Dear all,
this is a pre-information to a talk on the topic "Gaussian Processes come with (almost) free explainability" given by Sarem Seitz on the 2nd of December during our virtual colloquium. Everyone is invited to join with the typical credentials given in this VC course and to discuss with us. The abstract can be found below.
Best regards
Johannes
Abstract:
Gaussian Processes have proven themselves as a reliable and effective method in probabilistic Machine Learning. Thanks to recent and current advances, modeling complex data with Gaussian Processes is becoming more and more feasible. Thus, these types of models are, nowadays, an interesting alternative to classic Deep Learning, which is arguable the current state-of-the-art in Machine Learning. For the latter, we see an increasing interest in so-called explainable approaches, in essence methods that aim to make a deep model’s decision process transparent to humans. Given the necessity of explainability in many real-world applications, a transfer of successful explainable deep network approaches to the domain of Gaussian Processes could be a fruitful step in order to further improve their practical relevance. In particular, this work will focus on an adaption of Integrated Gradients as introduced by Sundararajan et al. [2017].