CS PhD Roundup — August 17, 2026: 5 active leads, no new opening cleared the filter

What changed this week

The week ending August 17, 2026 produced no newly dated opening that cleared both filters: a named advisor and a verifiable NeurIPS, ICML, ICLR, CVPR, or ACL publication from 2022–2025. Five active decision units remain, representing six seats because the EPFL entry covers two postdocs. One Inria postdoc stays watch-only because its official advert still does not show a posting date.

The deadline queue has narrowed to one immediate cluster: both UvA roles close on August 20. SDU's page remains live, but its current job-information block exposes a portal timestamp rather than a plainly labeled local deadline. EPFL and ISTA continue to accept rolling applications.

Quick triage

Direction	Advisor and institution	Position	Current status	First action
Mechanism-informed world models	Efstratios Gavves, University of Amsterdam	PhD	Closes August 20 1	Prepare the month-dated CV, motivation letter, project or publication list, and two references; apply through UvA
Physics-informed world models	Efstratios Gavves, University of Amsterdam	Postdoc	Closes August 20 2	Lead with physics-informed learning, simulation, robotics, or scientific-ML work; apply through UvA
Biological foundation models	Charlotte Bunne, EPFL AIMM	Two postdocs	Rolling; Fall 2026 preferred 3	Send the one-page research statement, CV, and two references through the EPFL application route
Neuromorphic reinforcement learning	Melih Kandemir, University of Southern Denmark	Postdoc	Live; portal block shows September 1, 2026, 05:59; timezone is not labeled 4	Open the portal and confirm the displayed cutoff before submitting the CV, publication list, and PhD evidence
Trustworthy and neuro-symbolic ML	Christoph Lampert, ISTA	Multiple postdocs	Rolling until filled; last updated June 2026 5	Email one PDF with the CV, research statement, publications, transcripts, and three references to `chl(at)ist.ac.at`

World models and embodied AI

University of Amsterdam: mechanism-informed world-model PhD

Fit. Efstratios Gavves is a University of Amsterdam professor whose current profile centers on physical AI, world models, causal vision, mechanistic learning, and biomedical applications. The profile lists Dream to Manipulate: Compositional World Models Empowering Robot Imitation Learning with Imagination and CaPo: Cooperative Plan Optimization for Efficient Embodied Multi-Agent Cooperation at ICLR 2025, Mechanistic Neural Networks for Scientific Machine Learning at ICML 2024, and Latent Field Discovery in Interacting Dynamical Systems with Neural Fields at NeurIPS 2023. The strongest match is an applicant working on causal or representation learning, generative modeling, robotics, dynamical systems, or embodied agents. 6

Role and application. This is a four-year PhD with an initial 18-month contract that can be extended after a satisfactory evaluation. The project asks how models can represent events, causes, and temporal relationships instead of only predicting the next frame. It also covers controllable agents for robot learning, embodied systems, and biomedical dynamics. The gross salary is €3,059–€3,881 per month, plus an 8% holiday allowance and an 8.3% year-end allowance. The preferred start is as soon as possible or by discussion. 1

The vacancy asks for an MSc in AI, computer science, electrical engineering, physics, mathematics, or a related field; strong machine-learning and programming skills; and professional English. The application packet needs a detailed CV with month-level education and work dates, a motivation letter, a publication or project list, and two references. The UvA application page closes on August 20, 2026. 1

Lab signal. The advert places the project in CyPhAI and describes work across robot learning, biomedical modeling, and physical-world data, with real robots, substantial GPU compute, reading groups, seminars, and collaborations. It also expects teaching support. These are conditions stated by the vacancy; the page does not provide independent student testimonials. 1

University of Amsterdam: physics-informed world-model postdoc

Fit. This role is also with Gavves and CyPhAI. It suits a researcher with a completed or near-completed PhD in machine learning, computer vision, robotics, physics, applied mathematics, or a related field, especially with physics-informed learning, simulation, dynamical systems, robot learning, or scientific ML experience. Gavves's ICLR 2025, ICML 2024, and NeurIPS 2023 work provides the advisor signal. 6

Role and application. The 21-month postdoc studies physics-informed and physics-grounded world models, including partial observation, faster simulation, transfer across embodiments and domains, and agents that act inside learned models. The gross salary is €3,546–€5,538 per month, plus the same holiday and year-end allowances. The preferred start is as soon as possible or by discussion. Applicants should bring a strong publication record, substantial experimental and programming experience, and the ability to work independently and mentor PhD and MSc students. 2

Apply by August 20, 2026 with a detailed CV, motivation letter, publication or project list, and two references through the UvA postdoc application page. The vacancy places the work in CyPhAI and VISLab, alongside robot learning, biomedical modeling, and mechanism-informed generative models. 2

Biological foundation models

EPFL AIMM: two postdocs with Charlotte Bunne

Fit. Charlotte Bunne's CV lists Modeling Complex System Dynamics with Flow Matching Across Time and Conditions at ICLR 2025 and Supervised Training of Conditional Monge Maps at NeurIPS 2022. Her recent work connects optimal transport, generative modeling, and biological or biomedical data. The opening fits a researcher who can work across large-scale model development and biological data rather than a general-purpose LLM profile. 7

Role and application. EPFL's Artificial Intelligence in Molecular Medicine Lab is recruiting two postdoctoral researchers for biological foundation-model architecture, pretraining objectives, tokenization, scalable learning, multimodal biological data, agentic interfaces, and applications in drug discovery and biomedicine. A completed or near-completed PhD in machine learning, computer science, computational biology, or a related field is expected, along with large-scale AI development, biological-data familiarity, and a strong publication record. Applications are reviewed on a rolling basis; Fall 2026 is preferred, but the posting gives no fixed deadline or salary figure. Send a CV, one-page research-interest statement, and two references through the EPFL application route. 3

