PhD student position in Machine Learning: Theoretically motivated deep learning
At the Division of Computational Science and Technology at KTH we are seeking a new PhD student in Machine Learning / Computer Vision to handle scale-dependent information in image data. In our research, we develop deep networks for processing image data that handle scaling transformations and other image transformations in a theoretically well-founded manner. Our research in this area comprises both theoretical modelling of the influence of image transformations on different architectures for deep networks as well as experimental evaluations of such networks on benchmark datasets to explore their properties. The work also comprises the creation of new benchmark datasets, to enable characterization of properties of deep networks that are not covered by existing datasets. For examples of our previous work in this area, see https://www.kth.se/profile/tony/page/deep-networks Within the scope of this PhD student position, you will work on and contribute to the research frontier regarding scale-covariant or scale-equivariant deep networks and/or deep networks parameterised in terms of Gaussian derivatives, on specific research topics that we choose together within the scope of the research project ”Covariant and invariant deep networks” that finances this position. The overall goal is to develop new architectures for deep networks that can generalise to scaling variations that are not spanned by the training data, and which can achieve higher robustness to variabilities in test data, as well as enable more efficient training with lower requirements concerning the amount of training data. The candidate should have very good knowledge in mathematics (analysis and linear systems, which we use for modelling convolution transformations and geometric image transformations) as well as in structured programming to write code that is easy to use for making experiments with, maintain and develop and share with colleagues. You must have very good knowledge about programming deep networks in Python, PyTorch is meritorious. Knowledge in computer vision and image analysis is strongly meritorious. For further information and information about how to apply, see https://www.kth.se/en/om/work-at-kth/lediga-jobb/what:job/jobID:567595/where:4/ [cid:[email protected]] Tony Lindeberg Professor of Computer Science — Computational Vision KTH Royal Institute of Technology Computational Brain Science Lab Division of Computational Science and Technology, CST SE-100 44 Stockholm, Sweden Phone: +46 8 790 6205 https://www.kth.se/profile/tony/
