CS PhD Roundup — September 21, 2026: a Physical AI PhD at Aarhus, and three calls closing by September 30

CS PhD Roundup — September 21, 2026: a Physical AI PhD at Aarhus, and three calls closing by September 30

A newly posted Physical AI PhD at Aarhus and three verified calls closing by September 30, one of them today, give applicants a short dated action list.

The September 15–21, 2026 window produced one newly posted PhD opening: a fully funded Physical AI position in the Adaptive & Agentic AI lab at Aarhus University, supervised by Behzad Bozorgtabar. 1 Two postdoc calls advertised earlier stay open and close on September 30, and the ELLIS Institute Finland's autumn recruitment closes today. 2
A role reaches the queue below only when its named supervisor has a recent paper at NeurIPS, ICML, ICLR, CVPR or ACL. The leads that failed that test, and the reason for each, sit in their own section at the end.

Quick triage

InstitutionPosition & unitKey adviser(s)DirectionDeadlineDirect application
Aarhus University (Denmark)Fully funded 3-year PhD, Adaptive & Agentic AI (A3) Lab, Dept. of Electrical and Computer EngineeringBehzad Bozorgtabar (main supervisor), Qi Zhang (co-supervisor)Physical AI: vision-language-action models, world models for planning, adaptation and edge intelligence1 Nov 2026, 23:59 CETAarhus PhD portal
University of Copenhagen (Denmark)2-year postdoc, CopeNLU, Dept. of Computer Science and Pioneer Centre for AIIsabelle Augenstein, Pepa AtanasovaMechanistic interpretability of large language models30 Sep 2026, 23:59 CETUniversity of Copenhagen job portal
Imperial College London (United Kingdom)Research Associate (postdoc), 2 years, AToM project, Dept. of ComputingEdoardo M. PontiLatent tokenization, sparse attention and permanent memory in foundation models30 Sep 2026Imperial job posting
ELLIS Institute Finland (Finland)Multiple PhD (4 years) and postdoc (3 years) posts across Finnish universitiesJörg Tiedemann, Samuel Kaski, Aki Vehtari, Francesco CroceFoundation models, multilingual NLP, probabilistic machine learning21 Sep 2026, 15:59 (UTC-05:00) — closes todayCall page · PhD portal · Postdoc portal

New this week: Physical AI at Aarhus

Behzad Bozorgtabar, Aarhus University

Fit. The A3 Lab builds trustworthy adaptive and agentic AI, and this project connects multimodal perception, reasoning and action with predictive learning and edge intelligence. 3 Three directions are named. Vision-language-action models turn visual observations and language instructions into robot behaviour, including learning from demonstrations and generalising to unfamiliar tasks or environments. World models predict how the physical world responds to an action, which is what supports planning and learning from interaction; the advert suggests video, demonstrations and simulation as sources, and transferring knowledge across robot configurations. Adaptation and edge intelligence covers reliable behaviour under changing environments and resource limits, including test-time and continual adaptation, uncertainty-aware decision-making and efficient inference. The stated goal is original research for NeurIPS, ICML, ICLR, CVPR, ICCV, CoRL, RSS and ICRA. 1
Bozorgtabar started at Aarhus University as an associate professor in February 2026, directs the A3 Lab, and holds a guest professorship at EPFL and a core membership of ELLIS. 4 His recent papers include LATA: Laplacian-Assisted Transductive Adaptation for Conformal Uncertainty in Medical VLMs at CVPR 2026, accepted for an oral presentation in roughly the top 1% of papers; ReservoirTTA and UniViT at NeurIPS 2025; A Simple Framework For Open-Vocabulary Zero-Shot Segmentation at ICLR 2025; and CrIBO at ICLR 2024, a spotlight paper in the top 5%. 4
Role and application. The fellowship runs three years under Aarhus University's Graduate School of Technical Sciences, with Qi Zhang, a professor in the same department, as co-supervisor. The preferred start date is 1 January 2027, and the application deadline is 1 November 2026 at 23:59 CET. Salary and terms follow the applicable collective agreement, and the advert publishes no figure. 1
Applicants need a master's degree of 120 ECTS, completed by enrolment, in computer science, electrical or computer engineering, robotics, machine learning or a related field. The advert asks for strong foundations in machine learning, linear algebra, probability and optimisation, and for strong Python with PyTorch or a comparable framework. It also states that only applicants who demonstrate substantial hands-on experience implementing, training and evaluating deep-learning models will be considered, and that experience limited to running tutorials or calling pretrained-model APIs falls short. 1
The dossier is a one-page statement of interest naming the research directions that fit and your strongest technical project, a CV with a project portfolio and any publications, bachelor's and master's transcripts and diplomas, and the project description copied from the announcement and uploaded as a PDF. Shortlisted candidates attend a technical interview and should be ready to defend their own implementation, experimental design, results and limitations. Questions go to behzad@ece.au.dk for the main supervisor and qz@ece.au.dk for the co-supervisor. 1 Submissions go through the Aarhus application form.
Lab signal. The A3 Lab is a founding-stage group, and the advert describes a selective process in which shortlisting means the evaluation committee reads only the most relevant applications. 1 The lab sits inside Denmark's national AI ecosystem through the Pioneer Centre for AI and the DIGIT Centre, and the position carries collaboration and research visits across the ELLIS network. Industrial partners in robotics and autonomous systems mean the methods get tested on real problems rather than benchmarks alone. The advert describes an international, English-speaking research environment with a competitive Danish PhD salary. 3
The listing is new this week. Aarhus's own PhD portal carries it with a revision stamp of 16 September 2026, and the ELLIS network's job board listed it on 21 September. 13

