Catégorie offres: PhD

Call for PhD : The Requirement of Human Oversight Applied to Artificial Intelligence in Healthcare: Foundations, Scope, and Implementation Modalities

Call for PhD : The Requirement of Human Oversight Applied to Artificial Intelligence in Healthcare: Foundations, Scope, and Implementation Modalities

Contexte et enjeux

La généralisation des technologies d’intelligence artificielle dans le domaine de la santé soulève des défis juridiques majeurs. Systèmes d’aide au diagnostic, robots chirurgicaux, outils de prédiction et de triage médical : tous impliquent une délégation partielle, voire totale, de fonctions décisionnelles critiques. Cette évolution interroge le rôle de l’humain dans la chaîne de responsabilité et de légitimité des décisions médicales. Dans ce contexte, l’exigence de « supervision humaine » constitue un principe émergent au croisement du droit de la santé, du droit de la responsabilité et du droit de la régulation des technologies numériques.

Problématique

Le principe de « supervision humaine », aussi nommé « garantie humaine » ou contrôle humain », est aujourd’hui requis tant par les règlementations européennes (AI Act, RGPD) que nationale (nouvelle loi bioéthique française du 2 août 2021). Il demeure néanmoins juridiquement flou. Sa détermination pose plusieurs difficultés : Comment mettre en oeuvre cette exigence ? Quelle place doit être réservé à l’humain dans l’usage des systèmes autonomes ? Comment articuler cette exigence avec les droits fondamentaux du patient, les principes de précaution et d’efficacité médicale, ainsi que la responsabilité des professionnels de santé et des concepteurs d’IA ? Quels sont les standards opérationnels à intégrer dans la mise en oeuvre concrète de cette exigence ?

Objectifs de la thèse

La recherche visera à :

Clarifier les fondements juridiques de l’exigence de supervision humaine dans le champ de l’IA en santé ;
Déterminer le contenu et la portée normative du principe
Proposer une grille d’analyse pour sa mise en oeuvre effective, tenant compte des contraintes techniques, cliniques et juridiques.

Profil recherché

Le/la candidat(e) devra être titulaire d’un Master 2 en droit de la santé ou droit du numérique (avec alors une appétence avérée pour les thématiques santé), faire preuve de solides capacités d’analyse juridique et d’un intérêt marqué pour les enjeux interdisciplinaires, la recherche impliquant de nombreux échanges avec les acteurs du secteur médical.
Il est indispensable d’avoir rédigé un mémoire durant le Master II et d’avoir au minimum obtenu la note de 13/20.

[PhD] Sujet de Thèse : « Modèles génératifs image vers 3D avec contraintes géométriques et physiques »

[PhD] Sujet de Thèse : « Modèles génératifs image vers 3D avec contraintes géométriques et physiques »

Contexte :
La thèse proposée se déroulera dans le cadre du projet PR[AI]RIE-PSAI (Paris School of Artificial Intelligence (AI)). Ce projet est le grand lauréat du programme « IA Cluster », porté par l’Université PSL, et a pour objectif de faire progresser les connaissances en matière d’IA, à proposer un enseignement supérieur de niveau international et à produire des innovations de rupture.

La thèse aura lieu au Centre de Robotique de Mines Paris – PSL. Le Centre de Robotique est un laboratoire de recherche spécialisé dans l’IA temps réel et les interactions Humain-Machine, appliquées aux véhicules automatisés, à la robotique mobile et collaborative, ainsi qu’à l’Industrie du Futur. Au sein de ce centre, l’axe Nuages de Points et Modélisation 3D (NPM3D) développe des techniques d’acquisition, traitement et rendu de nuages de points 3D, issues de la photogrammétrie ou du LiDAR, pour des applications variées (cartographie, robotique, patrimoine, archéologie).

Call for PhD – Simulating the Dead: Historical AI as Tools for the Humanities

Call for PhD – Simulating the Dead: Historical AI as Tools for the Humanities

Scientific context

Humanities scholars have long sought to understand how past people thought, felt, and interpreted the world. Traditional methods — such as close reading, archival analysis, and philology — offer rich, interpretive insights, but they are often labor-intensive and limited in scope. Quantitative approaches like word frequency, topic modelling and word embedding analysis have expanded our methodological toolkit, but remain indirect proxies for psychological or cultural traits [1-4].

