Exploring Complexity - Hanna Tseran's PhD Thesis on Deep Maxout Neural Networks

Hanna Tseran successfully defended her PhD thesis on activation functions in neural networks. She continues to work on deep learning theory in the Matsuo Laboratory at the University of Tokyo. Well done Hanna, and our best wishes!

Learning with neural networks depends on the particular parametrization of the functions represented by the network, i.e. the assignment of parameters to functions. It also depends on the identity of the functions, which are assigned typical parameters at initialization, and later on the parameters that emerge during training. The choice of activation function is a critical aspect of network design that influences these function properties and warrants investigation. Hanna's thesis focuses on analyzing the expected behavior of networks with maxout (multi-argument) activation functions. Beyond enhancing the practical applicability of maxout networks, these findings contribute to the theoretical exploration of activation functions beyond the common choices. She believes that this work can advance the study of activation functions and complicated neural network architectures. Her thesis entitled "Expected Complexity and Gradients of Deep Maxout Neural Networks and Implications to Parameter Initialization" was supervised by Guido Montúfar and Bernd Sturmfels.

Hanna is broadly interested in machine learning, with a focus on the mathematical aspects. She is particularly fascinated by deep learning theory. She believes it can benefit our understanding of deep learning mechanics, facilitate the informed design of machine learning models, and help ensure positive societal impact. In her current position as a project researcher at the Matsuo Laboratory at the University of Tokyo in Japan, she plans to further pursue these goals.

Hanna Tseran is originally from Minsk, Belarus. She received a specialist degree in computer science from the Belarusian State University in 2015, and a master's degree from the Graduate School of Information Science and Technology at the University of Tokyo in 2018. After that, she worked for a year as a research software engineer at Microsoft in the UK, before starting her PhD at MPI MiS in Leipzig in 2020. She also has extensive experience in research and software engineering. Before her Master's studies, she worked as a software engineer and later interned at Google, Amazon, and RIKEN.