CS PhD Roundup — July 27, 2026

What changed this week

Two openings have verifiable publication dates inside the July 20–27 window: an Oxford postdoc on interpretable and safe machine learning, and a University of Basel PhD or postdoc opening on LLM training and reasoning. The immediate deadline queue is Inria ALMANaCH on July 30, ETH Zurich and Basel on July 31, and Oxford on August 6. One strong Inria Grenoble video-generation postdoc is kept as a watch item because the official page does not expose a posting date, so it is not counted as a verified new opening this week. 1 2 3

Quick triage

Research direction	Advisor and institution	Type	Deadline or status	First action
Safe, interpretable ML, LLMs, RAG	Philip Torr, University of Oxford	Postdoc	August 6, 2026 at noon UK time	Apply early; resolve the contract wording
LLM training, reasoning, and adaptation	Aurelien Lucchi and Ilija Bogunovic, University of Basel	PhD or postdoc	July 31 recommended; August 18 shown in page metadata; rolling text also appears	Prepare the two-PDF application now
NLP and metalinguistic abilities	Benoit Sagot and Rachel Bawden, Inria Paris	PhD	July 30, 2026	Submit through Inria
Trustworthy and human-centred AI	Hans Gersbach and an unnamed CS professor, ETH Zurich	PhD	July 31, rolling until filled	Confirm the CS co-supervisor
Mechanism-informed world models	Efstratios Gavves, University of Amsterdam	PhD	August 20, 2026	Match your proposal to causal or embodied learning
Physics-informed world models	Efstratios Gavves, University of Amsterdam	Postdoc	August 20, 2026	Show a strong top-venue record and simulator experience
Biological foundation models	Charlotte Bunne, EPFL AIMM Lab	Two postdocs	Rolling; Fall 2026 preferred	Send the one-page research statement
Neuromorphic reinforcement learning	Melih Kandemir, University of Southern Denmark	Postdoc	August 31 at 23:59 local time; portal shows September 1 at 05:59	Prepare the publication list and PhD proof
Trustworthy ML, continual learning, neuro-symbolic AI	Christoph Lampert, ISTA	Multiple postdocs	Rolling, no fixed deadline	Email one PDF to the PI
Controllable text-to-video generation	Stephane Lathuiliere, Inria Grenoble	Postdoc, watch-only	August 20; posting date not exposed	Apply only if the date caveat is acceptable

New openings with a dated posting

University of Oxford: safe and interpretable machine learning

Advisor and fit: Philip Torr leads the Torr Vision Group. His group publication list includes Prompting a Pretrained Transformer Can Be a Universal Approximator at ICML 2024, When Do Prompting and Prefix-Tuning Work? A Theory of Capabilities and Limitations at ICLR 2024, and Universal In-Context Approximation By Prompting Fully Recurrent Models at NeurIPS 2024. The research fit is strongest for applicants working on LLMs, retrieval-augmented generation, continual learning, agentic systems, model interpretability, or AI safety. 4

Opening: The full-time Postdoctoral Research Assistant will work in the Torr Vision Group and the Department of Engineering Science on interpretability, safety, and alignment. The project covers LLMs and related architectures for generative tasks, continuous learning, indexing and retrieval, long-term-history RAG, and extraction of useful information from noisy multimodal observations. The advert was placed on July 23, 2026. It gives a one-year term in the body, although the header labels the contract type as permanent. Treat that inconsistency as an item to clarify before accepting an offer. 1

Application: The deadline is noon on August 6, 2026. Applicants should have a PhD or DPhil, or be close to completion, in machine learning. The application requires a covering letter or supporting statement with a short research-interest statement, a CV, and details for two referees. The salary is listed as £39,424–£47,779, Grade 7. Apply through the linked Oxford recruitment route. 1

Lab signal: The advert describes close collaboration with policymakers on empirical safety research, AI regulation, and governance. It also notes an Athena Swan Bronze award at the department level. No lab-size figure or student testimonials are published. 1

Apply to the Oxford Torr Vision postdoc

University of Basel: LLM training, reasoning, and adaptation

Advisor and fit: Aurelien Lucchi's public profile covers deep learning, stochastic optimization, fine-tuning, and reasoning. The OMLS group lists recent work including Adaptive Methods through the Lens of SDEs at ICLR 2025 and Enhancing Optimizer Stability: Momentum Adaptation of The NGN Step-size at NeurIPS 2025. Ilija Bogunovic's publication record includes Group Robust Preference Optimization in Reward-free RLHF at NeurIPS 2024 and a NeurIPS 2022 paper on corruption-tolerant Gaussian-process bandits. The combined fit is unusually broad: LLM post-training, mathematical reasoning, optimization, reinforcement learning, and robust decision-making. 5 6 7

Opening: The University of Basel accepts either a PhD candidate or a postdoc for work on LLM pre-training and post-training, reasoning, mathematical problem solving, decision-making, specialized-domain adaptation, fine-tuning, and reinforcement learning. 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 page was published on July 20, 2026. 2

