Applied Maths Seminar: Fulcher -- Unifying interdisciplinary literatures of mathematical methods through large-scale data-driven comparison

Ben Fulcher (School of Physics) will give a talk at 12pm Wednesday 18 March in Carslaw
451.  After that we will go to lunch (all welcome, students get a free lunch).  

Title: Unifying interdisciplinary literatures of mathematical methods through
large-scale data-driven comparison 

Abstract: When approaching a data-analysis problem, the analyst often needs to select an
appropriate theoretical tool to apply that will be most accurate or informative for the
question at hand.  In the case of time-series analysis, in which hundreds of
quantitative analysis methods have been developed over many decades across many
disciplinary contexts, the limitations of the analyst’s subjectivity are clear.  How
can they know whether they have selected the right tool from their all-too-human limited
knowledge of an expansive and diverse methodological literature? In this talk I will
introduce a potential way to address this problem, termed the 'highly comparative
approach', in which a large library containing thousands of diverse time-series features
are compared systematically for their utility on a given problem.  Through mass
methodological comparison, the analyst can be pointed to the most promising areas of
theory from across a broad literature to draw on for their problem.  I will demonstrate
the breadth and key successes of this approach to date, focusing particularly on two key
problems: (i) inferring the distance to a bifurcation of a system with variable noise;
and (ii) indexing the degree of time-reversal asymmetry from univariate time-series
data.