Postdoctoral position in Sample-Efficient Probabilistic Machine Learning

The is an international scientific community of 40,000 students and researchers. It is one of the leading multidisciplinary research universities in Europe and ranks among the top 100 international universities in the world.

The , which is part of the Faculty of Science, is a leading Computer Science research and teaching unit in Finland. The research themes of the Department cover machine learning and algorithms, computer networks and distributed systems, software systems and bioinformatics.

The Department of Computer Science, Faculty of Science, University of Helsinki invites applications for a

Postdoctoral Researcher in Sample-Efficient Probabilistic Machine Learning

Position description

The led by Principal Investigator Luigi Acerbi is looking for a postdoctoral researcher eager to work on new machine learning methods for smart, robust, sample-efficient probabilistic inference, with applications in scientific modeling — such as, but not limited to, computational and cognitive neuroscience. The candidate will join an established research group with strong links to the .

In our group, we develop machine learning approaches for building approximate Bayesian posteriors using only a small number of likelihood evaluations, which can be a game-changer for complex models or when resources are limited. Think of Bayesian optimization, a very effective technique to optimize black-box functions, but scaled up to perform full Bayesian inference. This project will build on the state-of-the-art framework developed in our group, , which combines Gaussian process surrogates, active learning, variational inference and Bayesian quadrature (, ).

Example research directions for this project include increasing the flexibility and robustness of VBMC to more complex posteriors, for example moving beyond the standard Gaussian kernel for Gaussian processes; exploring alternative deep-learning based surrogates such as normalizing flows; exploiting recent advances in Gaussian process inference for superior scalability; extending surrogate-based methods to hierarchical Bayesian inference; exploring highly distributed and "embarrassingly parallel" approaches for surrogate-based inference (). Concrete applications include fitting state-of-the-art models of human and animal decision making in computational and cognitive neuroscience (e.g., ).

The project will include research visits abroad to work with collaborators, such as 's group at the University of Geneva (Switzerland) and  and at the Ecole Normal Supérieure in Paris (France).

The position is full-time, funded for 2 years and will be filled as soon as possible, with a target starting date in Aug-Dec 2024 (later dates can be negotiated). This project is funded by the Research Council of Finland.

Requirements

The work involves research-related activities, including conducting theoretical and applied research, designing and programming machine learning software, computational and data analyses, writing research articles, participating in and presenting research at academic conferences, taking part in international research visits, and teaching-related activities.

The ideal candidate has a strong background in computational statistics and/or machine learning and experience with Bayesian modelling and methods (e.g., MCMC, variational inference, probabilistic programming, simulator-based inference). The candidate should also match at least one of the following requirements, and ideally more than one:

• Experience in implementing efficient code in Python (e.g., PyTorch, JAX).
• Experience with Gaussian processes and/or Bayesian optimization.
• Experience with computational modeling and statistical model fitting in an applied field (especially computational and cognitive neuroscience).

Excellent written and oral communication skills in English are needed, as well as previous research experience and publications.

Applicants should have – or be about to obtain – a PhD in machine learning, computer science, data science, applied mathematics and statistics, physics, computational neuroscience, or a related field.

Salary and benefits

The position is full-time and funded for 2 years. The starting salary is 3800–4500 euros/month, depending on previous qualifications and experience.

The University of Helsinki offers comprehensive services to its employees, including occupational health care and health insurance, sports facilities, and opportunities for professional development. The assists employees from abroad with their transition to work and live in Finland.

How to apply

Please submit your application with its attachments to the group leader, Luigi Acerbi (please find email address below), specifying "Application for postdoc in Sample-Efficient Probabilistic Machine Learning" as the email subject.

The application should include the following attachments as pdf files (in English):

1. CV with possible list of publications and previous projects;
2. A copy of your transcripts (list of courses completed during BSc/MSc and grades);
3. Cover letter (1-2 pages) with motivation, research interests and match to the project;
4. Contact details of two referees who could provide a letter of recommendation (can be included at the end of the CV);
5. An extended sample of your own scientific writing (e.g., your PhD thesis, or a paper as long as you were the main writer; could be a link if available online).

Applications will be considered until the position is filled.

Additional information

For more information on the group and the position, please:

• visit the
• contact the group leader Luigi Acerbi ()

These are to move to Finland.

Our group is committed to principles of diversity, equality and inclusion. We encourage applications from women, racial and ethnic minorities, and other under-represented individuals.

Appendix: Background references and material

Work related to the VBMC framework:

• Acerbi (2018). Variational Bayesian Monte Carlo. NeurIPS ().
• Acerbi (2020). Variational Bayesian Monte Carlo with Noisy Likelihoods. NeurIPS ().
• Huggins*, Li*, Tobaben*, Aarnos & Acerbi (2023). PyVBMC: Efficient Bayesian inference in Python. JOSS ().

Approaches for "embarrassingly parallel" inference:

• de Souza, Mesquita, Kaski & Acerbi (2022). Parallel MCMC Without Embarrassing Failures. AISTATS ().