Postdoc symmetries and their breaking in deep learning

Laplace Expert Ltd; Netherlands; Remote; EUR 41,000 - 64,000; Full time

Job summary

The University of Amsterdam invites applications for a postdoctoral researcher to study symmetries and their breaking in deep learning, supported by Simons funding. The aim is to connect physical notions of symmetry and universality with neural network behavior and robustness.

The appointment is an independent research position with potential teaching and project duties; candidates should propose a concrete prior project and a credible plan to test claims and compare explanations in AI and

Qualifications

• Researchers with experience in computer science, applied mathematics, implementation skills and empirical evaluation.
• Appointment length and division of research, teaching and project responsibilities must be checked against the employer’s full instructions.
• In preparing an application, the candidate should use a concrete prior project to show why the advertised topic fits their skills and identify any important method that still needs to be learned.

Responsibilities

• Independent research reading and development of a concrete plan to study symmetries and their breaking in deep learning.
• Test and compare explanations for neural network behavior using computational models, software systems and algorithm evaluation, with clear baselines and uncertainty assessment.

Job description

Postdoc symmetries and their breaking in deep learning

University of Amsterdam

Netherlands

Apply by 08 Dec 2026

About This Opportunity

The Simons-supported postdoc studies physical notions of symmetry and universality as explanations for neural network behaviour.

Opportunity Overview and ContextThe Simons-supported postdoc studies physical notions of symmetry and universality as explanations for neural network behaviour. From an applicant’s perspective, the key decision is whether this specific research problem and the host’s disciplinary setting support a coherent next step. This is an independent research reading, not a reproduction of the original vacancy copy.

Research Questions, Methodology and EvidenceThe central subject is postdoc symmetries and their breaking in deep learning. For computing, artificial intelligence and networks, a useful analysis asks how the underlying claims would be tested and how alternative explanations would be separated. Relevant methods to discuss with the research team could include computational models; software systems; algorithm evaluation; these are methodological considerations inferred from the field, not a claim that every technique is officially mandatory. A credible research plan would make explicit how the work contributes to turning computational advances into falsifiable claims about robustness, generalization and system performance, with defensible baselines and a clear account of uncertainty.Candidate Profile and Applicant FitThis direction is likely to appeal to researchers with experience in computer science, applied mathematics, implementation skills and empirical evaluation. A prospective postdoc should assess whether existing publications and research methods can support an independent contribution on this precise topic. Appointment length and the actual division of research, teaching and project responsibilities need to be checked against the employer’s full instructions. In preparing an application, the candidate should use a concrete prior project to show why the advertised topic fits their skills and identify any important method that still needs to be learned.

Compensation, Eligibility and How to ApplyA published application date was independently located: 2026-12-08 (calendar date; no closing hour inferred). A source-backed monetary range is EUR 3,706–5,760; its exact payment period, FTE basis and applicability must be checked before financial planning unless explicitly detailed in the source. Degree equivalence, language scores, work authorization, relocation, applicant-specific conditions, research supervision and required documents must be checked on the institution’s own application instructions. Where an official vacancy-specific submission form has not been independently verified, the record deliberately leaves the official application destination unfilled.

Potential Challenges and Overall Opportunity AssessmentThe main research risk concerns reproducibility, robustness under distribution shifts and unclear downstream requirements. Applicants should assess whether they prefer the problems implied by postdoc symmetries and their breaking in deep learning and can show a credible route to evaluating results, rather than relying on the title alone. Laplace interprets this as a potentially relevant postdoctoral research opening for those with a focused methodological match; actual selection odds and recruitment conditions remain with the institution.

computational models software systems algorithm evaluation

Level Postdoc

Field Computing, artificial intelligence and networks