Application: The page asks for two PDF files containing a motivation statement of no more than 150 words, a CV with publications if applicable, scanned bachelor's and master's transcripts, contact details for recommenders, and a thesis or publication where applicable. The body encourages applications before July 31, while page metadata shows August 18, 2026 at 23:59 Europe/Zurich and also says there is no fixed deadline. Use July 31 as the practical target and treat the later date as a portal-status check, not a reason to wait. 2

Lab signal: The advert promises close supervision, national and international collaboration, conference participation, and a diverse working environment. It does not publish a lab-size figure or independent student feedback. 2

Apply to the University of Basel LLM opening

NLP and language

Inria Paris ALMANaCH: metalinguistic abilities of LLMs

Advisor and fit: Benoit Sagot and Rachel Bawden co-supervise the Inria ALMANaCH PhD. Sagot's ACL Anthology record includes SpeechMatrix: A Large-Scale Mined Corpus of Multilingual Speech-to-Speech Translations and Tackling Ambiguity with Images: Improved Multimodal Machine Translation and Contrastive Evaluation, both at ACL 2023, as well as Data-Efficient French Language Modeling with CamemBERTa at Findings of ACL 2023. The fit is strongest for multilingual NLP, low-resource translation, language evaluation, and explicit linguistic knowledge. 8

Opening: This is a three-year Inria Paris PhD on how LLMs represent, explain, and use explicit linguistic knowledge. The research axes cover measurement of metalinguistic knowledge across phonology, morphology, syntax, and semantics; in-context use of grammars and dictionaries; and training or reinforcement-learning methods that turn declarative language knowledge into stable skills. The start date is October 1, 2026, and the salary is €2,300 gross per month. 9

Application: The deadline is July 30, 2026. Submit a CV, a one-page motivation letter, and degree and transcript documents through the Inria recruitment platform. English or French applications are accepted. 9

Lab signal: The listing names standard Inria benefits, flexible working arrangements, training, and a diversity-oriented recruitment policy. It does not publish a lab-size figure or student testimonials. A restricted-area security authorization may be required. 9

Apply to the Inria ALMANaCH PhD

Vision, world models, and embodied AI

University of Amsterdam: mechanism-informed world models

Advisor and fit: Efstratios Gavves leads the CyPhAI lab. His public profile lists 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. Applicants should be comfortable with world models, causal or representation learning, generative modeling, reinforcement learning, robotics, or dynamical systems. 10

Opening: The four-year PhD will develop models that expose mechanisms, events, causes, and temporal relationships rather than only predicting the next frame. The vacancy is open and closes on August 20, 2026. The salary runs from €3,059 in year one to €3,881 in year four, plus an 8% holiday allowance and an 8.3% year-end allowance. The preferred start is as soon as possible or by discussion. 11

Application: Submit a detailed CV with months for education and work experience, a motivation letter, a publication or project list where available, and two referees. The applicant should hold an MSc in AI, computer science, electrical engineering, physics, mathematics, or a related field, with solid ML and programming skills. 11

Lab signal: CyPhAI is part of VISLab and the UvA Informatics Institute. The vacancy names substantial GPU compute, real robots, biomedical and physical-world data, and collaborations with Toyota Research, Qualcomm, Elekta, and the Netherlands Cancer Institute. It also expects publishing at CVPR, ICLR, ICML, NeurIPS, or ECCV and assisting with teaching. 11

Apply to the UvA mechanism-informed world-model PhD

University of Amsterdam: physics-informed world models

Advisor and fit: The postdoc is also with Efstratios Gavves and CyPhAI. The strongest match is for researchers with publications in machine learning, computer vision, robotics, or scientific ML, plus experience with physics-informed learning, simulation, dynamical systems, robot learning, or world models. 12

Opening: The role studies physics-informed and physics-grounded world models, including partial observation, faster simulation, cross-embodiment transfer, and agents that act inside learned models. It closes on August 20, 2026 and pays €3,546–€5,538 gross per month, plus the same holiday and year-end allowances. The preferred start is as soon as possible or by discussion. 12

Application: Applicants should submit a detailed CV, motivation letter, publication or project list, and two references. A completed or near-completed PhD is required, along with strong programming and large-scale experimentation skills. The advert also expects the postdoc to guide PhD and MSc students. 12

Lab signal: The lab advertises GPU compute, real robots, biomedical and scientific data, and collaborations across industry and medicine. It is part of the ELLIS network. 12

Apply to the UvA physics-informed world-model 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 public research page lists 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. The fit is for researchers who can combine large-scale AI with biological data, multimodal learning, generative models, or agentic systems. 13

Opening: EPFL is recruiting two postdoctoral researchers for biological foundation-model work spanning 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 expected, along with large-scale pre-training or fine-tuning experience, biological background, and a strong top-tier publication record. The ideal start is Fall 2026, with later dates possible by arrangement. 14

Application: Applications are reviewed on a rolling basis. Send a CV, a one-page research-interest statement, and the names of two references. No fixed salary is stated. 14

