Abstract

In modern machine learning,raw data is the preferred input for our models. Where a decade ago data scientists were still engineering features, manually picking out the details we thought salient, they now prefer the data in their raw form. As long as we can assume that all relevant and irrelevant information is present in the input data, we can design deep models that build up intermediate representations to sift out relevant features. However, these models are often domain specific and tailored to the task at hand, and therefore unsuited for learning on heterogeneous knowledge: information of different types and from different domains. If we can develop methods that operate on this form of knowledge, we can dispense with a great deal more ad-hoc feature engineering and train deep models end-to-end in many more domains. To accomplish this, we first need a data model capable of expressing heterogeneous knowledge naturally in various domains, in as usable a form as possible, and satisfying as many use cases as possible. We argue that the knowledge graph is a suitable candidate for this data model. We further discuss some of the promises and challenges of this approach, and how we are currently broadening our efforts to multi-modal knowledge graphs.

Conference

ConferenceICT Open 2018
Abbreviated titleICT Open
CountryNetherlands
CityAmersfoort
Period19/03/1820/03/18
Internet address

Cite this

@conference{64f02e34e3a240238486b9894c6c4a17,
title = "The Knowledge Graph for End-to-End Learning on Heterogeneous Knowledge",
abstract = "In modern machine learning,raw data is the preferred input for our models. Where a decade ago data scientists were still engineering features, manually picking out the details we thought salient, they now prefer the data in their raw form. As long as we can assume that all relevant and irrelevant information is present in the input data, we can design deep models that build up intermediate representations to sift out relevant features. However, these models are often domain specific and tailored to the task at hand, and therefore unsuited for learning on heterogeneous knowledge: information of different types and from different domains. If we can develop methods that operate on this form of knowledge, we can dispense with a great deal more ad-hoc feature engineering and train deep models end-to-end in many more domains. To accomplish this, we first need a data model capable of expressing heterogeneous knowledge naturally in various domains, in as usable a form as possible, and satisfying as many use cases as possible. We argue that the knowledge graph is a suitable candidate for this data model. We further discuss some of the promises and challenges of this approach, and how we are currently broadening our efforts to multi-modal knowledge graphs.",
keywords = "Knowledge Graphs, End-to-End Learning, Heterogeneous Knowledge, Multimodal Embeddings",
author = "W.X. Wilcke and P. Bloem and {de Boer}, Viktor",
year = "2018",
month = "3",

}

Wilcke, WX, Bloem, P & de Boer, V 2018, 'The Knowledge Graph for End-to-End Learning on Heterogeneous Knowledge' ICT Open 2018, Amersfoort, Netherlands, 19/03/18 - 20/03/18, .

The Knowledge Graph for End-to-End Learning on Heterogeneous Knowledge. / Wilcke, W.X.; Bloem, P.; de Boer, Viktor.

2018. Abstract from ICT Open 2018, Amersfoort, Netherlands.

Research output: ScientificAbstract

TY - CONF

T1 - The Knowledge Graph for End-to-End Learning on Heterogeneous Knowledge

AU - Wilcke,W.X.

AU - Bloem,P.

AU - de Boer,Viktor

PY - 2018/3/19

Y1 - 2018/3/19

N2 - In modern machine learning,raw data is the preferred input for our models. Where a decade ago data scientists were still engineering features, manually picking out the details we thought salient, they now prefer the data in their raw form. As long as we can assume that all relevant and irrelevant information is present in the input data, we can design deep models that build up intermediate representations to sift out relevant features. However, these models are often domain specific and tailored to the task at hand, and therefore unsuited for learning on heterogeneous knowledge: information of different types and from different domains. If we can develop methods that operate on this form of knowledge, we can dispense with a great deal more ad-hoc feature engineering and train deep models end-to-end in many more domains. To accomplish this, we first need a data model capable of expressing heterogeneous knowledge naturally in various domains, in as usable a form as possible, and satisfying as many use cases as possible. We argue that the knowledge graph is a suitable candidate for this data model. We further discuss some of the promises and challenges of this approach, and how we are currently broadening our efforts to multi-modal knowledge graphs.

AB - In modern machine learning,raw data is the preferred input for our models. Where a decade ago data scientists were still engineering features, manually picking out the details we thought salient, they now prefer the data in their raw form. As long as we can assume that all relevant and irrelevant information is present in the input data, we can design deep models that build up intermediate representations to sift out relevant features. However, these models are often domain specific and tailored to the task at hand, and therefore unsuited for learning on heterogeneous knowledge: information of different types and from different domains. If we can develop methods that operate on this form of knowledge, we can dispense with a great deal more ad-hoc feature engineering and train deep models end-to-end in many more domains. To accomplish this, we first need a data model capable of expressing heterogeneous knowledge naturally in various domains, in as usable a form as possible, and satisfying as many use cases as possible. We argue that the knowledge graph is a suitable candidate for this data model. We further discuss some of the promises and challenges of this approach, and how we are currently broadening our efforts to multi-modal knowledge graphs.

KW - Knowledge Graphs

KW - End-to-End Learning

KW - Heterogeneous Knowledge

KW - Multimodal Embeddings

M3 - Abstract

ER -