We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cond-mat.stat-mech

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Condensed Matter > Statistical Mechanics

Title: Exploring Predictive States via Cantor Embeddings and Wasserstein Distance

Abstract: Predictive states for stochastic processes are a nonparametric and interpretable construct with relevance across a multitude of modeling paradigms. Recent progress on the self-supervised reconstruction of predictive states from time-series data focused on the use of reproducing kernel Hilbert spaces. Here, we examine how Wasserstein distances may be used to detect predictive equivalences in symbolic data. We compute Wasserstein distances between distributions over sequences ("predictions"), using a finite-dimensional embedding of sequences based on the Cantor for the underlying geometry. We show that exploratory data analysis using the resulting geometry via hierarchical clustering and dimension reduction provides insight into the temporal structure of processes ranging from the relatively simple (e.g., finite-state hidden Markov models) to the very complex (e.g., infinite-state indexed grammars).
Comments: 9 pages, 4 figures; this http URL
Subjects: Statistical Mechanics (cond-mat.stat-mech); Machine Learning (cs.LG); Dynamical Systems (math.DS); Machine Learning (stat.ML)
DOI: 10.1063/5.0102603
Cite as: arXiv:2206.04198 [cond-mat.stat-mech]
  (or arXiv:2206.04198v1 [cond-mat.stat-mech] for this version)

Submission history

From: James P. Crutchfield [view email]
[v1] Thu, 9 Jun 2022 00:09:47 GMT (761kb,D)

Link back to: arXiv, form interface, contact.