Recent advances in artificial intelligence (AI) have opened up novel avenues for understanding the human experience across time. Among the most intriguing frontiers is the development of Historical Large Language Models (HLLMs) — language models trained on corpora of historical texts [5]. These models offer the potential to simulate plausible psychological responses and cultural representations from individuals who lived in past societies, effectively creating populations of ‘virtual ancestors’. An HLLM trained on a specific corpus — say, 18th-century French political tracts, or Qing dynasty administrative documents — can respond to prompts with outputs that reflect the linguistic and conceptual patterns present in its training data. These simulated responses can be interrogated using psychological instruments or thematic surveys, generating data that, while artificial, may reveal the distribution of beliefs or values latent in a cultural moment. One could, for instance, estimate levels of authoritarianism, concern for purity, or belief in free will.

MonadGPT (https://huggingface.co /Pclanglais/MonadGPT) provides an example of what we have in mind. This is a fine-tuned version of the Mistral-Hermes 2 model, trained on a corpus of 11,000 early modern texts in English, French, and Latin, primarily sourced from Early English Books Online (EEBO) and Gallica. This model is designed to emulate the language and conceptual frameworks of the 17th c., offering insights into the discourse of that era.

Call for PhD: Scientific deep learning for Anomaly Detection in ductile DAmage Modelingapplied to metal forming

Call for PhD: Scientific deep learning for Anomaly Detection in ductile DAmage Modelingapplied to metal forming

Summary of the project: Damage is a particular form of anomaly in material forming. These anomalies
come from materials microstructure heterogeneity that drives ductile damage mechanisms. We propose
to combine deep learning for anomaly detection and mechanical modeling of damage. This work is
limited to the use of synthetic data produced with mechanical models calibrated in the context of previous
work in materials mechanics. However, these models remain imperfect, in particular for dealing with
recycled materials or, in general, materials with a high variability of their physical properties. In this case,
an anomaly may be caused by unusual properties or an unsuitable mechanical model. The anomalies
will be identified as cases out-of-distribution of so-called normal data. The objective of this project is to
develop: (i) self-supervised learning of a latent space of normal data, (ii) an anomaly detection task
using this latent space, (iii) a final stage of scientific explanation of the causes of anomalies based on
explainable AI. All this in the context of large deformations of point cloud.

PhD Position in AI accelerated simulations of chemical reactivity

PhD Position in AI accelerated simulations of chemical reactivity

Research Topic
One of the main research interests within the Chemical Theory and Modelling (CTM) research group is the use of AI accelerated computational chemistry to explore chemical reaction spaces, e.g., with the aim to discover new performant catalysts, and analyze/engineer complex reaction networks, including plausible prebiotic (auto-)catalytic cycles that could shed light on the origin of life.
In recent years, we have contributed both application-driven and methodological advances in this area. For instance, we have developed a high-throughput screening protocol to identify bioorthogonal click reactions from a chemical space exceeding 10 million possibilities.(1) Additionally, we have created TS-tools, a software package for the automated generation of diverse reaction profiles for unknown reactions.(2)
However, a significant limitation in our current approaches is the lack of an
accurate and e\icient description of solvent e\ects, particularly for reactions occurring
in polar environments. This hinders our ability to extend high-throughput reaction
screening methods to many biologically and industrially relevant processes. The goal of
this PhD project is to leverage machine learning interatomic potentials (MLIPs) to
enable large-scale reactivity exploration in solution.
MLIPs are neural network-based models trained to predict energies – and atomic forces – based on molecular geometries.(3) When trained on high-quality Density Functional Theory (DFT) data, these models can simulate (the dynamics associated with) a complete reaction path, including explicit solvation e\ects, at a fraction of the computational cost of full quantum chemical simulations. However, obtaining
representative training data for reactive events remains a challenge, as conventional molecular dynamics (MD) simulations often fail to sample rare reaction events e\ectively.
To overcome this, most researchers employ enhanced sampling MD simulations at the DFT level, combined with active learning, to generate training data.(4)
While e\ective, these approaches require pre-defining reaction coordinates, inherently biasing the training data generated, and hence also the resulting MLIP. This PhD project will firstly aim to develop MLIPs for reactions in solution with minimal pre-imposed mechanistic assumptions, by training them on snapshots from many, diverse reaction pathways, generated by TS-tools, in combination with our in-house reaction pathway enumeration software (currently under development).
More specfically, the methodology will involve:

