PhD student position in machine learning at the Department of Computer Science and Engineering, Chalmers University of Technology and University of Gothenburg.
Information about the project
Machine learning is now an essential tool for scientists and engineers. It is used in diverse applications to predict outcomes from inputs by training models to minimize prediction error in training data. As widespread adoption reaches beyond research and the developers of such systems, the seams have started to show in this attractively simple idea. State-of-the-art models which achieve top accuracy on benchmark tasks fail to generalize to new examples and to highly related problem domains. In this project, we will study the combination of causal inference and learning using auxiliary information to improve the efficiency and domain robustness of learning algorithms.
Generalization in machine learning refers to a trained system performing well on previously unseen examples. These new examples are often assumed to follow the same distribution as training examples, which guarantees good generalization if the number of samples is large enough. In real-world applications, however, data often follows different patterns: 1) We often have access to different (auxiliary) information at training time than we do when the trained system is deployed, 2) The distribution of in-deployment samples often differs from those collected for training, 3) The number of samples is rarely as large as we would like it to be. In this project, we will develop sample-efficient learning algorithms which makes the best possible use of small numbers of training examples by exploiting auxiliary information and causal assumptions.
The position is placed in the research group led by Fredrik Johansson, currently comprised of 5 PhD students working on topics related to machine learning for improved decision making with applications in healthcare.