Lab signal: The listing points to dedicated access to CSCS and the Swiss AI Initiative, collaboration with pharmaceutical partners, an international and collaborative environment, and mentorship toward an independent academic or industry research career. 14

Apply to the EPFL AIMM postdocs

University of Southern Denmark: neuromorphic reinforcement learning

Advisor and fit: Melih Kandemir's public publication record includes Deterministic Uncertainty Propagation for Improved Model-Based Offline Reinforcement Learning at NeurIPS 2024, Deep Actor-Critics with Tight Risk Certificates at NeurIPS 2025, and a NeurIPS 2022 paper on latent Gaussian-process ODEs. The role suits applicants with both RL theory and engineering ability, especially in uncertainty, non-stationarity, continuous control, or spiking neural networks. 15

Opening: The ADIN Lab postdoc develops sample- and energy-efficient reinforcement learning for high-dimensional dynamical systems, combining deep RL, probabilistic modeling, and spiking neural networks. It is full time for two years and starts no earlier than October 1, 2026. The official job record was posted on June 17 and remains active. 16

Application: The official portal's job-info block shows an apply-before value of September 1, 2026 at 05:59, while its page description gives August 31, 2026 at 23:59 local Danish time. The latter is the local deadline to use. Submit a CV with employment dates, a full publication list, and a PhD diploma or supervisor statement. Salary follows the Danish collective agreement; no figure is published. 16

Lab signal: IMADA brings mathematicians and computer scientists together, and the Data Science and Statistics group is described as a cross-field research platform. The postdoc will collaborate on top-tier publications and take on standard teaching-assistant duties. 16

Apply to the SDU ADIN Lab postdoc

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

Advisor and fit: Christoph Lampert's group lists recent work at NeurIPS, ICML, ICLR, and CVPR. Examples include ASIDE: Architectural Separation of Instructions and Data in Language Models at ICLR 2025, Federated Learning with Unlabeled Clients: Personalization Can Happen in Low Dimensions at ICML 2025, and 1-Lipschitz Layers Compared: Memory, Speed, and Certifiable Robustness at CVPR 2024. 17

Opening: Multiple one- or two-year postdocs are available at ISTA with flexible starting dates and no fixed deadline. Topics include trustworthy ML, privacy, fairness, robustness, federated and continual learning, LLM security, transfer learning, and the interface between symbolic and statistical ML. Applicants should hold a PhD in ML or a related area and have strong mathematics, programming, analytical reasoning, and first-tier publications. 18

Application: Send a single PDF to Christoph Lampert containing a CV, research statement, publication list, academic transcripts, and contact details for three references. Applications are considered until the positions are filled. 18

Lab signal: The post is explicitly 100% research, with no mandatory teaching or administrative duties. ISTA lists state-of-the-art compute, equipment and conference-travel funding, an English-language institute, and optional supported teaching at PhD level. 18

Apply to the ISTA Lampert group postdocs

Watch-only lead

Inria Grenoble: controllable text-to-video generation

Advisor and fit: Stéphane Lathuilière supervises the postdoctoral research visit in the ROBOTLEARN team. His profile places him in machine learning for computer vision, foundation-model adaptation, continual learning, image and video generation, and multimodal learning. DBLP lists his Collaborating Foundation Models for Domain Generalized Semantic Segmentation at CVPR 2024, An eye for an ear: zero-shot audio description leveraging an image captioner with audio-visual token distribution matching at NeurIPS 2024, and Zero-shot spatial layout conditioning for text-to-image diffusion models at CVPR 2023. 19 20

Opening: The two-year postdoc focuses on editing and conditional generation with text-to-video models. The research covers diffusion and transformer architectures, conditioning with image references, motion skeletons and semantic masks, video editing, temporal consistency, semantic fidelity, efficient training, and reinforcement-learning post-training. The start date is October 1, 2026, the deadline is August 20, and the salary is €2,788 gross per month. 3

Application: A PhD or equivalent is required, with a background in speech processing, computer vision, or machine learning and solid programming skills. Experience with LLMs, SpeechLMs, RL algorithms, or robotics is useful but not mandatory. Applications must be submitted online through Inria. 3

Lab signal: The advert describes Inria Grenoble's interdisciplinary research environment and expects publication and collaboration with the ROBOTLEARN team. The official page does not show a publication timestamp, so this lead is not included in the dated-new count. 3 19

Read the Inria Grenoble text-to-video postdoc

Deadline watch

July 30: Inria ALMANaCH PhD.
July 31: ETH Zurich trustworthy AI PhD; Basel's practical application target.
August 6: Oxford Torr Vision postdoc, noon UK time.
August 20: UvA world-model PhD and postdoc; Inria Grenoble text-to-video postdoc, watch-only.
August 31: SDU neuromorphic RL postdoc at 23:59 Denmark time; the portal renders this as September 1 at 05:59.
Rolling: EPFL AIMM biological foundation-model postdocs and ISTA Lampert group postdocs.

The University of Basel page is the only dated new listing in this issue with two different deadline signals: an applicant-facing July 31 recommendation, an August 18 metadata date, and rolling language in the body. Treat the earliest practical date as the one that matters.