  1. Generating diverse reaction pathways, with a range of intermediate geometries or snapshots along them, for a given molecular system using TS-tools and our pathway enumerator.
  2. Solvating all generated intermediate geometries/snapshots in an automated manner with explicit solvent clusters.
  3. Training an MLIP on these diverse solvated geometries, leveraging the approach pioneered by Fernanda Duarte and co-workers, who demonstrated that MLIPs trained on cluster models of water are transferable to simulations in bulk solution.(5)
    Subsequently, we will also aim to integrate the TS-tools approach into the active learning-based refinement of the developed MLIPs, i.e., the sampling of additional snapshots in regions of the PES where the initial MLIP is uncertain about its predictions.
    In a final part of the project, the impact of transfer learning, either by finetuning an existing general purpose MLIP, such as ANI,6 or by transfering an in-house developed MLIP from its original reactive system to a new one, on data-e\iciency and generalizability, will be considered.
    Overall, our aim is to rapidly develop generalizable, unbiased MLIPs capable of mechanistic discovery without imposing (strong) human preconceptions in the training data. The resulting models could potentially transform high-throughput reaction exploration in aqueous environments, with applications spanning catalysis, prebiotic chemistry, and beyond.
    References
  4. Stuyver, T.; Coley, C. W. Machine Learning-Guided Computational Screening of
    New Candidate Reactions with High Bioorthogonal Click Potential. Chem.—Eur.
    J. 2023, 29, e202300387.
  5. Stuyver, T. TS-Tools: Rapid and Automated Localization of Transition States Based
    on a Textual Reaction SMILES Input. J. Comput. Chem. 2024, 45, 2308–2317.
  6. Behler, J. Perspective: Machine Learning Potentials for Atomistic Simulations. J.
    Chem. Phys. 2016, 145, 170901.
  7. David, R.; de la Puente, M.; Gomez, A.; Anton, O.; Stirnemann, G.; Laage,
    D. ArcaNN: Automated Enhanced Sampling Generation of Training Sets for
    Chemically Reactive Machine Learning Interatomic Potentials. Digit. Discov.
    2025, 4, 54–72.
  8. Zhang, H.; Juraskova, V.; Duarte, F. Modelling Chemical Processes in Explicit
    Solvents with Machine Learning Potentials. Nat. Commun. 2024, 15, 6114.
  9. Smith, J. S.; Isayev, O.; Roitberg, A. E. ANI-1: An Extensible Neural Network
    Potential with DFT Accuracy at Force Field Computational Cost. Chem.
    Sci. 2017, 8, 3192–3203.

    Eligibility and Selection Criteria
    Candidates will be evaluated based on:
  • Academic excellence
  • Relevance of their background to the research topic
    The selection process follows an open, transparent, and merit-based (OTM) recruitment
    procedure.
    Non-discrimination, openness, and transparency:
    All PR[AI]RIE-PSAI partners are committed to supporting and promoting equality,
    diversity, and inclusion within their communities. We encourage applications from
    diverse backgrounds and ensure a fair selection process.
    Application Requirements
    Applicants must submit the following documents:
  1. Curriculum Vitae (CV)
  2. Motivation Letter (max. 1 page) describing:
  • Your interest in the research topic
  • How your background aligns with the project
  1. Copy of your most recent diplomas
    Application Procedure
  • Deadline: May 20, 2025, at 17h00
  • Applications should be submitted to: thijs.stuyver@chimieparistech.psl.eu
  • The evaluation process consists of two phases:
  1. Pre-selection by the supervisor.
  2. Final selection by an expert committee, evaluating applications based on excellence
    and alignment with PR[AI]RIE-PSAI’s scientific program.
    Final results will be communicated by June 15, 2025.
    For additional information, please visit https://thijsstuyver.com or contact
    thijs.stuyver@chimieparistech.psl.eu.

