Talk on “Piecewise Linear Differentiation in Data Analysis and Machine Learning”

Dr. Torsten Bosse from the Mathematics and Computer Science Division Argonne National Laboratory, USA talks about Piecewise Linear Differentiation in Data Analysis and Machine Learning.

Abstract:

In the past decade, there has been tremendous progress in data analysis and machine learning, which became important research directions in mathematics and computer science. Several applications within these research directions involve non-smooth functions and give rise to non-smooth optimization algorithms. For an efficient and robust treatment of these problems, good approximations of the original functions are required that reflect the structure of the underlying problem. Within this talk, we will discuss an extension of algorithmic differentiation to compute piecewise linear approximations of these functions and a suitable representation of their models. Furthermore, we will give some ideas, how some of these techniques can be used in the considered context and point out some challenges for this approach.