PhD in Explainable Machine Learning for Nuclear Safeguards

PhD in Explainable Machine Learning for Nuclear Safeguards

Quick Decision

This Chalmers employed doctorate connects explainable AI with spent nuclear fuel verification: making neural-network results understandable to inspectors, then using those explanations to improve model reliability. It offers four funded research years, potentially five calendar years with 20% departmental duties, starting at SEK 35,725 gross/month. Deadline: 23 November 2026. A useful opportunity for computational researchers interested in high-consequence human decisions, with data access, validation and supervision arrangements still to clarify. The exact post is REF2026-0498, job 15136. Official vacancy.

Research Problem and Doctoral Work

The advertised problem is the gap between strong ANN predictions and explanations that support a safeguards inspector's judgment about consistency with declared information. A correct classification alone does not explain its reasoning or delimit conditions where it should be trusted. The doctoral researcher will explore, test and evaluate an XAI framework, integrate different explanation methods and semantic layers that combine predictions into actionable explanations, and use the findings to guide ANN architecture, training and robustness. Scientific concepts, journal papers and conference communication are expected; no specific XAI algorithm, dataset size or deployed inspection system is promised.

The intellectual challenge is partly human: an explanation useful for developers may be unclear to inspectors, while a persuasive explanation may fail to reflect the model's actual behaviour. NIST's general framework distinguishes evidence, user meaning, explanation accuracy and knowledge limits. These provide possible evaluation questions, not an advertised prescribed study protocol. Useful early work could compare generic explanation stability and faithfulness, assess uncertainty and define an evaluation process with intended users. Such work should use institutionally approved data and the team's established governance. General explanation principles.

The vacancy establishes computational research, not a guaranteed placement at an inspection agency or direct access to nuclear facilities. Clarify the balance of simulation and measured data, what can be published, access approvals, available computing, and who will assess explanation usefulness. Generic coding evidence on an open, non-sensitive classification dataset can demonstrate transferable skill without reproducing safeguards operating details.

School, Subject and Training

Chalmers University of Technology in Gothenburg, Sweden, is the employer and degree institution. The host is Reactor Physics, Modelling and Safety, in Subatomic, High Energy and Plasma Physics, Department of Physics and Astronomy. Registration is in the Graduate School of Nuclear Engineering. The use of ML does not make this a Computer Science doctoral award. No partner university award is advertised. Host department; Research group.

QS places Chalmers joint 174th overall in 2027, following 165th in 2026; this is institutional context rather than a nuclear-engineering/XAI subject ranking. More direct subject evidence comes from the group's safeguards, diagnostics and computational modelling activity, and Vinai's sustained publication record. Dated QS institutional ranking.

The published Nuclear Engineering syllabus requires 240 credits: 60 coursework and 180 thesis, with an English thesis normally preferred and an approved defence. A licentiate can be an intermediate qualification; otherwise a midway seminar marks that level. At least two supervisors are appointed; the advertisement names Vinai as contact but does not specify the complete team. Generic and Transferable Skills training includes 7.5 mandatory credits, with access up to 15, in communication, pedagogy, leadership and ethics. Doctoral education has no programme fees, and subject graduate schools oversee courses and follow-up. Confirm the individual study plan and precise degree title at admission. Subject syllabus; Employment and training; Graduate schools and courses.

Supervisor, Papers and Research Setting

Paolo Vinai is Associate Professor, with work in numerical reactor modelling, diagnostics and safeguards ML. The official portfolio records Swedish Radiation Safety Authority collaborations; an earlier 2021–22 project shows continuity but does not identify the new position's sponsor. Group resources encompass computational modelling and signal analysis. Prior partnerships do not automatically provide this student's data or an external secondment. Vinai research portfolio; Earlier verification research.

The following discussions focus on each study’s aims and conclusions and their relevance to ML reliability; they do not establish external validation of the reported models:

Al-Dbissi et al., 2023: confidence and validation

The paper in Annals of Nuclear Energy 193, 110005, studies ANN interpretation of simulated verification signals. Its conclusion highlights confident errors, label imbalance and overlap of configurations across validation folds. This motivates careful separation of predictive confidence from correctness. 2023 study.

Al-Dbissi et al., 2024: generalisation limits

The paper in volume 204, 110536, compares physical information supplied to models and tests generalisation to held-out configurations. Coverage limits remain; simulation performance does not establish behaviour on actual inspection measurements. This supports asking how new-project data and evaluation will address transfer. 2024 study.

Al-Dbissi et al., 2026: model comparison

The paper in volume 237, 112465, evaluates CNN pattern analysis against a feed-forward baseline. Aggregate metrics conceal case variation and limited training diversity. It shows an active method-development line, not an established XAI system or inspector user trial. 2026 study.

