CS PhD Roundup - July 20, 2026

What changed this week

The clearest new lead is at the University of Basel: Aurelien Lucchi and Ilija Bogunovic are recruiting for a PhD or postdoctoral project on LLM training, reasoning, and reliable generalization. The posting is dated July 20, 2026. Its body encourages applications before July 31, while the page metadata shows August 18 as the application deadline and also says the search remains open until a suitable candidate is found. Treat July 31 as the practical target and confirm the portal status when applying. 1

The University of Twente deadline is still live, but it is a same-day action: the listing now gives July 20, 2026 at 23:59 Europe/Amsterdam. The position is a fully funded four-year PhD with Nicola Strisciuglio on compositional learning for data-efficient vision foundation models. 2

Inria WILLOW's July 17 deadline has passed and its official page now returns a not-found response. The ETH Zurich oncology postdoc's stated collection window ended July 19; no extension is stated in the listing retrieved for this issue, so it is not carried as an active opening. 3 4

Quick triage

Research direction	Advisor and institution	Type	Deadline or status	First action
LLM training and reasoning	Aurelien Lucchi and Ilija Bogunovic, University of Basel	PhD or postdoc	July 31 recommended; August 18 shown in page metadata; no fixed deadline stated in body	Prepare application now
Computer vision and compositional learning	Nicola Strisciuglio, University of Twente	PhD	July 20, 23:59 Europe/Amsterdam	Apply today
Trustworthy and human-centred AI	Hans Gersbach, ETH Zurich, with an unnamed computer science professor	PhD	July 31, rolling review	Confirm the CS co-supervisor first
NLP and LLM metalinguistic ability	Benoit Sagot and Rachel Bawden, Inria Paris	PhD	July 30	Apply through Inria
Mechanism-informed world models	Efstratios Gavves, University of Amsterdam	PhD	August 20	Apply if world models or causal learning fit
Physics-informed world models	Efstratios Gavves, University of Amsterdam	Postdoc	August 20	Apply if you can lead an independent line
Biological foundation models	Charlotte Bunne, EPFL AIMM Lab	Two postdocs	Rolling	Send the requested materials
Neuromorphic reinforcement learning	Melih Kandemir, University of Southern Denmark	Postdoc	Online while the job remains listed; start no earlier than October 1	Check the live portal
Trustworthy ML and neuro-symbolic AI	Christoph H. Lampert, ISTA	Multiple postdocs	Rolling	Email one PDF to the PI

Language models and NLP

University of Basel: LLM training, reasoning, and adaptation

Advisor and fit: Aurelien Lucchi is an associate professor at Basel whose public profile lists deep learning, stochastic optimization, fine-tuning, and reasoning as core interests. The OMLS group reports an ICLR 2025 paper, Adaptive Methods through the Lens of SDEs, and a NeurIPS 2025 paper, Enhancing Optimizer Stability: Momentum Adaptation of The NGN Step-size. Ilija Bogunovic's public publication page lists Group Robust Preference Optimization in Reward-free RLHF at NeurIPS 2024 and A Robust Phased Elimination Algorithm for Corruption-Tolerant Gaussian Process Bandits at NeurIPS 2022. The combined fit is strongest for applicants interested in LLM post-training, reasoning, reinforcement learning, optimization, and robust decision-making. 5 6 7

Opening: The posting accepts either a PhD candidate or a postdoc. The PhD is funded for four years; the postdoc starts with two years and may extend to four. The expected start is around October 2026, with later starts possible. The work covers pre-training and post-training, mathematical reasoning, decision-making, domain adaptation, fine-tuning, and reinforcement learning. Applicants should have a strong ML background, good implementation skills, and the ability to run large-scale experiments. 1

Application: Send two PDF files: a short motivation statement of up to 150 words, CV, transcripts, recommendation contacts, and a thesis or publication where applicable. The page says applications are encouraged before July 31, but its metadata shows August 18, 2026 at 23:59 Europe/Zurich and the body says there is no fixed deadline. Use July 31 as the applicant-side target, and rely on the linked page for the live portal state.

Lab signal: The listing explicitly promises close supervision, access to national and international collaborators, conference travel, and a diverse working environment. It does not publish a lab size or student testimonials.

