• Home
  • Research
  • Teaching
  • Tutorials
  • CV

Cumulative topology for time-series classification and regime detection.

Research · Plate I
Euler Characteristic Methods
Cumulative topology · classification · regime detection Three preprints under review Updated MMXXVI
An Euler characteristic surface drawn as a heatmap over filtration parameter and threshold.
The Euler characteristic surface: a 2D map χ(ε,τ), encoding topological summary across filtration scale and threshold.
“The Euler characteristic is the simplest topological invariant—and, after enough integration, the most useful.”

The Euler characteristic is the modest invariant: an alternating sum of simplex counts that, written down, fits on a postcard. Computed across a filtration, however, it encodes considerably more structure than the postcard would suggest. Recent work uses Euler characteristic profiles and Euler characteristic surfaces as fast, interpretable descriptors for time-series classification, regime detection in chaotic dynamical systems, and unsupervised classification of two-phase flow regimes.

The advantage of these descriptors over their persistence-based cousins is computational: the Euler characteristic is linear in simplex count and can be tracked through a filtration in a single pass. The cost is that the descriptor is not, in general, complete—different topologies can share an Euler characteristic surface. The work navigates that trade-off honestly.

Active threads

Filed under Theory · Application Three preprints, MMXXVI
  • Interpretable Classification of Time Series Using Euler Characteristic Surfaces—a fully interpretable pipeline for time-series classification, with Atish Mitra and the NIT Sikkim group; under review at Nature Scientific Reports.
  • Detecting Regime Transitions in Dynamical Systems via the Mixup Euler Characteristic Profile—a mixup-style ECP for regime change in chaotic systems, submitted to Chaos.
  • Topological Characterization of Churn Flow and Unsupervised Correction to the Wu Flow-Regime Map—ECS for two-phase flow classification in vertical pipes, submitted to International Journal of Multiphase Flow.

Collaborators

Filed under Co-authors Two institutions, several students
  • Atish Mitra, Montana Technological University
  • Md. Nurujjaman, NIT Sikkim
  • Buddha Nath Sharma, Salam Rabindrajit Luwang—NIT Sikkim
  • Brady Koenig, Abigail Stein, Burt Todd—Montana Tech
  • Vishal Mandal, Santanu Nandi
❦

Back to the research overview, or read about the closely related work on TDA for time series & dynamics.

Washington, D.C.

Data Science · The George Washington University

  • Report an issue

© MMXXVI Sushovan Majhi