Multi-source probabilistic inference /

Open positions

Open positions

Postdoctoral researcher in Machine learning for sensing

We use flexible probabilistic models, from Bayesian deep learning to Gaussian processes, for enhancing the capability of physical sensors. For example, we design simple sensors together with the computational models to create a combination where both elements best complement each other. We are working on example cases in low-cost ultrasonic sensing, acoustic levitation, spectral imaging on ordinary cameras, and adaptive optics for exoplanet discovery. We already have personnel and collaborators working on these cases, and are now looking for a postdoc to work on general machine learning research topics to support them.

Your task would be to conduct fundamental research on machine learning, ideally building on your already strong expertise in a topic that relates to the general goal. We have already identified several directions of theoretical and practical interest, but you are also encouraged to bring in topic of personal interest and expertise and explain why you would want to work on those with me.  For all topics, the main goal is to conduct fundamental research on machine learning that aims for publications in top-level venues (ICML, NeurIPS, AISTATS, ICLR, JMLR etc).

A non-exhaustive list of example directions:

  1. Efficient approximate inference (geometric MCMC and/or variational approximations) for flexible Bayesian models. See our recent NeurIPS paper for an example of the kind of work we do. We are also currently working on extremely promising idea for speeding up geometric MCMC algorithms.
  2. Hybrid methods combining strong domain-specific models with data-driven elements. See our recent UAI paper that maps ultrasonic physics to a Gaussian process formulation, and our deep learning architecture motivated by optical diffraction for low-cost spectral imaging.
  3. Unsupervised or weakly supervised methods for signal data, and techniques for coping with limited labels (domain adaptation, few-shot learning).
  4. Real-time control for active sensing using e.g. model-based reinforcement learning. See our paper on adaptive optics based on reinforcement learning.

An ideal candidate has strong publication record in machine learning, statistics or signal processing, and wants to continue fundamental research on core algorithms and models. We offer the opportunity to do this for interesting cases with unique data, while working on other ML researchers focusing on similar methodological questions.

Apply via the HICT postdoctoral call by August 9, 2021!

Postdoctoral researcher for FCAI core topics

There are also several open positions for more open-ended projects in the same call, with possibility of working with me as one of the supervisors. 

Outside the calls

Exceptional students interested in doctoral studies and strong candidates for post-doctoral researcher positions are encouraged to contact the group leader directly also outside the calls. Check also https://fcai.fi/open-positions for possible open positions in artificial intelligence research in several research groups in the Helsinki area.