Title: Recent developments in hybrid causal discovery
Authors: Liam Solus - KTH Royal Institute of Technology (Sweden) [presenting]
Abstract: Combinatorics, algebra and discrete geometry have come to play an increasingly significant role in the development of methods for causal discovery, the goal of which is to infer the cause-effect relations amongst a system of variables based on available data. Since the basic model selection problem of causal discovery is NP-hard, a variety of techniques for learning a causal model efficiently have been studied. One family of such methods are the hybrid causal discovery algorithms, which rely on a mixture of conditional independence testing and greedy optimization. We will discuss some recently explored connections between causal discovery and classic problems in combinatorial optimization that yield hybrid algorithms with improved performance over the state-of-the-art. We will see that the geometric perspective promoted by fields such as algebraic statistics plays a key role in the identification of this new methodology.