Latitude

A model for mixed linear-tropical matrix factorization

Sanjar Karaev, James Hook, Pauli Miettinen

Research output: Contribution to conferencePaper

Abstract

Nonnegative matrix factorization (NMF) is one of the most frequently-used matrix factorization models in data analysis. A significant reason to the popularity of NMF is its interpretability and the 'parts of whole' interpretation of its components. Recently max-times, or subtropical, matrix factorization (SMF) has been introduced as an alternative model with equally useful 'winner takes it all' interpretation. In this paper we propose a new mixed linear-tropical model and a new algorithm called Latitude that combines NMF and SMF, being able to smoothly alternate between the two. In our model the data is modeled using latent factors and additional latent parameters that control whether the factors are interpreted as NMF or SMF features or as mixtures of both. We present an algorithm for our novel matrix factorization. Our experiments show that it improves over both baselines and can yield interpretable results that reveal more of the latent structure than either NMF or SMF alone.

Original languageEnglish
Pages360-368
Number of pages9
Publication statusPublished - 5 May 2018
Event2018 SIAM International Conference on Data Mining, SDM 2018 - San Diego, USA United States
Duration: 3 May 20185 May 2018

Conference

Conference2018 SIAM International Conference on Data Mining, SDM 2018
CountryUSA United States
CitySan Diego
Period3/05/185/05/18

Fingerprint

Factorization

Keywords

  • Matrix factorization
  • NMF
  • Subtropical algebra

ASJC Scopus subject areas

  • Computer Science Applications
  • Software

Cite this

Karaev, S., Hook, J., & Miettinen, P. (2018). Latitude: A model for mixed linear-tropical matrix factorization. 360-368. Paper presented at 2018 SIAM International Conference on Data Mining, SDM 2018, San Diego, USA United States.

Latitude : A model for mixed linear-tropical matrix factorization. / Karaev, Sanjar; Hook, James; Miettinen, Pauli.

2018. 360-368 Paper presented at 2018 SIAM International Conference on Data Mining, SDM 2018, San Diego, USA United States.

Research output: Contribution to conferencePaper

Karaev, S, Hook, J & Miettinen, P 2018, 'Latitude: A model for mixed linear-tropical matrix factorization' Paper presented at 2018 SIAM International Conference on Data Mining, SDM 2018, San Diego, USA United States, 3/05/18 - 5/05/18, pp. 360-368.
Karaev S, Hook J, Miettinen P. Latitude: A model for mixed linear-tropical matrix factorization. 2018. Paper presented at 2018 SIAM International Conference on Data Mining, SDM 2018, San Diego, USA United States.
Karaev, Sanjar ; Hook, James ; Miettinen, Pauli. / Latitude : A model for mixed linear-tropical matrix factorization. Paper presented at 2018 SIAM International Conference on Data Mining, SDM 2018, San Diego, USA United States.9 p.
@conference{fef23f28b2734c7790ff6bdec8eaab28,
title = "Latitude: A model for mixed linear-tropical matrix factorization",
abstract = "Nonnegative matrix factorization (NMF) is one of the most frequently-used matrix factorization models in data analysis. A significant reason to the popularity of NMF is its interpretability and the 'parts of whole' interpretation of its components. Recently max-times, or subtropical, matrix factorization (SMF) has been introduced as an alternative model with equally useful 'winner takes it all' interpretation. In this paper we propose a new mixed linear-tropical model and a new algorithm called Latitude that combines NMF and SMF, being able to smoothly alternate between the two. In our model the data is modeled using latent factors and additional latent parameters that control whether the factors are interpreted as NMF or SMF features or as mixtures of both. We present an algorithm for our novel matrix factorization. Our experiments show that it improves over both baselines and can yield interpretable results that reveal more of the latent structure than either NMF or SMF alone.",
keywords = "Matrix factorization, NMF, Subtropical algebra",
author = "Sanjar Karaev and James Hook and Pauli Miettinen",
year = "2018",
month = "5",
day = "5",
language = "English",
pages = "360--368",
note = "2018 SIAM International Conference on Data Mining, SDM 2018 ; Conference date: 03-05-2018 Through 05-05-2018",

}

TY - CONF

T1 - Latitude

T2 - A model for mixed linear-tropical matrix factorization

AU - Karaev, Sanjar

AU - Hook, James

AU - Miettinen, Pauli

PY - 2018/5/5

Y1 - 2018/5/5

N2 - Nonnegative matrix factorization (NMF) is one of the most frequently-used matrix factorization models in data analysis. A significant reason to the popularity of NMF is its interpretability and the 'parts of whole' interpretation of its components. Recently max-times, or subtropical, matrix factorization (SMF) has been introduced as an alternative model with equally useful 'winner takes it all' interpretation. In this paper we propose a new mixed linear-tropical model and a new algorithm called Latitude that combines NMF and SMF, being able to smoothly alternate between the two. In our model the data is modeled using latent factors and additional latent parameters that control whether the factors are interpreted as NMF or SMF features or as mixtures of both. We present an algorithm for our novel matrix factorization. Our experiments show that it improves over both baselines and can yield interpretable results that reveal more of the latent structure than either NMF or SMF alone.

AB - Nonnegative matrix factorization (NMF) is one of the most frequently-used matrix factorization models in data analysis. A significant reason to the popularity of NMF is its interpretability and the 'parts of whole' interpretation of its components. Recently max-times, or subtropical, matrix factorization (SMF) has been introduced as an alternative model with equally useful 'winner takes it all' interpretation. In this paper we propose a new mixed linear-tropical model and a new algorithm called Latitude that combines NMF and SMF, being able to smoothly alternate between the two. In our model the data is modeled using latent factors and additional latent parameters that control whether the factors are interpreted as NMF or SMF features or as mixtures of both. We present an algorithm for our novel matrix factorization. Our experiments show that it improves over both baselines and can yield interpretable results that reveal more of the latent structure than either NMF or SMF alone.

KW - Matrix factorization

KW - NMF

KW - Subtropical algebra

UR - http://www.scopus.com/inward/record.url?scp=85048329760&partnerID=8YFLogxK

M3 - Paper

SP - 360

EP - 368

ER -