Paid PhD Position at Oldenburg: Applied Cryptography, Privacy and Trustworthy AI for Healthcare

Paid PhD Position at Oldenburg: Applied Cryptography, Privacy and Trustworthy AI for Healthcare

Quick Decision

This is a 100% E13 TV-L doctoral research employment position connecting cryptographic privacy with trustworthy machine learning and healthcare infrastructure. It is particularly relevant to Computer Science or Mathematics graduates who can document both cybersecurity and machine-learning topics in their studies, alongside strong programming. The main advantage is a concrete privacy-computation team within a new healthcare collaboration, supported by relevant recent papers and a full employment fraction.

A strong scientific fit if the methods and application setting align. The deadline is 8 November 2026. The captured listing gives 23:59 (Europe/Berlin), while the university employer page is date-only. As of 6 November 2026, its snapshot displayed “STATUS: EXPIRED”; a 25 October 2026 UTC check had found it inviting applications. These dated signals do not establish current acceptance; contact the application mailbox before submitting. The contract is full-time and ends 30 September 2030; a later start does not automatically extend it. Applications require a separate editable form and one merged PDF. The notice does not state a nationality restriction; foreign-qualification recognition, doctoral admission, employment checks and immigration permission remain separate decisions. This is an evidence-based interpretation of the published criteria, not an individual admission guarantee. Employer vacancy notice

Research Problem and Doctoral Work

Hospitals, healthcare providers and home-care environments need to collaborate without exposing sensitive medical information. Keeping data local does not alone establish privacy; encrypting computation does not alone establish robustness, fairness or useful explanations. The research challenge is to define meaningful protections and build systems whose privacy, utility and practical costs can be evaluated together.

The project develops secure, trustworthy and privacy-preserving infrastructure for healthcare data processing and communication. Possible directions include homomorphic encryption, secure multi-party computation, differential privacy and secure collaborative machine learning, alongside robustness, explainability, fairness and auditability. These are a menu of group research directions, not a requirement that one thesis solve every property and technique simultaneously. Official research scope and PATAI group

The research directions protect or evaluate different things:

Direction	Main object	A useful comparison question
Homomorphic encryption and MPC	Computation and collaboration without exposing raw information	Who holds keys, which parties are trusted, and what are the computation/communication costs?
Differential privacy	Protection associated with released outputs	What is released, and are the trust assumptions and utility measures comparable?
Secure collaborative ML	Information exposed by parameters and training intermediates	What can be learned beyond data remaining local?
Robustness, explainability, fairness and auditability	Model behaviour and evidence that can be inspected	Which property is selected and how does its evaluation interact with privacy?

These distinctions help scope a thesis; they are not an additional required protocol. Related papers show different limits: encrypted primitives have scale/memory trade-offs, selective protection needs a defined leakage goal, and locating a time-series change differs from testing whether a change exists. USENIX paper DBSec paper PoPETs paper

The full advert adds duties beyond the shorter project summary: publish research at leading conferences; co-supervise students; collaborate on joint papers with PATAI, SSI and other incubator groups; consolidate interdisciplinary project results in activities and workshops; and co-develop the incubator's infrastructure. The recruitment introduction also names leading journal outputs and academic/non-academic collaboration. It advertises full conference-travel and international-networking support, whose actual budget, approvals and covered items need confirmation. Ask how these responsibilities fit around an independent thesis.

A useful initial research plan would select a concrete healthcare collaboration and threat model, specify what participants and released outputs may reveal, and evaluate a limited technical contribution against appropriate baselines. That is analyst interpretation, not a prescribed methodology. Accuracy, privacy assumptions, communication, computation, numerical approximation and deployability may trade off. Medical-data access, ethics and clinical partnership must be arranged; the advert does not provide a dataset licence or guarantee clinical benefit.

