Multimedia LEGO: LEarning structured model by probabilistic loGic Ontology tree

Shiyu Chang, Guo Jun Qi, Jinhui Tang, Qi Tian, Yong Rui, Thomas S. Huang

Research output: Contribution to journalConference article

Abstract

Recent advances in Multimedia research have generated a large collection of concept models, e.g., LSCOM and Media mill 101, which become accessible to other researchers. While most current research effort still focuses on building new concepts from scratch, little effort has been made on constructing new concepts upon the existing models already in the warehouse. To address this issue, we develop a new framework in this paper, termed LEGO, to seamlessly integrate both the new target training examples and the existing primitive concept models. LEGO treats the primitive concept models as a lego toy to potentially construct an unlimited vocabulary of new concepts. Specifically, LEGO first formulates the logic operations to be the lego connectors to combine existing concept models hierarchically in probabilistic logic ontology trees. LEGO then simultaneously incorporates new target training information to efficiently disambiguate the underlying logic tree and correct the error propagation. We present extensive experimental results on a large vehicle domain data set from Image Net, and demonstrate significantly superior performance over existing state-of-the-art approaches which build new concept models from scratch.

Original languageEnglish (US)
Article number6729585
Pages (from-to)979-984
Number of pages6
JournalProceedings - IEEE International Conference on Data Mining, ICDM
DOIs
StatePublished - Dec 1 2013
Event13th IEEE International Conference on Data Mining, ICDM 2013 - Dallas, TX, United States
Duration: Dec 7 2013Dec 10 2013

Fingerprint

Probabilistic logics
Ontology
Warehouses

Keywords

  • Concept recycling
  • Logical operations
  • Model warehouse
  • Multimedia LEGO
  • Probabilistic logic ontology tree

ASJC Scopus subject areas

  • Engineering(all)

Cite this

Multimedia LEGO : LEarning structured model by probabilistic loGic Ontology tree. / Chang, Shiyu; Qi, Guo Jun; Tang, Jinhui; Tian, Qi; Rui, Yong; Huang, Thomas S.

In: Proceedings - IEEE International Conference on Data Mining, ICDM, 01.12.2013, p. 979-984.

Research output: Contribution to journalConference article

@article{0f987b2d3a324f7e83e554b95531a4b7,
title = "Multimedia LEGO: LEarning structured model by probabilistic loGic Ontology tree",
abstract = "Recent advances in Multimedia research have generated a large collection of concept models, e.g., LSCOM and Media mill 101, which become accessible to other researchers. While most current research effort still focuses on building new concepts from scratch, little effort has been made on constructing new concepts upon the existing models already in the warehouse. To address this issue, we develop a new framework in this paper, termed LEGO, to seamlessly integrate both the new target training examples and the existing primitive concept models. LEGO treats the primitive concept models as a lego toy to potentially construct an unlimited vocabulary of new concepts. Specifically, LEGO first formulates the logic operations to be the lego connectors to combine existing concept models hierarchically in probabilistic logic ontology trees. LEGO then simultaneously incorporates new target training information to efficiently disambiguate the underlying logic tree and correct the error propagation. We present extensive experimental results on a large vehicle domain data set from Image Net, and demonstrate significantly superior performance over existing state-of-the-art approaches which build new concept models from scratch.",
keywords = "Concept recycling, Logical operations, Model warehouse, Multimedia LEGO, Probabilistic logic ontology tree",
author = "Shiyu Chang and Qi, {Guo Jun} and Jinhui Tang and Qi Tian and Yong Rui and Huang, {Thomas S.}",
year = "2013",
month = "12",
day = "1",
doi = "10.1109/ICDM.2013.49",
language = "English (US)",
pages = "979--984",
journal = "Proceedings - IEEE International Conference on Data Mining, ICDM",
issn = "1550-4786",

}

TY - JOUR

T1 - Multimedia LEGO

T2 - LEarning structured model by probabilistic loGic Ontology tree

AU - Chang, Shiyu

AU - Qi, Guo Jun

AU - Tang, Jinhui

AU - Tian, Qi

AU - Rui, Yong

AU - Huang, Thomas S.

PY - 2013/12/1

Y1 - 2013/12/1

N2 - Recent advances in Multimedia research have generated a large collection of concept models, e.g., LSCOM and Media mill 101, which become accessible to other researchers. While most current research effort still focuses on building new concepts from scratch, little effort has been made on constructing new concepts upon the existing models already in the warehouse. To address this issue, we develop a new framework in this paper, termed LEGO, to seamlessly integrate both the new target training examples and the existing primitive concept models. LEGO treats the primitive concept models as a lego toy to potentially construct an unlimited vocabulary of new concepts. Specifically, LEGO first formulates the logic operations to be the lego connectors to combine existing concept models hierarchically in probabilistic logic ontology trees. LEGO then simultaneously incorporates new target training information to efficiently disambiguate the underlying logic tree and correct the error propagation. We present extensive experimental results on a large vehicle domain data set from Image Net, and demonstrate significantly superior performance over existing state-of-the-art approaches which build new concept models from scratch.

AB - Recent advances in Multimedia research have generated a large collection of concept models, e.g., LSCOM and Media mill 101, which become accessible to other researchers. While most current research effort still focuses on building new concepts from scratch, little effort has been made on constructing new concepts upon the existing models already in the warehouse. To address this issue, we develop a new framework in this paper, termed LEGO, to seamlessly integrate both the new target training examples and the existing primitive concept models. LEGO treats the primitive concept models as a lego toy to potentially construct an unlimited vocabulary of new concepts. Specifically, LEGO first formulates the logic operations to be the lego connectors to combine existing concept models hierarchically in probabilistic logic ontology trees. LEGO then simultaneously incorporates new target training information to efficiently disambiguate the underlying logic tree and correct the error propagation. We present extensive experimental results on a large vehicle domain data set from Image Net, and demonstrate significantly superior performance over existing state-of-the-art approaches which build new concept models from scratch.

KW - Concept recycling

KW - Logical operations

KW - Model warehouse

KW - Multimedia LEGO

KW - Probabilistic logic ontology tree

UR - http://www.scopus.com/inward/record.url?scp=84894679974&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84894679974&partnerID=8YFLogxK

U2 - 10.1109/ICDM.2013.49

DO - 10.1109/ICDM.2013.49

M3 - Conference article

AN - SCOPUS:84894679974

SP - 979

EP - 984

JO - Proceedings - IEEE International Conference on Data Mining, ICDM

JF - Proceedings - IEEE International Conference on Data Mining, ICDM

SN - 1550-4786

M1 - 6729585

ER -