Unsupervised Neural Segmentation and Clustering for Unit Discovery in Sequential Data - Archive ouverte HAL Access content directly
Conference Papers Year :

Unsupervised Neural Segmentation and Clustering for Unit Discovery in Sequential Data

Abstract

We study the problem of unsupervised segmentation and clustering of handwritten lines with applications to character discovery. We propose a constrained variant of Vector Quantized Variational Autoencoder (VQ-VAE) which produces a discrete and piecewise-constant encoding of the data. We show that the constrained quantization task is dual to a Markovian dynamics prior placed on the latent codes. Such view facilitates a probabilistic interpretation of the constraints and allows efficient inference. We demonstrate the effectiveness of the proposed method in the context of unsupervised handwriting character discovery in 17th-century scanned manuscripts.
Fichier principal
Vignette du fichier
PGR009.pdf (291.17 Ko) Télécharger le fichier
Origin : Files produced by the author(s)
Loading...

Dates and versions

hal-02399138 , version 1 (08-12-2019)

Identifiers

  • HAL Id : hal-02399138 , version 1

Cite

Jan Chorowski, Nanxin Chen, Ricard Marxer, Hans J G A Dolfing, Adrian Łańcucki, et al.. Unsupervised Neural Segmentation and Clustering for Unit Discovery in Sequential Data. NeurIPS 2019 workshop - Perception as generative reasoning - Structure, Causality, Probability, Dec 2019, Vancouver, Canada. ⟨hal-02399138⟩
391 View
570 Download

Share

Gmail Facebook Twitter LinkedIn More