NIST's AI Risk Management Framework offers relevant evaluation context, while the European Commission's EHDS programme illustrates the practical significance of secure health-data use and exchange. The framework is voluntary, and EHDS implementation is phased rather than all obligations applying immediately. Neither establishes that this doctorate's future system will meet all regulatory requirements. NIST context and EHDS official overview

School, Programme and Field Context

The University of Oldenburg is the employer and university doctoral setting. The position sits in the Department of Computing Science, School II, within Privacy-Aware Trustworthy AI (PATAI), embedded in Safety-Security-Interaction (SSI). Connected Health Nordwest brings together university medicine, computing, OFFIS and DFKI around healthcare AI. Their participation does not imply a joint degree or a degree from OFFIS, DFKI or Twente. The exact doctoral title, current regulations and admission sequence should be confirmed with the department. Vacancy, PATAI and SSI

A reliable year-specific current overall or Computer Science ranking was not obtained: the QS and THE primary profiles could not be substantively accessed. No numeric or elite-status claim is made. Specialist strength is better supported here by the named team's recent cryptographic papers, SSI's broader publication activity and the healthcare collaboration. A new PATAI group should also be distinguished from a long-established laboratory with a documented record of its own completed doctoral students.

The contract is full-time, suitable for part-time arrangements by agreement, and fixed to September 2030. Doctoral qualification is explicitly supported. OLTECH currently provides voluntary membership, scientific/interdisciplinary skills, data management, research integrity and career development. Its 2022 special-admission document describes particular structured programmes, including non-EU degree recognition by the relevant board; it does not establish that this post requires a particular 30-ECTS curriculum or a three-year degree schedule. The OLTECH travel-grant banner is limited to Schools V/VI and does not establish a travel entitlement for this School II post. Employment recruitment and formal doctoral admission should not be merged into one automatic approval. OLTECH voluntary training provision, 2022 programme regulation and scope and doctoral counselling

Supervisor and Lab Assessment

Dr Federico Mazzone leads PATAI and is an SSI postdoctoral researcher. His official biography records Mathematics/Cryptography training at Trento, a Twente PhD in October 2025 and appointment at Oldenburg from June 2026. Twente's defence notice identifies his thesis on fully homomorphic encryption for privacy-preserving collaborative machine learning. He is therefore an early-career group leader with directly relevant work, rather than someone whose own long-term PhD completion record can already be evaluated. Official biography and Twente defence notice

Prof Andreas Peter, SSI head, provides the other named source of guidance. His official academic trajectory covers a Darmstadt Computer Science doctorate, professorial work at Twente and the Oldenburg chair since 2022. The combination offers early-career technical leadership and experienced professorial support; actual meeting frequency, feedback and responsibility for formal supervision are unpublished. Peter biography

Related papers: contributions and limits

1. Federico Mazzone, Maarten Everts, Florian Hahn and Andreas Peter, USENIX Security 2025: Efficient Ranking, Order Statistics, and Sorting under CKKS. Homomorphic matrix re-encoding and SIMD parallel comparisons reduce sequential comparison depth for ranking, order statistics and sorting. A comparison depth of at most two does not mean the entire encrypted computation takes two operations. This provides reusable privacy-preserving ML building blocks while exposing time/memory trade-offs. Official paper and final PDF

Selected final-paper evaluation and limitations show quadratic space complexity: splitting larger vectors across ciphertexts increases comparisons, and gains diminish when parallel resources are exhausted. The evaluation uses bounded random vectors and a specified server, not real medical data. Approximate comparison also trades precision against runtime for closely spaced values. It supports the team's technical experience, not universal superiority at every hospital data scale.

2. Federico Mazzone, Ahmad Al Badawi, Yuriy Polyakov, Maarten Everts, Florian Hahn and Andreas Peter, DBSec 2025: Encrypt What Matters. The final abstract defines selective model protection relative to leakage through the eventual prediction interface. Deep models may require more encrypted components, and output-based black-box leakage remains. Its relevance is the need to define observable information and adversarial capabilities before promising efficiency, rather than assuming one encrypted layer always suffices. Institutional publication record Publisher abstract and version statement

The publisher explicitly calls the final chapter a significantly reworked version of the 2024 preprint. The final body was not obtained. Substantive selected reading instead covers that earlier SmartCryptNN version: privacy/performance depends on exposed layers and adversarial capabilities; its semi-honest proof does not cover arbitrary malicious deviations. The prototype emphasises relative comparisons and substantial communication costs. Earlier numerical results are not presented as final-chapter benchmarks. Earlier selected reading of the 2024 preprint

