PhD in generative AI for biological imaging | Tissue images and spatial molecular data

PhD in generative AI for biological imaging | Tissue images and spatial molecular data

Quick Decision

Promising with material uncertainties. This is a three-year, full-time CNRS doctoral contract at IBENS in Paris 05, focused on generative machine learning for joint analysis of tissue images and spatial molecular measurements. The advertised methods are flow matching and diffusion; the work is computational and includes large-image analysis, model evaluation and collaboration with biology teams. It should be most relevant to applicants interested in applied mathematics, machine learning and computational imaging. The vacancy states €2,300 gross per month and lists leave and transport benefits.

Confirm that applications are still being accepted before preparing a submission. The CNRS detail page displayed New when checked on 8 November 2026, while the captured listing record showed EXPIRED. The employer page sets a deadline of 27 November 2026 at 23:59 but does not state a timezone; the captured record labels it UTC. The CNRS regional vacancies list dates this offer 6 November 2026, while the captured record says 7 November 2026. Other important unknowns are the required prior degree and language levels, application materials, the doctoral school and degree-awarding route, dataset access, and the evaluation plan.

Research Problem and Doctoral Work

What Is Unsolved and Why It Matters

The advertised project asks how deep generative models can support joint analysis of tissue images and spatial molecular measurements. It is not just an image-classification project: the vacancy calls for analysis across image and spatial molecular data and says the methods will be applied to biological questions with partner teams. The exact scientific question, data type and intended output are not published, so applicants should ask what the model is expected to generate or represent and how the two modalities will be linked.

This problem sits within an active computational-imaging field. Recent work explores cross-modal representations and generative approaches that relate image structure to spatial molecular measurements, including GenST, STFlow and iStar. These papers show that the area is moving quickly, but they do not establish that the same data or modeling choices are planned here. The important practical issue is evaluation: a model can produce plausible outputs without reliably preserving task-relevant information. The image-analysis validation recommendations in Metrics Reloaded argue for choosing metrics around the actual problem rather than relying on generic scores.

Work, Methods, and Potential Outputs

The CNRS notice specifies that the doctoral researcher will:

• design, implement and evaluate deep generative models, explicitly naming flow matching and diffusion;
• process and analyse large imaging datasets;
• work with biology collaborators on biological questions; and
• publish and present results at conferences.

The notice describes the position as computer-based and lists access to GPU computing resources. It does not publish the datasets, tissue or measurement platform, partner commitments, model baselines, quantitative metrics, validation plan or expected thesis outputs. It also does not detail the amount of programming, data preparation or partner-facing work relative to method development.

Possible questions to discuss—not promised work packages—include how the two modalities will be aligned, what happens when paired data are limited or incomplete, how uncertainty and robustness will be assessed, and which task-specific criteria will be used to compare models. A 2025 preprint on whole-slide flow matching and a 2026 workshop paper on generative cross-modal prediction illustrate nearby directions; the project notice does not say that either approach will be adopted.

School, Programme and Field Context

The CNRS vacancy identifies Institut de biologie de l'École normale supérieure (IBENS) in Paris 05 as the workplace and lists a fixed-term doctoral contract. ENS-PSL is the host academic environment; CNRS is the recruiting employer named on the vacancy. The general ENS framework describes a three-year target doctorate prepared in an ENS-PSL laboratory under a PSL co-accredited doctoral school, while IBENS PhD guidance says prospective candidates should contact the team leader and identifies PSL's ED Life Sciences as the institute's affiliated doctoral school. These are general institutional routes, not confirmation of this vacancy's enrolment arrangement. The offer does not name its registering institution, doctoral school, formal specialty or degree-awarding institution. Confirm each in writing; do not assume that co-supervision or the host institution determines the award.

PSL reports 28th overall in the QS World University Rankings 2026 and 57th overall in the THE World University Rankings 2027. These are broad university rankings, not subject rankings or measures of IBENS, this research group or the doctorate. No subject-specific ranking for computational imaging was verified. PSL's ED Sciences du Vivant page includes computational biology and biomathematics within a broad life-sciences remit, which provides relevant doctoral-school context if this route applies.

Supervisor and Lab Assessment

Supervisor Research Directions

The vacancy names Auguste Genovesio's image-analysis and AI team at IBENS, with co-supervision from N. Spassky's IBENS team. It describes an interdisciplinary environment spanning AI, applied mathematics and biology, with GPU resources. Genovesio's official profile says that his group develops methods for large-scale image and data analysis at the interface of applied mathematics, statistics, computer science and molecular/cell biology. The 2026 team listing includes Genovesio, CNRS researcher Clémence Réda, PhD students and an engineer.

The public pages do not specify the day-to-day supervisor, how the two teams will divide supervision, GPU allocation, data engineering support or project-specific compute capacity. The listing's reference to a group roster should not be read as a promise of staffing on this thesis.

Recent Important Publications

• PhenDiff (MICCAI 2024) uses conditional diffusion to identify image shifts by translating real images between conditions. With Genovesio among the authors, it is a direct group precedent for generative image analysis. Its task differs from the advertised joint spatial-molecular analysis.
• ChAda-ViT (CVPR 2024) studies channel-adaptive attention for joint representation learning from microscopy images. It is relevant to computational handling of heterogeneous imaging inputs, but is not the generative method specified by this vacancy.
• DiffEx (2025 preprint) proposes diffusion-based visual explanations for image classifiers. It raises useful questions about interpretability and evaluation, but remains a preprint on a narrower task.
• Cross-modal knowledge distillation (ICML 2025) uses weakly paired morphology and transcriptomic representations. It is relevant to cross-modal learning and limited pairing, though it is not the same as spatially resolved joint generation.