PhD Candidate/Institute Pasteur – PhD proposal: An Olfactory Pathway to Consciousness for Severe Brain-Injured Patients

PhD Candidate/Institute Pasteur – PhD proposal: An Olfactory Pathway to Consciousness for Severe Brain-Injured Patients

PhD proposal: An Olfactory Pathway to Consciousness for Severe Brain-Injured Patients

Introduction

Consciousness is characterised by the ability to awake and interact with our environment. This state of wakefulness promotes sensory perceptions, whether from outside (the environment) or inside (for example, memory). These perceptions are then processed and made conscious through the action of specialised neural networks called correlates of consciousness (NCCs; Mashour et al., Neuron 2020). Studying NCCs has led to multiple theories explaining their functioning, including the global neuronal workspace theory, developed by the Dehaene and Changeux group (2004) and inspired by Baars (1988), and the integrated information theory, proposed by Tononi (Boly et al., J. Neurosci. 2017), and inspired by Edelman (1989). These theories emphasise the crucial role of the prefrontal-parietal network, particularly
the anterior cingulate cortex and the dorsal prefrontal cortex. They also highlight the necessity of functional integrity in the circuits of wakefulness, including the basal ganglia (ventral tegmental area) and brainstem nuclei integrated into the ascending reticular activating system, involving ponto-mesencephalic neurons and the locus coeruleus (Koch et al., Nat Rev Neurosci, 2016).


Acute primary brain injuries, such as intraparenchymal hematomas, subarachnoid haemorrhages, ischemic strokes, and traumatic brain injuries, can seriously threaten the functionality of central neural networks and cause disorders of consciousness (TDC), ranging from transient coma to prolonged states of altered consciousness (non-responsive wakefulness state, minimally conscious state). These disorders reflect the severity of the initial brain impact (Maas et al., Lancet Neurol. 2017). Measuring the extent of neuronal damage and predicting the speed and degree of consciousness recovery represents a double challenge for science and medical practice (Maciel, Contin Lifelong Learn Neurol, 2021).


The severity of brain injuries often necessitates admission to an intensive care unit and in some cases, neurosurgical or neuroradiological intervention. The management of potential complications, such as intracranial hypertension or secondary complications (e.g., hyperthermia, infections, hemodynamic, hydro-electrolytic, and metabolic disturbances), frequently requires deep sedation. This is often necessary, even in patients with pre-existing TDC, to optimise brain metabolism and perfusion (Oddo et al., Crit Care. 2016). Deep sedation for several days or more, in cases of refractory intracranial hypertension, can mask changes in the neurological status of the patient, complicating neurological assessment and prognostication. However, critical decisions regarding the limitation or withdrawal of treatment are often made in this complex context (Lazaridis et al., Neurocrit Care, 2019).

The consequences of these decisions, affecting patients, their families, society, and the healthcare system, are significant, placing the evaluation of the prognosis of severely brain-injured patients at the heart of an important debate for public health involving scientific, medical, and ethical considerations.

Predicting the recovery of consciousness and wakefulness depends on our ability to evaluate consciousness-related neural networks’ anatomical and functional integrity (NCCs). This involves examining the brain’s ability to perceive and integrate, in a conscious manner, stimuli ranging from simple to complex, thus revealing the integrity of the circuits responsible for wakefulness and consciousness. A multimodal approach is ideal, combining clinical assessment, advanced imaging techniques such as a positron emission tomography and functional magnetic resonance imaging, and neurophysiological methods including electroencephalography (EEG) and event-related potentials. However, healthcare professionals often face a dilemma between the accuracy of these diagnostic tools and their immediate
availability, particularly in critically ill patients in intensive care units.


EEG is particularly appreciated for its non-invasiveness, the possibility of bedside performance, and availability in most intensive care units. Additionally, it offers high temporal and spatial resolution, enabling the precise evaluation of neural changes in the brain in response to external stimulation, or “EEG reactivity” (Admiraal et al., Neurology, 2020). Significant advances have been made in interpreting EEG signals in patients with chronic disorders of consciousness (TDC), starting with the introduction of complex mathematical analysis methods for signal processing (quantitative analysis, such as spectral analysis or connectivity) (Sitt et al., Brain J Neurol, 2014), and the development of more sophisticated sensory stimulation techniques (Arzi et al., Nature, 2020). EEG reactivity to verbal motor commands has been detected in comatose patients, a condition known as cognitive-motor dissociation, and this reactivity has been linked to better long-term cognitive and functional prognosis (Claassen et al., NEJM, 2019). However, the effectiveness of these quantitative analyses and their general applicability are largely dependent on the quality of the EEG recording, which may be compromised in an intensive care setting due to factors such as artefacts, a limited number of electrodes due to intracranial medical devices, or the effect of sedation. It is hypothesised that overcoming these obstacles by removing the abovementioned
constraints would lead to more reliable qualitative and quantitative analyses.

