Postdoc Position in Quantized Reinforcement Learning

Syddansk Universitet; Odense; On-site; DKK 420,000 - 540,000; Full time

Job summary

Syddansk Universitet invites applications for a two-year postdoctoral fellowship in quantized reinforcement learning at the ADIN Lab, SDU, Odense. The project targets how continuous sensory input is distilled into discrete tokens to guide control in complex environments.

The successful candidate will develop uncertainty-aware world models, provide theoretical guarantees, and benchmark across state-of-the-art simulation suites, collaborating toward high-impact publications in top venues.

Qualifications

• PhD required in a relevant field (CS/Math/Stats/Theoretical Physics).
• At least two first-author papers at flagship ML venues (NeurIPS/ICML/ICLR/AISTATS).
• Strong background in statistical learning theory and RL foundations.
• Excellent command of English, both written and spoken.

Responsibilities

• Advance research on quantized reinforcement learning and embodied discretization.
• Develop scalable prototypes and benchmarks in high-dimensional simulations.
• Bridge theory with practical algorithms; publish in top ML venues.
• Fulfill standard teaching assistant duties as part of the residency.

Skills

Python

Reinforcement Learning

PyTorch/JAX

Mathematical Maturity

English Communication

Education

PhD in Computer Science/Mathematics/Statistics/Theoretical Physics

Tools

Git

DL Frameworks

Job description

The SDU Adaptive Intelligence Lab (ADIN Lab) (https://adinlab.github.io/ ) located under the Data Science and Statistics Section of the Department of Mathematics and Computer Science (IMADA) at the University of Southern Denmark invites applications for a postdoctoral research fellowship position within the field of quantized reinforcement learning to be filled earliest by 1 November 2026 for a period of two years.

About the Project:

The project focuses on the theoretical grounding and algorithmic realization of embodied quantization—investigating how continuous sensory inputs are distilled into discrete tokens (such as concepts and objects) to facilitate control in complex environments. Key responsibilities include bridging high-capacity generative representations and discrete world models with non-stationary, risk-sensitive reinforcement learning theory. To achieve this, the candidate will develop uncertainty-aware parsimonious world models (e.g., via evidential learning) and prove finite-sample or regret guarantees under non-stationary latent state spaces, validating these algorithms by building scalable prototypes and benchmarking across high-dimensional simulation suites.

Research Environment:

IMADA uniquely brings mathematicians and computer scientists together within a single department to foster theoretically well-backed, high-quality data science research. The department is home to numerous externally funded research projects, and the Data Science and Statistics Group serves as a vibrant synergy platform for experts across fields. The successful candidate will join the ADIN Lab, collaborate on publishing at top-tier venues (NeurIPS, ICML, ICLR, AISTATS), and fulfill standard teaching assistantship duties.

Expected Skills and Qualifications:

We are seeking a candidate with a strong desire to make significant contributions to fundamental machine learning research, possessing a combination of mathematical maturity and advanced engineering skills:

Education:

A PhD in Computer Science, Mathematics, Statistics, or Theoretical Physics at the time of employment.

Publication Track Record:

At least two first-author research papers at flagship venues of core machine learning research (e.g., NeurIPS, ICML, ICLR, AISTATS).

Theoretical Rigor:

A deep understanding of statistical learning theory and reinforcement learning foundations, with the ability to conduct performance, convergence, or regret bound analysis (e.g., optimistic posterior sampling) in discrete or latent non-stationary environments.

Implementation Expertise:

Outstanding scientific programming skills (Python, PyTorch/JAX) with a proven track record of developing, debugging, and scaling deep generative models, discrete codebook world models, or complex RL pipelines. Clean public repositories or released source code from past publications is a strong plus.

Algorithmic Breadth:

Familiarity with probabilistic machine learning, evidential learning, discrete variational autoencoders (VQ-VAEs), transformers, or model-based RL is highly desirable.

Communication:

Excellent spoken and written communication skills in English.

Application deadline: 7. September 2026 at 23:59 hours local Danish time