Persistence paths and signature features in topological data analysis
We introduce a new feature map for barcodes as they arise in persistent homology computation. The main idea is to first realize each barcode as a path in a convenient vector space, and to then compute its path signature which takes values in the tensor algebra of that vector space. The composition o...
Main Authors: | , , |
---|---|
Format: | Journal article |
Language: | English |
Published: |
IEEE
2018
|
_version_ | 1797108846287650816 |
---|---|
author | Chevyrev, I Nanda, V Oberhauser, H |
author_facet | Chevyrev, I Nanda, V Oberhauser, H |
author_sort | Chevyrev, I |
collection | OXFORD |
description | We introduce a new feature map for barcodes as they arise in persistent homology computation. The main idea is to first realize each barcode as a path in a convenient vector space, and to then compute its path signature which takes values in the tensor algebra of that vector space. The composition of these two operations - barcode to path, path to tensor series - results in a feature map that has several desirable properties for statistical learning, such as universality and characteristicness, and achieves state-of-the-art results on common classification benchmarks. |
first_indexed | 2024-03-07T07:33:48Z |
format | Journal article |
id | oxford-uuid:b5f8c42c-0354-4284-b19c-24af2b74b1a6 |
institution | University of Oxford |
language | English |
last_indexed | 2024-03-07T07:33:48Z |
publishDate | 2018 |
publisher | IEEE |
record_format | dspace |
spelling | oxford-uuid:b5f8c42c-0354-4284-b19c-24af2b74b1a62023-02-09T12:56:12ZPersistence paths and signature features in topological data analysisJournal articlehttp://purl.org/coar/resource_type/c_dcae04bcuuid:b5f8c42c-0354-4284-b19c-24af2b74b1a6EnglishSymplectic Elements at OxfordIEEE2018Chevyrev, INanda, VOberhauser, HWe introduce a new feature map for barcodes as they arise in persistent homology computation. The main idea is to first realize each barcode as a path in a convenient vector space, and to then compute its path signature which takes values in the tensor algebra of that vector space. The composition of these two operations - barcode to path, path to tensor series - results in a feature map that has several desirable properties for statistical learning, such as universality and characteristicness, and achieves state-of-the-art results on common classification benchmarks. |
spellingShingle | Chevyrev, I Nanda, V Oberhauser, H Persistence paths and signature features in topological data analysis |
title | Persistence paths and signature features in topological data analysis |
title_full | Persistence paths and signature features in topological data analysis |
title_fullStr | Persistence paths and signature features in topological data analysis |
title_full_unstemmed | Persistence paths and signature features in topological data analysis |
title_short | Persistence paths and signature features in topological data analysis |
title_sort | persistence paths and signature features in topological data analysis |
work_keys_str_mv | AT chevyrevi persistencepathsandsignaturefeaturesintopologicaldataanalysis AT nandav persistencepathsandsignaturefeaturesintopologicaldataanalysis AT oberhauserh persistencepathsandsignaturefeaturesintopologicaldataanalysis |