INF526 Selected Topics in Applied Computer Science IV: Probabilistic Machine Learning
CSE127 Machine Learning with Astronomical Data
Department of Computer Science, Turkish-German University, Istanbul
Instructor: Emre Işık
Parameter inference in science & engineering is of utmost importance. In natural dynamical systems (such as an ecosystem or a galaxy) we attempt to estimate parameters without an exact understanding of the intrinsic, often too complex, physics that controls the system. In artificial dynamical systems (such as a mega-city or a deep neural network), we also need to infer some not-directly-observable parameters, along with their uncertainties.
Probabilistic (especially Bayesian) approaches in machine learning are becoming increasingly attractive in modelling complex systems with all sorts of noise and degeneracies. This course will be an overview of the topic, with examples from astronomical data, which is good because they really come from complex black-box systems, such as planets or stars. They are practical examples to get you accustomed to knowledge discovery and prediction business, even if your aim is to deal with more ‘Earthly’ stuff, like stock exchanges or medical imaging.
Until the mid-terms, I will basically follow the relevant parts of the book Math for ML by Deisenroth et al. (2020), with applications (for recitation hours) from here and there, as well as from @Emre ‘s astro-data and simulations. After the mid-term, lecture & recitation materials will be from diverse resources.
(each week has 90 mins of lecture, ~90 mins of practice/recitation sessions)