Apply to the University of Basel LLM training opening

Inria Paris ALMANaCH: metalinguistic abilities of LLMs

Advisor and fit: Benoit Sagot leads Inria's ALMAnaCH team, and Rachel Bawden is a co-supervisor on the listing. Their public records include Sagot's 2023 EMNLP paper Generative Spoken Language Model based on continuous representations and Bawden's 2023 EAMT paper Investigating the Translation Performance of a Large Multilingual Language Model: the Case of BLOOM. The relevant fit is NLP, multilingual and low-resource translation, evaluation, and language models. 8 9

Opening: This is a three-year PhD at Inria Paris. The project asks how LLMs represent, explain, and apply explicit linguistic knowledge, with directions covering measurement of metalinguistic knowledge, in-context acquisition from grammars and dictionaries, and training methods for turning declarative linguistic knowledge into usable skill. A master's degree in computer science, ML, NLP, or computational linguistics is required, along with deep learning experience and a serious interest in language and linguistics. 10

Application: The deadline is July 30. Apply through the Inria recruitment platform with a CV, one-page motivation letter, and degree and grade documents. English or French applications are accepted. The listing does not state a salary or a lab-size figure.

Apply to the Inria ALMANaCH PhD

Vision, world models, and embodied AI

University of Twente: compositional learning for vision foundation models

Advisor and fit: Nicola Strisciuglio is an associate professor at Twente working in machine learning and computer vision. His public publication record includes Data-efficient large scale place recognition with graded similarity supervision at CVPR 2023, What do neural networks learn in image classification? A frequency shortcut perspective at ICCV 2023, and Fourier-basis Functions to Bridge Augmentation Gap at CVPR 2024. 11

Opening: One fully funded four-year PhD will study compositional learning for data-efficient vision foundation models. The project targets object-attribute compositionality, bias detection and machine unlearning, and perceptual and conceptual priors that reduce dependence on huge datasets. A master's degree in computer science, computer engineering, or mathematics with an AI, ML, or computer vision focus is expected, together with strong mathematics and programming. 2

Application: The deadline is July 20 at 23:59 Europe/Amsterdam. Submit a two-page CV, one-page cover letter, academic transcript, and an English-language test where required. Salary runs from EUR 3,059 per month in year one to EUR 3,881 in year four. The position includes a graduate-school training plan, conference travel budget, and partial work-from-home flexibility.

Lab signal: The listing asks for a good team spirit in an interdisciplinary and international environment. It also names structured supervision and training through the Twente Graduate School. The deadline leaves little room for an exploratory email, so apply first and contact the advisor with a focused question only if needed.

Apply to the University of Twente PhD

University of Amsterdam: mechanism-informed world models and agents

Advisor and fit: Efstratios Gavves leads the CyPhAI lab within the University of Amsterdam's VISLab. His public profile lists world models, robot learning, causal vision, mechanisms, safety, and biomedical Physical AI. Recent listed work includes Dream to Manipulate and CaPo 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. 12

PhD opening: The four-year PhD will develop world models that expose events, causes, and temporal relationships rather than only predicting the next frame. The vacancy spans robot manipulation, embodied agents, and biomedical dynamics. It asks for an MSc in a relevant quantitative field, ML and programming experience, and interest in world models, causal or representation learning, RL, dynamical systems, or robotics. The salary is EUR 3,059 to EUR 3,881 gross per month. 13

Postdoc opening: The related postdoc will lead research on physics-informed and physics-grounded world models, including partial observation, faster simulation, cross-embodiment transfer, and agents that act inside the models. It is a 21-month contract with possible extension, paid at EUR 3,546 to EUR 5,538 gross per month, plus an 8% holiday allowance and 8.3% year-end allowance. A PhD and an independent publication record are expected. 14

Application and lab signal: Both vacancies close on August 20. The pages mention reading groups, seminars, collaborations, and teaching. The PhD requires a CV, motivation letter, publications or representative work where available, and two referee contacts. The postdoc asks for the same core materials and expects the researcher to mentor PhD and MSc students.

Apply to the UvA PhD or apply to the UvA postdoc

Biology, reinforcement learning, and trustworthy ML

EPFL AIMM Lab: biological foundation models

Advisor and fit: Charlotte Bunne leads EPFL's Artificial Intelligence in Molecular Medicine Lab. The lab's research page places its work at the intersection of AI and human biology, with foundation models, generative AI, agentic systems, and multimodal modeling. Recent listed work includes Modeling Complex System Dynamics with Flow Matching Across Time and Conditions at ICLR 2025 and MTBBench: A Multimodal Sequential Clinical Decision-Making Benchmark in Oncology at NeurIPS 2025. 15 16

Opening: EPFL is recruiting two postdoctoral researchers with large-scale AI experience and a biology background. The roles cover foundation-model architecture, multimodal learning, agentic interfaces, and translation toward drug discovery and biomedicine. A PhD in ML, computer science, computational biology, or a related field is required, along with large-scale pre-training or fine-tuning experience and a strong ML or life-science publication record. The expected start is Fall 2026, with later starts possible by arrangement. 17

Application and lab signal: Applications are reviewed on a rolling basis. Send a CV, a one-page research-interest statement, and two references. The listing promises dedicated access to CSCS and the Swiss AI Initiative, collaboration with pharmaceutical partners, and mentorship toward an independent academic or industry career. It does not state a fixed salary.