Lab signal. The advert promises dedicated access to the Swiss National Supercomputing Centre and the Swiss AI Initiative, direct pharmaceutical collaborations, mentorship toward an independent academic or industry career, and an international environment. The page does not provide a lab-size figure or independent student feedback. 3

Reinforcement learning

University of Southern Denmark: neuromorphic reinforcement-learning postdoc

Fit. Melih Kandemir's publication page lists Adaptive Ensemble Aggregation for Actor-Critics and ObjectRL: An Object-Oriented Reinforcement Learning Codebase from 2025, along with 2025 work on evidential proximal policy optimization. The research direction is deep reinforcement learning under uncertainty, probabilistic modeling, non-stationary environments, and bio-inspired architectures. 8

Role and application. This is a two-year postdoc in the ADIN Lab, starting no earlier than October 1, 2026. The project combines deep RL, probabilistic modeling, and spiking neural networks to study sample- and energy-efficient control in high-dimensional, non-stationary systems. The advert requires a PhD in computer science, mathematics, statistics, or theoretical physics; at least two first-author papers at flagship ML venues such as NeurIPS, ICML, or ICLR; strong RL theory; and Python with PyTorch or JAX. Salary follows the Danish collective agreement, with no figure published. 4

The current page's Job Info block shows Apply Before 09/01/2026, 05:59 AM, but it does not label the timezone in the retrieved text. Open the SDU vacancy and application portal and confirm the portal's displayed cutoff before uploading a CV with employment dates, a full publication list, and a PhD diploma or supervisor statement. 4

Lab signal. ADIN sits in SDU's Data Science and Statistics Group, which the advert describes as a joint mathematics-and-computer-science environment. The postdoc will collaborate on top-tier publications and perform standard teaching-assistant duties. 4

Trustworthy and neuro-symbolic ML

ISTA: Lampert group postdocs

Fit. Christoph Lampert's group lists Fast Rate Bounds for Multi-Task and Meta-Learning with Different Sample Sizes and Neural Collapse is Globally Optimal in Deep Regularized ResNets and Transformers at NeurIPS 2025; ASIDE: Architectural Separation of Instructions and Data in Language Models at ICLR 2025; Banded Square Root Matrix Factorization for Differentially Private Model Training at NeurIPS 2024; and 1-Lipschitz Layers Compared: Memory, Speed, and Certifiable Robustness at CVPR 2024. The opening is a match for work on privacy, fairness, robustness, federated or continual learning, LLM security, and symbolic-statistical ML. 9

Role and application. Multiple one- or two-year postdocs are available with flexible starts and no fixed deadline. Applicants should hold a PhD in machine learning or a related area, with strong mathematics, programming, analytical reasoning, and first-tier publications. Send one PDF containing a CV, research statement, publication list, transcripts, and contact details for three references to Prof. Christoph Lampert at chl(at)ist.ac.at. The page says applications remain open until the positions are filled. 5

Lab signal. The group advertises a 100% research role with no mandatory teaching or administrative duties, state-of-the-art compute, and funding for equipment and conference travel. ISTA's working language is English, and voluntary PhD-level teaching is encouraged rather than required. These are explicit group conditions; the page does not provide independent student feedback. 5

Watch-only: Inria Grenoble text-to-video postdoc

Stéphane Lathuilière supervises a two-year postdoc on editing and conditional generation with text-to-video models. The work covers diffusion models, transformers, video editing, reinforcement-learning post-training, temporal consistency, semantic fidelity, and efficient training. His DBLP record lists Ask and Remember: A Questions-Only Replay Strategy for Continual Visual Question Answering and Di[M]O: Distilling Masked Diffusion Models Into One-Step Generator at ICCV 2025, plus 2025 work on machine unlearning. 10

The opening starts October 1, 2026, runs for two years, pays €2,788 gross per month, and closes August 20. It requires a PhD or equivalent and skills in speech processing, computer vision, machine learning, and programming; LLMs, SpeechLMs, reinforcement learning, or robotics are useful but optional. Apply through the Inria online route. The official advert still exposes no posting date, so confirm that it is live before applying. 11

New adverts that stayed outside the queue

Oxford mathematical and computational foundations of AI, placed August 14. The official advert offers two 24-month postdocs across structured and geometric models, low-dimensional learning, implicit regularization, and stochastic-control RL, with a September 14 deadline. It does not name an advisor or investigator in the public advert, so it does not clear the named-advisor filter. 12
Imperial AI for turbomachinery design, placed August 13. The postdoc would combine machine learning and computational fluid dynamics under Dr Teng Cao, start on November 1, and close on September 6. The advert gives a strong engineering brief but does not establish the required recent target-venue publication record for the named supervisor, so it stays outside the main roundup. 13

Deadline watch

August 20: UvA mechanism-informed world-model PhD and physics-informed world-model postdoc.
August 20, watch-only: Inria text-to-video postdoc; the official page still does not expose its posting date.
Portal timestamp September 1, 2026, 05:59: SDU neuromorphic-RL postdoc; confirm the timezone in the application system.
Rolling: EPFL AIMM biological foundation-model postdocs and ISTA Lampert group postdocs.

The action order is straightforward: handle the two UvA applications first, verify the Inria page before spending time on its packet, then confirm the SDU portal cutoff. For SDU, the two-first-author-paper threshold is a hard screen. For EPFL and ISTA, rolling review makes an application-ready packet more useful than waiting for a nominal cutoff.