These pixel errors cannot be modelled accurately in the wavefront as they naturally arise directly on the detectors and are unrelated to the telescope’s optic aberrations. Therefore, as a first direction, we will extend the PSF modelling approach, considering the detector-level effect by combining a parametric and data-driven (learned) approach. We will exploit the automatic differentiation capabilities of machine learning frameworks (e.g. TensorFlow, Pytorch, JAX) of the WaveDiff PSF model to accomplish the objective.
As a second direction, we will consider the joint estimation of the PSF field and the stellar Spectral Energy Densities (SEDs) by exploiting repeated exposures or dithers. The goal is to improve and calibrate the original SED estimation by exploiting the PSF modelling information. We will rely on our PSF model, and repeated observations of the same object will change the star image (as it is imaged on different focal plane positions) but will share the same SEDs.
Another direction will be to extend WaveDiff for more general astronomical observatories like JWST with smaller fields of view. We will need to constrain the PSF model with observations from several bands to build a unique PSF model constrained by more information. The objective is to develop the next PSF model for JWST that is available for widespread use, which we will validate with the available real data from the COSMOS-Web JWST program.
The following direction will be to extend the performance of WaveDiff by including a continuous field in the form of an implicit neural representations [8], or neural fields (NeRF) [9], to address the spatial variations of the PSF in the wavefront space with a more powerful and flexible model.
Finally, throughout the PhD, the candidate will collaborate on Euclid’s data-driven PSF modelling effort, which consists of applying WaveDiff to real Euclid data, and the COSMOS-Web collaboration to exploit JWST observations.
References
[1] R. Mandelbaum. “Weak Lensing for Precision Cosmology”. In: Annual Review of Astronomy and Astro- physics 56 (2018), pp. 393–433. doi: 10.1146/annurev-astro-081817-051928. arXiv: 1710.03235.
[2] T. I. Liaudat et al. “Multi-CCD modelling of the point spread function”. In: A&A 646 (2021), A27. doi:10.1051/0004-6361/202039584.
[3] T. I. Liaudat, J.-L. Starck, and M. Kilbinger. “Point spread function modelling for astronomical telescopes: a review focused on weak gravitational lensing studies”. In: Frontiers in Astronomy and Space Sciences 10 (2023). doi: 10.3389/fspas.2023.1158213.
[4] T. I. Liaudat, J.-L. Starck, M. Kilbinger, and P.-A. Frugier. “Rethinking data-driven point spread function modeling with a differentiable optical model”. In: Inverse Problems 39.3 (Feb. 2023), p. 035008. doi:10.1088/1361-6420/acb664.
[5] C. M. Casey et al. “COSMOS-Web: An Overview of the JWST Cosmic Origins Survey”. In: The Astrophysical Journal 954.1 (Aug. 2023), p. 31. doi: 10.3847/1538-4357/acc2bc.
[6] A. Acebron et al. “The Next Step in Galaxy Cluster Strong Lensing: Modeling the Surface Brightness of Multiply Imaged Sources”. In: ApJ 976.1, 110 (Nov. 2024), p. 110. doi: 10.3847/1538-4357/ad8343. arXiv: 2410.01883 [astro-ph.GA].
[7] B. Y. Feng et al. “Exoplanet Imaging via Differentiable Rendering”. In: IEEE Transactions on Computational Imaging 11 (2025), pp. 36–51. doi: 10.1109/TCI.2025.3525971.
[8] Y. Xie et al. “Neural Fields in Visual Computing and Beyond”. In: arXiv e-prints, arXiv:2111.11426 (Nov.2021), arXiv:2111.11426. doi: 10.48550/arXiv.2111.11426. arXiv: 2111.11426 [cs.CV].
[9] B. Mildenhall et al. “NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis”. In: arXiv e-prints, arXiv:2003.08934 (Mar. 2020), arXiv:2003.08934. doi: 10.48550/arXiv.2003.08934. arXiv:2003.08934 [cs.CV].
Design artificial intelligence tools for tracking Fission Product release out of nuclear fuel
The Laboratory for the Analysis of Radionuclide Migration (LAMIR), part of the Institute for Research on Nuclear Systems (IRESNE) at CEA Cadarache, has developed a set of advanced measurement methods to characterize the release of fission products from nuclear fuel during thermal transients. Among these innovative tools is an operando in situ imaging system that enables real-time observation of these phenomena. The large amount of data generated by these experiments requires dedicated digital processing techniques that account for both the specificities of nuclear instrumentation and the underlying physical mechanisms.
The goal of this PhD project is to develop an optimized data processing approach based on state-of-the-art Artificial Intelligence (AI) methods.
In the first phase, the focus will be on processing thermal sequence images to detect and analyze material movements, aiming to identify an optimal image-processing strategy defined by rigorous quantitative criteria.
In the second phase, the methodology will be extended to all experimental data collected during a thermal sequence. The long-term objective is to create a real-time diagnostic tool capable of supporting experiment monitoring and interpretation.
This PhD will be carried out within a collaborative framework between LAMIR, which has recognized expertise in nuclear fuel behavior analysis and imaging, and the Institut Fresnel in Marseille, known for its strong background in image analysis and artificial intelligence.
The candidate will benefit from a multidisciplinary and stimulating research environment, with opportunities to present and publish their work at national and international conferences and in peer-reviewed journals.
Adaptive and explainable Video Anomaly Detection
Video Anomaly Detection (VAD) aims to automatically identify unusual events in video that deviate from normal patterns. Existing methods often rely on One-Class or Weakly Supervised learning: the former uses only normal data for training, while the latter leverages video-level labels. Recent advances in Vision-Language Models (VLMs) and Large Language Models (LLMs) have improved both the performance and explainability of VAD systems. Despite progress on public benchmarks, challenges remain. Most methods are limited to a single domain, leading to performance drops when applied to new datasets with different anomaly definitions. Additionally, they assume all training data is available upfront, which is unrealistic for real-world deployment where models must adapt to new data over time. Few approaches explore multimodal adaptation using natural language rules to define normal and abnormal events, offering a more intuitive and flexible way to update VAD systems without needing new video samples.
This PhD research aims to develop adaptable Video Anomaly Detection methods capable of handling new domains or anomaly types using few video examples and/or textual rules.
The main lines of research will be the following:
• Cross-Domain Adaptation in VAD: improving robustness against domain gaps through Few-Shot adaptation;
• Continual Learning in VAD: continually enriching the model to deal with new types of anomalies;
• Multimodal Few-Shot Learning: facilitating the model adaptation process through rules in natural language.