Together, these papers make the advertised topic a plausible continuation of the group’s computational-imaging activity. They are context for a conversation with the supervisors, not evidence of the thesis's exact model, data, biological scope or planned publication.

Lab, Resources, and Collaboration Environment

The CNRS notice says the work will involve partner biology teams and access to GPU computing. Its statement that the role is computer-based suggests that the advertised emphasis is computational rather than hands-on laboratory work; it does not remove the need to coordinate with domain collaborators or clarify data governance. Ask what GPUs and support are available, whether data access is already arranged, who handles approvals or permissions, and what scientific/technical support the co-supervising teams provide.

Funding, Application, and Living Cost

Contract, Salary, and Fees

The official vacancy lists a full-time, 36-month fixed-term doctoral contract, starting 1 January 2027, at €2,300 gross per month. A French government order sets the doctoral-contract minimum at €2,300 gross per month from 1 January 2026, so the listed salary matches that floor. Neither source states take-home pay. The CNRS page also lists 44 annual leave/RTT days, 75% transport-cost coverage, and a sustainable-mobility allowance up to €300. Remote-work practice and compensation are mentioned without further terms. Confirm benefit eligibility and conditions.

The employer page does not name a project-specific funder. The captured source record says the position is not funded under an EU framework programme, but does not identify another funder or budget line. Treat the funding source beyond the stated CNRS contract as unconfirmed.

For context only, Dauphine-PSL lists a €398 doctoral fee for 2026–27, excluding the €105 CVEC shown by the government portal. This is not confirmation that the project registers at Dauphine or that these exact fees apply. Ask the team which institution handles registration and what fees or exemptions apply.

Paris Cost Context

Crous Paris's 2026–27 tariff table, effective 1 September 2026, lists selected studio examples of roughly €416–630 per month. These specific student-residence prices are not a private-market average and do not guarantee eligibility or availability. A PSL student budget guide from 2024 gives older rent and food estimates; it is useful only as historical context, not as a current complete budget for an employed doctoral researcher.

The selected Crous rent examples equal approximately 18–27% of the stated gross salary before deductions. This ratio is not an affordability estimate. Net pay, rent outside Crous, utilities, insurance, transport and other personal costs remain to be checked.

Applicant Fit and Preparation

Required Background and Skills

The CNRS vacancy does not state a required degree level, years of experience, French threshold or English threshold. A captured record lists “PhD or equivalent” and “French: Basic,” but the direct employer notice does not substantiate these as entry requirements. Do not treat them as confirmed eligibility criteria; ask the recruitment contact for the formal application rules and required documents.

Preferred Experience and Suggested Preparation

The task description points toward preparation in applied mathematics or statistics, programming, machine/deep learning, computational imaging and handling large datasets. Prior work with diffusion or flow-matching models, GPU workflows, reproducible code and model evaluation would help an applicant explain how they could approach the advertised work. These are analyst-suggested preparation areas, not official requirements. The recent team papers offer relevant discussion points, especially around generative image analysis, heterogeneous imaging inputs, cross-modal alignment and interpretability.

The profile is less aligned with applicants seeking primarily hands-on laboratory practice: the vacancy describes computer-based work, though collaboration with biology teams is explicitly part of the role. APEC's data-scientist and image/signal-processing engineer profiles point to adjacent skill-transfer directions. They do not establish placement rates, hiring demand or salary outcomes for graduates of this project.

Next Actions and Questions to Verify

Before Applying

1. Check the CNRS application page directly to confirm that it is still accepting applications. It displayed New on 8 November, but a captured source record showed EXPIRED.
2. Confirm whether 27 November 2026 at 23:59 is local time or another timezone. The employer page does not say; the captured record labels it UTC.
3. Ask which prior degree or qualification is required, what language levels are expected, which documents to submit, and what the selection timeline is. The public vacancy does not answer these questions.

At Interview

1. Which doctoral school and specialty will register the thesis? Which institution handles enrolment, and which institution awards the degree?
2. Who will be the day-to-day supervisor, and how are responsibilities divided between the Genovesio and Spassky teams?
3. Which datasets and permissions are already secured? How will model outputs be validated, what baselines will be used, and what does “joint analysis” mean for this project's concrete question?
4. What GPU capacity, software support and data-engineering support will be allocated?

Before Accepting an Offer

1. Confirm the contract terms, net monthly pay, leave/RTT calculation, transport and mobility-benefit conditions, and any remote-work arrangements.
2. Confirm the doctoral fee, CVEC liability or exemption and the registration institution.
3. Get the degree-awarding route and doctoral specialty in writing, and discuss realistic housing options and timing.

Evidence and Sources

Vacancy and doctoral route

• CNRS doctoral vacancy
• CNRS Île-de-France vacancies list
• IBENS PhD entry guidance
• PSL ED Sciences du Vivant
• ENS doctoral studies
• ENS Biology Department contact and address

Institution, team, and research

• PSL QS World University Rankings 2026 summary
• PSL THE World University Rankings 2027 summary
• Auguste Genovesio — IBENS profile
• Genovesio team members — IBENS
• PhenDiff — MICCAI 2024
• ChAda-ViT — CVPR 2024
• DiffEx — 2025 preprint
• Cross-modal knowledge distillation — ICML 2025
• iStar — PubMed record
• SpatialEx — Nature Methods
• GenST — PMLR 2026 workshop paper
• STFlow — 2025 preprint
• Metrics Reloaded — image-analysis validation recommendations
• Stem — 2025 preprint

Contract, costs, and career context

• French doctoral remuneration order
• Dauphine-PSL tuition fees
• Official CVEC portal
• Crous Paris 2026–27 residence tariffs
• PSL student budget guide (2024)
• APEC data scientist career profile
• APEC signal-processing engineer career profile