Also open, advertised earlier

Both postdocs below were posted before this window and stay live. Their employers publish no listing date on the advert itself, so treat them as an active queue rather than as this week's news. Both close on 30 September.

Isabelle Augenstein and Pepa Atanasova, University of Copenhagen

Fit. The postdoc works on how large language models acquire, integrate and reason over evolving, pluralistic knowledge, using mechanistic interpretability methods. 5 CopeNLU, the group that hosts the position, researches interpretability, explainability and bias detection, with applications in fact checking and cross-cultural learning. 5
Augenstein leads the group with Atanasova and holds an ERC Starting Grant for the project Explainable and Robust Automatic Fact Checking. 6 The group placed seven papers at ACL 2026 and seven at EMNLP 2026 across its two most recent acceptance rounds. 6 One of those ACL papers, on explaining the sources of uncertainty in automated fact-checking, is in the anthology. 7
Role and application. The appointment runs two years from 1 February 2027 or as close to that date as possible, with an extension contingent on funding. Applications close on 30 September 2026 at 23:59 CET, and the advert writes CET while Denmark stays on summer time until 25 October, so the cut-off may land an hour earlier than the label suggests. Interviews are expected in the week of 19–23 October 2026. Monthly salary starts at DKK 39,900, about EUR 5,300 at August 2026 levels, with an additional 18.07% employer pension contribution; applicants holding a PhD who have lived outside Denmark for the past ten years may qualify for tax reductions. 5
Candidates need a PhD in computer science or a related field and good written and oral English. The advert assesses qualifications on previous publications and relevant experience, and describes a background in natural language processing, machine learning or explainable AI as ideal. The dossier is unusually long: a cover letter, a research plan covering current and future work, a CV, master's and PhD diplomas, a complete publication list, reprints of three particularly relevant papers, and the names and contact details of three referees. Enquiries go to augenstein@di.ku.dk. 5
Lab signal. CopeNLU sits in the Department of Computer Science and the Pioneer Centre for AI in central Copenhagen, and the advert cites CSRankings placing the Danish capital's NLP environment second in Europe. The postdoc takes on a small amount of teaching or supervision. 5

Edoardo M. Ponti, Imperial College London

Fit. The AToM project, Adaptive Tokenization and Memory in Foundation Models, is an ERC Starting Grant that attacks the cost of long sequences in transformer models. Four directions are in scope: latent tokenization that segments raw, modality-agnostic data into learned hierarchical units; permanent adaptive-size memory with hierarchical sparse attention; retrofitting state-of-the-art models and releasing the recipes, data and checkpoints; and inference-time hyper-scaling for reasoning, agentic workflows and long-horizon world modelling. 8
The group has already shipped retrofitted open-weight models with NVIDIA and with the Allen Institute for AI — Qwen3-8B-DMS-8x and Bolmo-7B — and reports speedups of up to eight times without accuracy loss. 8 Ponti is an associate professor in natural language processing in the Department of Computing, and his group holds best paper and SAC highlight awards from ACL and EMNLP. He delivered the NeurIPS 2024 tutorial on dynamic sparsity. 9
Role and application. The post is a Research Associate role, job number ENG04037, full time and fixed term for two years on the South Kensington campus, starting around winter 2026/27 with flexibility. The closing date is 30 September 2026. Salary runs from £50,733 to £59,484 per year, or £45,399 to £48,876 on the Research Assistant scale if the PhD is not yet awarded. 810
Applicants need a PhD in computer science, a strong publication record in machine learning or natural language processing venues, and hands-on experience training and evaluating large language models in PyTorch or JAX. Familiarity with CUDA or Triton kernels and with inference engines helps. The application is a full CV with a publication list plus a research statement of at most two pages, focused on the research questions in adaptive and efficient architectures that interest the candidate most. Imperial reserves the right to close the advert before the stated date if applications run high. 810
Lab signal. The project owns B200 GPUs and cloud credits, and the department and faculty add H200s. 8 The postdoc sets an independent research agenda inside the project, co-supervises PhD and MSc students, and can deliver tutorials at conferences and summer schools; the project funds two international conferences a year. Ponti is recruiting two fully funded PhD studentships on the same grant as well, through the department's central admissions with his name as proposed supervisor and a rolling deadline. 8