Furthermore, regarding stimulation paradigms, the nature of the stimulus plays a crucial role in determining the types and localisation of neural networks activated. The complexity of the stimulus, whether passive or active and the sensory modality used are key factors influencing the specific activation of neural networks. The addition of an emotional component to the stimulus, such as the use of a patient’s name instead of a simple sound, can activate additional neural networks (e.g., temporal and limbic circuits) associated with interoceptive mechanisms, which are unique to each individual (Holeckova et al., Brain Res, 2006).


The hypothesis is that it is possible to find better access to the consciousness of severely brain-damaged patients by using stimuli known for their strong emotional and memory associations: olfactory stimulation. An approach combining generative models and advanced statistical structure is proposed to develop a new method for probing altered states of consciousness through olfactory stimulation.

This project consists of three initiatives

  • The generative approach: Develop a generative algorithm of synthetic EEGs and a statistical structure procedure to create a high-resolution EEG (32 to 64 electrodes), free of missing electrodes, artefacts, and the effects of sedatives, using advanced modelling techniques known for their better generalisation capabilities. A detailed statistical structuring of the data will be carried out, as the problem involves multiple null models that need to be considered to ensure the generative model is conditioned to these models. We will train the generative model progressively using EEGs from healthy subjects, sedated subjects without brain injuries, brain-injured subjects without sedation, and finally, EEGs from brain-injured subjects under sedation. We will also use high-resolution databases (32 to 64 electrodes) to learn the generation of high-resolution EEGs from clinical recordings with 12 or less in cases where electrodes are missing
  • The olfactory stimulus: Develop and optimise the olfactory stimulus to understand its effects on patients
  • Predicting the fate of patients: Use the developed generative model and the optimised olfactory stimulus to characterise the evolution of consciousness in brain-injured patients in collaboration with Eleonore Bouchereau, a medical chief resident, and her forthcoming thesis. This will include a clinical study with the ethical paperwork submitted before the project’s start. The objective is to follow a variety of patients admitted to the service of Professor Sharshar, a director of neurophysiology. Based on these “experimental” trajectories, we will use physical modelling and Bayesian probability to create a “physics” of the trajectory of patients in the latent space learned by the generative model. This “physics” will then be used to predict the medical fate of new patients. This will deliver not only predictive models but also trajectories, along with the probabilities of these trajectories, providing a complete tool for medical decision-making.

The medical environment beyond the laboratories of PM Lledo and JB Masson

Our project is fundamental and highly theoretical in its execution, but its applications in medicine are immediate. Patients are recruited from the departments of anesthesiology-intensive care, neurology, and neurophysiology (Pr Martine Gavaret) of the GHU-Psychiatry and Neurosciences in Paris. These departments are expert centres for managing and neuro-prognostication patients with severe acute brain injuries

Regarding EEG data from healthy subjects, we have access to a public database and our own database, supported by the GHU-Psychiatry and Neurosciences in Paris. For sedated subjects without brain injuries, we have a database of approximately 250 EEGs collected from patients undergoing cardiac surgery at the European Georges Pompidou Hospital (validation by CERAR and CNIL already obtained for data exploitation). We are building a clinical database of patients with 64 high-resolution electrode EEGs collected in the neurology department (Pr. Guillaume Turc) of the GHU-Psychiatry and Neurosciences in Paris for brain-injured subjects not under sedation. For sedated subjects, whether brain-injured or not, we have a database of 322 patients collected during a multicenter study (PRoReTro – NCT02395868) that we have conducted.

Applications in medicine beyond the project

The production of a synthetic EEG enables the extrapolation of a high-resolution EEG from a standard clinical recording of a few electrodes, which will have multiple applications, particularly in neurology and psychiatry. This could lead to the development and commercialisation of a startup (PM Lledo and JB Masson have already launched their startups) In addition to its prognostic value, synthetic EEG could contribute to a better understanding of new neural networks and mechanisms involved in consciousness, as well as cognitive and functional disorders. This could also help refine medical diagnoses with improved
spatial resolution (e.g., in epilepsy). Additionally, our methods for removing the effects of sedation from EEG signals could be developed for use in perioperative monitoring or for optimising the titration of sedation during anesthesia-intensive care management.

The candidate

The successful candidate for this fully funded proposal should have a strong background in either physics or applied mathematics, advanced computational skills and the willingness to work in a highly interdisciplinary environment on highly complex medical data. This PhD will be joined to the one of Eleonore Bouchereau an MD who was recently granted “a poste Acceuil Inserm”. Candidates should send their application to dbc-epi-recrutement@pasteur.fr.