Together they provide a credible reason to study trustworthy explanations. They do not fix the new student's architecture or establish that explanations alone resolve validation problems.

Pay, Living Costs and Eligibility

The exact vacancy states funding from the start and SEK 35,725/month starting salary, effective 1 May 2026. This is taxable employment pay. The retained employee terms give annual leave of 28 days, increasing to 31 from age 30 and 35 from age 40; occupational pension and insurance apply under employee terms. Teaching/other duties up to 20% can extend the four-year appointment to five. Obtain the individual salary progression and duty allocation before accepting. Employee conditions.

Chalmers' February 2026 student budget estimates SEK 14,360–18,360/month, including SEK 4,000–8,000 rent. Its SEK 3,000 monthly course-literature allowance is unusually large for a computational PhD, so treat the total as a reference requiring a personal budget, rather than guaranteed spending or savings. Gross salary cannot be directly compared with net expenses. Housing guidance reserves guarantees for specified student categories; doctoral employees are not listed, so ask HR about their separate support and deposits. Dated living-cost reference; Housing categories. National rules confirm doctoral application and tuition fees are waived. Doctoral fee rules.

Required qualifications are a relevant 120- or 60-credit Master's, with an explicit vacancy footnote accepting a four-year Bachelor's earned outside Sweden. Fields are Physics, Mathematics, Computer Science or equivalent. Applicants need strong English, computational methods and programming such as Python, plus experience in at least one of ML, DL, XAI, reactor physics or safeguards. All five are not required. Teamwork, responsibility and analytical skills matter; equivalence and English evidence are assessed individually.

Physical presence is required throughout. A valid residence permit must be available at the study start or admission may be withdrawn. The Migration Agency's doctoral route accepts salary as maintenance; the saved 2026 minimum is SEK 10,656/month, a permit threshold rather than a Gothenburg budget. Start dates, permit duration and processing are individual decisions. Official doctoral residence route.

Application, Outlook and Questions

The deadline is 23 November 2026 (Europe/Stockholm); the recruitment-system timestamp is 24 November 2026, 06:59:59 UTC+08:00. Current acceptance is unconfirmed; check the official portal before applying. Submit English PDF files: CV and personal letter with a brief introduction and motivation; Bachelor's thesis and transcripts; Master's thesis if available. Each file can be up to 40 MB; ZIP, email and incomplete applications are not accepted. References are requested after interview; a background check may occur. Questions: vinai@chalmers.se.

Career and transferable skills

The combination of interpretable ML, uncertainty evaluation, scientific computing and regulated-domain communication can transfer to trustworthy-AI and energy research. A 2026 JRC/ESARDA publication confirms active AI support research while retaining human safeguards judgment; its text-analysis task is different and offers no placement or salary statistics for this PhD. Current field activity.

Questions before applying, interviewing and accepting

Before applying, confirm foreign-degree equivalence and documents. At interview, ask about measured/simulated data, explanation users, validation criteria, supervision division and publications. Before accepting, obtain contract/pay progression, teaching allocation, housing/relocation, permit timing, approved data/compute access and completion arrangements. The opportunity's principal advantage is funded interdisciplinary training; the remaining decision depends on access and supervision quality.

Evidence and Sources

The vacancy lists 23 November 2026 as the deadline; current acceptance, individual permit status and employment terms remain unconfirmed. Incomplete author bylines remain unresolved.

Exact post, institution and training

• Doctoral vacancy REF2026-0498/job 15136
• Department of Physics and Astronomy
• Nuclear Engineering doctoral syllabus
• Chalmers doctoral employment and GTS training
• Chalmers graduate schools and courses
• QS Chalmers institutional profile 2027

Supervisor and research evidence

• Paolo Vinai research portfolio
• Reactor physics, modelling and safety group
• Earlier SSM-funded safeguards research project
• Identification of diversions in spent PWR fuel assemblies by PDET signatures using Artificial Neural Networks (ANNs)
• On the use of neutron flux gradient with ANNs for the detection of diverted spent nuclear fuel
• Reconstructing patterns of missing fuel pins in PWR spent nuclear fuel assemblies via CNNs
• NISTIR 8312 Four Principles of Explainable AI
• JRC/ESARDA 2026 AI support in safeguards information analysis

Pay, fees and practical arrangements

• Chalmers employee benefits and conditions
• Chalmers Gothenburg living-cost guide 2026
• Chalmers housing guidance
• Swedish Migration Agency doctoral residence route
• Swedish University Admissions doctoral fee rules