In this talk Dr Xavier Alameda-Pineda will discuss several works on automatic understanding of human behavior. First, He will present recent variational expectation-maximisation techniques for multi-speaker tracking, where a combinatorial problem over time yields to a high-complexity exact solution that is intractable in practice.
Variational approximation of the posterior leads to an efficient, yet performant, way to tackle the problem. Secondly, He will present robust deep regression techniques based on probabilistic models and discuss the joint training of the graphical model and the network parameters.
Finally, He will propose one way to exploit behavioral diversity for data generation.
Bio: Xavier Alameda-Pineda received M.Sc. in Mathematics (2008), in Telecommunications (2009) and in Computer Science (2010). He obtained his Ph.D. in Mathematics and Computer Science from Universite Joseph Fourier in 2013. Since 2016, he is a Research Scientist at Inria Grenoble, with the Perception Team. He served as Area Chair at ICCV 2017, at ACM MM 2019 and at ICIAP 2019. He is the recipient of several paper awards, and of the ACM SIGMM Rising Star 2018. He is a Senior Member of the IEEE. His scientific interests lie in computer vision, machine learning and signal processing for human behavior understanding and robotics.