References

Mashour GA, Roelfsema P, Changeux JP, Dehaene S. Conscious Processing and the Global Neuronal Workspace Hypothesis. Neuron. 2020 Mar 4;105(5):776-798. doi: 10.1016/j.neuron.2020.01.026. PMID: 32135090; PMCID: PMC8770991.

Sergent C, Dehaene S. Neural processes underlying conscious perception: experimental findings and a global neuronal workspace framework. J Physiol Paris. 2004 Jul-Nov;98(4-6):374-84. doi: 10.1016/j.jphysparis.2005.09.006. Epub 2005 Nov 15. PMID: 16293402.

Boly M, Massimini M, Tsuchiya N, Postle BR, Koch C, Tononi G. Are the Neural Correlates of Consciousness in the Front or in the Back of the Cerebral Cortex? Clinical and Neuroimaging Evidence. J Neurosci. 2017 Oct 4;37(40):9603-9613. doi: 10.1523/JNEUROSCI.3218-16.2017. PMID: 28978697; PMCID: PMC5628406.

Koch C, Massimini M, Boly M, Tononi G. Neural correlates of consciousness: progress and problems. Nat Rev Neurosci. mai 2016;17(5):307‐21

Maas AIR, Menon DK, Adelson PD, Andelic N, Bell MJ, Belli A, et al. Traumatic brain injury: integrated approaches to improve prevention, clinical care, and research. Lancet Neurol. 1 déc 2017;16(12):987‐1048

Maciel CB. Neurologic Outcome Prediction in the Intensive Care Unit. Contin Lifelong Learn Neurol. oct 2021;27(5):1405

Oddo M, Crippa IA, Mehta S, Menon D, Payen JF, Taccone FS, et al. Optimizing sedation in patients with acute brain injury. Crit Care. 2016;20:128.

Lazaridis C. Withdrawal of Life-Sustaining Treatments in Perceived Devastating Brain Injury: The Key Role of Uncertainty. Neurocrit Care. 1 févr 2019;30(1):33‐41

Admiraal MM, Horn J, Hofmeijer J, Hoedemaekers CWE, Kaam CR van, Keijzer HM, et al. EEG reactivity testing for prediction of good outcome in patients after cardiac arrest. Neurology. 11 août 2020;95(6):e653‐61

Sitt JD, King JR, El Karoui I, Rohaut B, Faugeras F, Gramfort A, et al. Large scale screening of neural signatures of consciousness in patients in a vegetative or minimally conscious state. Brain J Neurol. août 2014;137(Pt 8):2258‐70

Arzi A, Rozenkrantz L, Gorodisky L, Rozenkrantz D, Holtzman Y, Ravia A, et al. Olfactory sniffing signals consciousness in unresponsive patients with brain injuries. Nature. mai 2020;581(7809):428‐33

Claassen J, Doyle K, Matory A, Couch C, Burger KM, Velazquez A, et al. Detection of Brain Activation in Unresponsive Patients with Acute Brain Injury. N Engl J Med. 27 2019;380(26):2497‐505

Holeckova I, Fischer C, Giard MH, Delpuech C, Morlet D. Brain responses to a subject’s own name uttered by a familiar voice. Brain Res. 2006 Apr 12;1082(1):142-52. doi: 10.1016/j.brainres.2006.01.089. PMID: 16703673.

Ho J, Jain A, Abbeel P. Denoising Diffusion Probabilistic Models [Internet]. arXiv; 2020 [cité 21 déc 2023]. Disponible sur: http://arxiv.org/abs/2006.11239

Khader F, Mueller-Franzes G, Arasteh ST, Han T, Haarburger C, Schulze-Hagen M, et al. Medical Diffusion: Denoising Diffusion Probabilistic Models for 3D Medical Image Generation [Internet]. arXiv; 2023 [cité 21 déc 2023]. Disponible sur: http://arxiv.org/abs/2211.03364

Verdier H, Laurent F, Cassé A, Vestergaard CL, Specht CG, Masson JB. Simulation-based inference for non-parametric statistical comparison of biomolecule dynamics. PLoS Comput Biol. 2023 Feb 2;19(2):e1010088. doi: 10.1371/journal.pcbi.1010088. PMID: 36730436; PMCID: PMC9928078.