3. Federico Mazzone, Giorgio Micali and Massimiliano Pronesti, PoPETs 2026(2), pp. 127–142: Secure Change-Point Detection for Time Series under Homomorphic Encryption. Mazzone and Micali share first authorship. The formal proceedings record and final paper support this account. The method connects block statistics/ordinal patterns to encrypted comparisons, CUSUM and argmax to locate mean, variance or frequency changes. Formal proceedings Author preprint

Final section 2.1 assumes a single data provider who receives the result, and an honest-but-curious server; it does not cover disclosure of the decrypted result to another recipient. The evaluation includes synthetic data, an EEG excerpt, one participant's heartbeat-interval recording and network data. Accuracy close to a plaintext statistical baseline is not clinical validation. The roughly three-minute million-point result is server processing under specified hardware; client work, transmission, memory and ciphertext volume still matter. Crucially, final section 5 assumes that a change point is known to exist: multiple changes and full testing including the no-change null hypothesis remain open problems. FHE protects the computation view while DP constrains released outputs under different assumptions, so the experiments do not establish universal FHE superiority over DP.

Together these papers support healthcare privacy-infrastructure research, while not covering every advertised trustworthy-AI property. At interview, ask how an existing computational technique becomes a defensible thesis contribution and which data permissions, fallback public data, computing and software-maintenance support are available.

Funding, Application and Living Cost

The employer confirms 100% E13 TV-L, annual special payment, occupational pension and 30 days annual leave, with flexible hours, possible partial remote work and development support. This is salary-based employment, with deductions, rather than a tax-free stipend. The Lower Saxony official table gives E13, Step 1: €4,759.37 gross monthly, valid 1 April–31 December 2026. It is a dated reference, not confirmation that a particular applicant starts in Step 1 or receives that net amount. HR must assess recognised experience and the table applicable at the agreed start; the state publishes further adjustments from March 2027 and January 2028, so the 2026 table is not a rate fixed for the whole contract. Official pay table and state update schedule

Oldenburg's current exchange-student guidance estimates around €1,000 monthly, or €12,000 annualised, for local living costs, with room costs €200–€450 and food €250–€350. Its roughly €140 student health-insurance estimate is not a payroll quote for an employed doctoral researcher. After deriving net pay, do not add that student premium again without checking the actual insurance arrangement. Room availability, personal housing choices, deposits and relocation can materially change the budget. Comparing a gross salary with this student budget cannot establish savings. Current local cost guidance

The university distinguishes tuition from semester contributions: most study programmes do not charge tuition, but enrolled students pay contributions for transport, student services and administration. The doctoral employee's registration status and exact contribution remain to be confirmed; exchange-student fee figures should not be used as doctoral rates. The phrase fully funded does not prove that relocation, visa and every registration cost are covered. University fees guidance

If the employer confirms that applications are still accepted, send the application to applications.ssi@uni-oldenburg.de using “Secure and Trustworthy AI” as the subject. The notice lists 8 November 2026 at 23:59 (Europe/Berlin), but as of 6 November the page also displayed “STATUS: EXPIRED”; confirm acceptance immediately. The application requires two attachments:

1. The official form, completed fully and truthfully, as a separate editable file such as XLSX. The vacancy does not reproduce every form question; review the linked template before completing it. Official template

2. One merged PDF containing the CV, Bachelor's and Master's transcripts, and both degree certificates if available, otherwise proof of enrolment. The job still formally requires the completed academic degree; allowing enrolment evidence does not establish that completion can be deferred indefinitely.

The employer instructions list the application address and attachments. The notice lists a precise deadline while the page displayed an “EXPIRED” marker, ask applications.ssi@uni-oldenburg.de whether it will accept the submission and request confirmation of receipt. Contact Mazzone at federico.mazzone@uni-oldenburg.de and Peter at andreas.peter@uni-oldenburg.de for project questions. Full employer instructions

Applicant Fit and Preparation

Required: a completed Diplom/Master in Computer Science, Mathematics with an appropriate computing specialisation, or a comparable field; cybersecurity and machine learning, both interpreted broadly, documented as parts of the applicant's studies; very good programming; excellent written/spoken English; and willingness to pursue the doctorate. The explicit study-content requirement is stronger and more specific than simply expressing enthusiasm for privacy-preserving AI.

