Dear all, The Brain and Data Science Lab of Stefan Haufe at Charité - Universitätsmedizin Berlin is looking to recruit two highly motivated PhD students. Appointments can be made for four years starting immediately. 1. Theory and practice of interpreting machine learning models Machine learning systems have recently set new standards for solving a wide range of problems by leveraging vast amounts of training data. But they have remained “black boxes” whose internal workings are too complex to be comprehensible by a human. Especially in the health domain, it is desirable to explain and visualize the decisions of ML models. Recently, it has been shown that many existing explanation methods can be misleading even when using simple linear ML models [Haufe et al., 2014; Kindermans et al., 2018]. In this PhD project, a bottom-up approach will be performed, in which correct interpretation will be defined axiomatically. The resulting definition will be used to benchmark novel and existing explanation methods using synthetic ground-truth data. The resulting methodology will also be applied to clinical use cases. 2. Estimating and characterizing EEG/MEG functional connectomes in aging and dementia The study of functional brain interactions promises to greatly enhance our understanding of mental diseases. EEG and MEG make it possible to brain dynamics at high temporal scales but suffer from low spatial resolution, which often leads to false detections of brain connectivity. While the problem has been overcome for linear connectivity metrics [Nolte et al., 2004; Haufe et al., 2013] it still persists for non-linear interactions such as phase-amplitude and amplitude-amplitude coupling, which have been postulated as possible mechanisms of brain communication. This PhD project will establish a best practice to reconstruct non-linear connectivity from EEG/MEG data. The developed pipeline will be applied to study brain connectivity in aging and neurological conditions (dementia) using big EEG/MEG datasets. Interpretable machine learning will further be used to relate functional connectomes to behavior and relevant clinical variables. The Brain and Data Science Group at Charité develops machine learning and signal processing methods for the analysis of non-invasive brain signals in health and disease. We are located at the Berlin Center for Advanced Neuroimaging (BCAN) on the beautiful historic Charité Campus Mitte in the center of Berlin, and are embedded in a stimulating inter-disciplinary research environment. The group is funded by an ERC starting grant of the European Union. Requirements: Candidates are expected to hold a very good MSc or equivalent degree preferably in a technical field (machine learning, computer science, statistics, mathematics, computational (neuro) science, data science, physics, electrical/biomedical engineering, etc.). All positions require a solid math/statistics background, proficiency in written English, and good coding skills (e.g., Matlab, Python, C++, Java). Prior experience with functional neuroimaging data is a plus. Applications should include a letter of motivation, a CV, transcripts and degree certificates, as well as (if available) references, an English-language writing sample, and a coding sample (e.g. link to a github project). Applications should be sent to stefan.haufe@charite.de until Oct 15th 2019. All attached documents should be contained in a single pdf. See braindata.charite.de for further information on the positions and our group. With best wishes, Stefan Haufe -- Stefan Haufe ERC Research Group Leader Charité - Universitätsmedizin Berlin Berlin Center for Advanced Neuroimaging (BCAN) Charitéplatz 1 Sauerbruchweg 4 10117 Berlin Tel: +49 30 450 639 639 Fax: +49 30 450 539 951 Web: braindata.charite.de