Closing today

The ELLIS Institute Finland closes its autumn call for doctoral students and postdocs today, 21 September 2026, at 23:59 EEST — 15:59 in this channel's timezone. The call covers multiple fully funded positions across Finnish universities and partner institutions, and last week's issue carried the terms in full. 2 Doctoral applicants submit through the ELLIS doctoral portal with reference R47606, and postdocs through the postdoc portal with reference R47586. 1112

Screened out this week

  • University of Göttingen: three openings under Tristan Manfred Stöber. A PhD on hippocampal representations in dynamic environments, a PhD on brain-inspired world models for navigation, and a postdoc on machine learning for epilepsy, all at TV-L E13 100%, with applications received by 24 September 2026 given full consideration and the posts staying open until filled. Stöber holds an ERC Starting Grant, and the world-models project aims squarely at model-based deep reinforcement learning. His publication record sits in neuroscience journals — Epilepsia, PLOS Computational Biology, Nature Communications, PNAS and Hippocampus — which lies outside the five venues this channel screens on. 13
  • Radboud University: PhD in physics-informed generative AI for synthetic energy data. An NWO project position inside the SHARE consortium, working with operational grid data from the distribution operator Alliander, supervised by Yuliya Shapovalova and Tom Heskes, with a deadline of 25 October 2026 and a page last modified on 18 September 2026. The scan found no qualifying 2022–2025 paper at NeurIPS, ICML, ICLR, CVPR or ACL for either adviser; both publish in statistics, energy and machine-learning journals. 1415
  • MPI for Biological Cybernetics, Tübingen: PhD positions in foundation models of human cognition. The CORE COGNITION lab's lead, Marcel Binz, clears the venue bar with Turning large language models into cognitive models at ICLR 2024, and the ELLIS job board lists a deadline of 1 October 2026 for fully funded, four-year E13 TVöD posts. 1617 The institute's own vacancy list currently shows a layer-fMRI PhD and student assistant roles, which leaves these positions unconfirmed from the employer's page. 18
  • VU Amsterdam: postdoc in causal machine learning and reinforcement learning. Posted 28 August 2026 with a 27 September 2026 deadline, three years, embedded in the Quantitative Data Analytics group and co-run with Amsterdam UMC, Erasmus MC and the University of Groningen. The group's lead, Mark Hoogendoorn, publishes in health machine learning and evolutionary computation, outside the five venues. 19
  • Empa and EPFL: postdoc in tabular foundation models for building and district energy systems. The role sits between Empa's Urban Energy Systems Laboratory and EPFL's IMOS laboratory under Olga Fink, with a 24 September 2026 deadline and immediate start possible. Empa's own job portal carries the advert without a posting date, which keeps it out of this week's window. 20

Deadline watch

  1. ELLIS Institute Finland — today, 21 September 2026, 15:59 (UTC-05:00). Finish the two-page motivation letter naming your preferred supervisors, confirm your referees are reachable, and upload through the doctoral or postdoc portal. 2
  2. University of Copenhagen — 30 September 2026, 23:59 CET. This dossier carries the most pieces: cover letter, research plan, publication list, three paper reprints and three referees. Ask for reprint permissions and referee confirmations now, before the interview window opens in the week of 19 October. 5
  3. Imperial College London — 30 September 2026. Write the two-page research statement against the AToM directions rather than your own thesis topic, and submit well before the closing date in case the advert closes early. 8
  4. Aarhus University — 1 November 2026, 23:59 CET. Copy the project description from the announcement into a PDF, prepare a one-page statement that names your strongest technical project, and be ready to defend your own implementation and experimental design in the technical interview. 1

References

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20

This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.

Related content