2 PhD student positions in AI

Information about the projects
Topic 1: Learning and Leveraging Rich Priors for Factorization Problems

In this project we are interested in developing methods that combine traditional (parametric) mathematical formulations induced by domain expertise with (non-parametric) models learned from examples. Parametric models inject domain knowledge into learning-based approaches and have therefore the potential to massively reduce the necessary amount of training data. Additionally, the output can be constrained e.g. to be physically plausible, which is difficult to guarantee with pure learning-based architectures. At the same time, being able to incorporate learned priors has the potential to regularize problems where a physical model is not sufficient to guarantee a well posed formulation. From a theoretical point of view we are interested in results that characterize formulations in terms of their expressiveness and generalization as well as developing efficient inference approaches.

Our main application of interest are factorization-based problems, in particular non-rigid structure-from-motion (NRSfM), which aims to infer 3D models of dynamic scenes or objects from videos or multiple images. In contrast to its rigid counterpart, NRSfM is far less mature and it is inherently an ill posed problem requiring suitable priors that disambiguate the effects of camera motion and object deformation. Hence, we see this application as ideal and non-trivial testbed for combining our 3D computer vision and matrix factorization expertise with machine learning methods.

Funding for this project has been obtained from the Wallenberg AI, Autonomous Systems and Software Program (WASP) which is Sweden’s largest ever individual research program, a major national initiative for strategically basic research, education and faculty recruitment. The program is initiated and generously funded by the Knut and Alice Wallenberg Foundation (KAW) with 2.6 billion SEK. In addition to this, the program receives support from collaborating industry and from participating universities to form a total budget of 3.5 billion SEK. Major goals are more than 50 new professors and more than 300 new PhDs within AI, Autonomous Systems and Software. The vision of WASP is excellent research and competence in artificial intelligence, autonomous systems and software for the benefit of Swedish industry. For more information about the research and other activities conducted within WASP please visit: http://wasp-sweden.org/

Topic 2: Energy-based models for supervised deep neural networks and their applications

Despite  deep  learning-based  methods  being  the  state-of-the-art  in  many  AI-related  applications,  there  is  a lack  of  consensus  of  how  to  understand  and  interpret  deep  neural  networks  in  order  to  reason  about  their strengths and weaknesses.  Energy-based models in machine learning have a long tradition as a framework to learn from unlabeled data, i.e. unsupervised learning.  Recently, it has been shown that supervised learning of deep neural networks using the back propagation method is a limiting case of a suitably defined approach for learning energy-based models using a so-called contrastive loss.  This connection is the basis for our interest in a tighter connection between deep learning and energy-based models. In contrast to feed-forward deep neural networks, energy-based models also allow information flow from later layers (considered to detect more abstract features) to earlier layers (low-level features).

Energy-based models leverage optimization methods at two levels: at the level of inferring the units’ activations, and at the level of learning the model parameters. This in turn enables a two-fold flexibility for shaping the underlying learning losses (at the inference level and at the learning  level), which is one of the motivating factors for this project. Overall, we aim to explore the benefits of energy-based models for deep neural networks in the following aspects: prediction uncertainty and input confidence, learning efficiency, formal robustness and explainability.

Funding for this position has been obtained from the Chalmers AI Research Centre (CHAIR). CHAIR is a relatively new research centre aiming to  increase Chalmers’ expertise and excellence in artificial intelligence. For more information about CHAIR please visit https://www.chalmers.se/en/centres/chair/Pages/default.aspx.

Major responsibilities
Your major responsibilities are to pursue your own doctoral studies. You are expected to develop your own scientific concepts and communicate the results of your research verbally and in writing, both in Swedish and in English. The position generally also includes teaching on Chalmers' undergraduate level or performing other duties corresponding to 20 per cent of working hours.

Position summary
Full-time temporary employment. The position is limited to a maximum of five years.

Qualifications
To qualify as a PhD student, you must have a master's level degree corresponding to at least 240 higher education credits in a relevant field (physics, mathematics or computer science).

The position requires sound verbal and written communication skills in Swedish and English. If Swedish is not your native language, you should be able to teach in Swedish after two years. Chalmers offers Swedish courses.

Application procedure
The application should be marked with Ref 20200040 and written in English. The application should be sent electronically and be attached as pdf-files. Find application procedure by reading the ad at Chalmers webpage.

Application deadline: 27th February - 2020

For questions, please contact:
Christopher Zach, Professor, Electrical Engineering
zach@chalmers.se