Advantageous: practical academic research, for example a relevant peer-reviewed paper; practical ML projects; strong mathematical understanding of ML; and knowledge of applied cryptography/PETs. A publication is an example of desirable research experience, not a universal admission prerequisite. The advert does not set a German-language condition or a named programming-language checklist. Official selection criteria

Analyst-suggested preparation: map security and ML coursework to the job; prepare one reproducible research/programming example that explains the threat model, utility and cost; and discuss why a particular healthcare collaboration needs the proposed protection. A generic AI portfolio without documented security studies, or pure cryptographic theory without the required ML studies and programming, would leave a material gap.

Career functions include cryptographic/privacy R&D, secure ML engineering, health-data infrastructure and trustworthy-AI evaluation. ENISA's researcher and architecture profiles provide a function-based comparison involving research, prototypes, security/privacy requirements and communication. NIST and EHDS show active evaluation and health-data needs. They do not quantify vacancies, salaries or the group's graduate placement rate. ENISA role profiles

Next Actions and Questions to Verify

Before applying: obtain the editable template and assemble both degree stages' documentation; make the security/ML study evidence easy to locate. Confirm degree-completion timing, English evidence, foreign-qualification documents, deadline clock and employment/doctoral admission sequence. The university's equal-qualification preferences for women and disabled applicants are preferences, not an exclusive first-round restriction.

At interview: agree a manageable first thesis question, threat model, privacy/utility/cost evaluation and selected trustworthy-AI properties. Ask about health-data permissions, ethics, fallback public datasets, computing and software support, Mazzone/Peter mentoring, feedback and the student-co-supervision workload.

Before accepting: confirm start date and completion plan within the fixed September 2030 end; formal doctoral degree route and supervision agreement; E13 experience step and net-pay calculation; semester contributions, permits and arrival costs; conference/networking support and any project-infrastructure obligations. The shorter description's travel support should be translated into an actual budget and approval process.

Evidence and Sources

The recruitment summary is dated 23 October 2026. The employer page was updated 22 October and was checked on 25 October 2026 UTC, when it was still inviting applications. As of 6 November 2026, the page displayed “STATUS: EXPIRED” while the notice listed 8 November 2026 at 23:59 (Europe/Berlin). These dated records conflict; applicants should contact the application mailbox to confirm whether submissions are being accepted.

The paper discussion draws on selected final USENIX evaluation and limitations sections and selected final PoPETs threat-model, evaluation and conclusion sections. For DBSec, the available reading covers the final abstract and version statement plus selected reading of the substantially reworked 2024 antecedent; the final paper body was not reviewed. This is not a full audit of all article methods, proofs, code or benchmarks. The exact doctoral regulations, a reliable year-specific ranking and all application-form questions remain unconfirmed.

Vacancy and doctoral training

• S01 — Exact Oldenburg Trustworthy AI doctorate vacancy
• S02 — PATAI official research group scope
• S03 — SSI official research and recent academic activity
• S10 — OLTECH voluntary doctoral training
• S11 — 2022 OLTECH special admission reference and nonEU recognition
• S12 — Oldenburg doctorate counselling
• S20 — Mandatory editable application template identity
• EURAXESS vacancy listing

Supervisors and recent papers

• S04 — Federico Mazzone official biography
• S05 — Andreas Peter official cryptography professor biography
• S06 — Twente records Mazzone’s 2025 FHE doctorate
• S07 — USENIX 2025 encrypted ranking/order statistics/sorting
• S08 — DBSec 2025 selective model encryption for federated learning
• S09 — 2026 encrypted time-series change-point detection

Funding, costs and field careers

• S13 — Oldenburg local living-cost comparator
• S14 — Oldenburg tuition versus semester contribution
• S15 — Lower Saxony official TV-L 2026 pay table
• S16 — Lower Saxony official pay update schedule
• S17 — NIST trustworthy AI risk-management context
• S18 — European Commission health-data infrastructure transition
• S19 — ENISA cybersecurity researcher and architecture role profiles

Paper records and version information

• DBSec publisher abstract and version statement
• 2024 SmartCryptNN preprint
• PoPETs 2026 proceedings paper