Apply to the EPFL AIMM postdoc posting

University of Southern Denmark: neuromorphic reinforcement learning

Advisor and fit: Melih Kandemir's public publication page lists Deep Actor-Critics with Tight Risk Certificates at NeurIPS 2025, Deterministic Uncertainty Propagation for Improved Model-Based Offline Reinforcement Learning at NeurIPS 2024, and Learning Interacting Dynamical Systems with Latent Gaussian Process ODEs at NeurIPS 2022. His profile centers on reinforcement learning, Bayesian and PAC-Bayes methods, uncertainty, bandits, and dynamic systems. 18

Opening: The SDU ADIN Lab postdoc focuses on neuromorphic reinforcement learning, spiking-neural-network architectures, robust and adaptive control in non-stationary or uncertain environments, and theory such as convergence, stability, and sample complexity. The posting describes a 10-to-12-month fixed-term position in one section and says the role is to start no earlier than October 1, 2026 for a period of two years in another. That contract-length inconsistency should be clarified with the department. 19

Application: The job page says the role remains open as long as it is online rather than giving a calendar deadline. Apply through the live portal with a CV, publication list, and PhD diploma or supervisor confirmation. Salary follows the Danish collective agreement; no amount is stated. The listing notes possible teaching-assistant duties and a cross-disciplinary data-science environment.

Apply to the SDU postdoc

ISTA: trustworthy ML, transfer learning, and neuro-symbolic AI

Advisor and fit: Christoph H. Lampert's group lists recent work at NeurIPS 2025, ICML 2025, ICLR 2025, and CVPR 2024, including ASIDE: Architectural Separation of Instructions and Data in Language Models, Logic Gate Neural Networks are Good for Verification, and Improved Modelling of Federated Datasets using Mixtures-of-Dirichlet-Multinomials. The open topics cover trustworthy ML, privacy and fairness, robustness, federated and continual learning, LLM security, and the interface between symbolic AI and statistical ML. 20 21

Opening and application: Multiple one- or two-year postdocs are available at ISTA, with flexible starting dates and no fixed deadline. Send one PDF containing a CV, research statement, publication list, transcripts, and three reference contacts to Christoph Lampert. Candidates should hold a PhD in ML or a related area and have a strong first-tier publication record.

Lab signal: The page explicitly says the role has 100% research with no mandatory teaching or administrative duties. Voluntary PhD-level teaching is supported, and the group cites state-of-the-art compute, equipment and conference-travel funding. The posting says applications are considered until the positions are filled.

Apply to the ISTA postdoc opening

One caveated lead to watch

ETH Zurich: trustworthy and human-centred AI development

This PhD position at KOF Swiss Economic Institute focuses on reliability, security, explainability, human-AI interaction, LLM evaluation, retrieval-augmented workflows, and high-stakes systems such as trustworthy AI voting assistants. The listed supervisor is Hans Gersbach, with a computer science professor named only by role. That supervisor structure is the reason this remains a caveated lead rather than a clean qualifying entry for the roundup. 22

The listing gives July 31 as the deadline, with rolling review, and asks for a CV, short cover letter, full transcript, and two or three recommendation letters. The start date is August or September 2026. Apply only after confirming who the computer science co-supervisor is and how the doctorate will be administered.

Read the ETH Zurich PhD listing

Deadline watch

July 20: Twente compositionality PhD, 23:59 Europe/Amsterdam.
July 30: Inria ALMANaCH PhD.
July 31: ETH Zurich trustworthy AI PhD; Basel's posting also recommends submitting by this date.
August 20: Both UvA world-model openings.
Rolling: EPFL AIMM and ISTA Lampert group.
Online while listed: SDU neuromorphic RL postdoc.

The expired Inria WILLOW and ETH oncology items are no longer in the active pool. The former has disappeared from the official jobs page; the latter's stated application window ended July 19 without a published extension in the retrieved listing.