Solving Raven's Progressive Matrices with Multi-Layer Relation Networks

Marius Jahrens; Thomas Martinetz

doi:10.1109/IJCNN48605.2020.9207319

Solving Raven's Progressive Matrices with Multi-Layer Relation Networks

Institute of Neuro- and Bioinformatics

Abstract

Raven's Progressive Matrices are a benchmark originally designed to test the cognitive abilities of humans. It has recently been adapted to test relational reasoning in machine learning systems. For this purpose the so-called Procedurally Generated Matrices dataset was set up, which is so far one of the most difficult relational reasoning benchmarks. Here we show that deep neural networks are capable of solving this benchmark, reaching an accuracy of 98.0 percent over the previous state-of-the-art of 62.6 percent by combining Wild Relation Networks with Multi-Layer Relation Networks and introducing Magnitude Encoding, an encoding scheme designed for late fusion architectures.

Original language	English
Title of host publication	2020 International Joint Conference on Neural Networks (IJCNN)
Publisher	IEEE
Publication date	07.2020
Article number	9207319
ISBN (Print)	978-1-7281-6927-9
ISBN (Electronic)	978-1-7281-6926-2
DOIs	https://doi.org/10.1109/IJCNN48605.2020.9207319
Publication status	Published - 07.2020
Event	2020 International Joint Conference on Neural Networks - Virtual, Glasgow, United Kingdom Duration: 19.07.2020 → 24.07.2020 Conference number: 163566

Research Areas and Centers

Research Area: Intelligent Systems
Centers: Center for Artificial Intelligence Luebeck (ZKIL)

Access to Document

10.1109/IJCNN48605.2020.9207319

Cite this

@inproceedings{fe6cf5e056424478ab98e869e3e89dcb,

title = "Solving Raven's Progressive Matrices with Multi-Layer Relation Networks",

abstract = "Raven's Progressive Matrices are a benchmark originally designed to test the cognitive abilities of humans. It has recently been adapted to test relational reasoning in machine learning systems. For this purpose the so-called Procedurally Generated Matrices dataset was set up, which is so far one of the most difficult relational reasoning benchmarks. Here we show that deep neural networks are capable of solving this benchmark, reaching an accuracy of 98.0 percent over the previous state-of-the-art of 62.6 percent by combining Wild Relation Networks with Multi-Layer Relation Networks and introducing Magnitude Encoding, an encoding scheme designed for late fusion architectures.",

author = "Marius Jahrens and Thomas Martinetz",

note = "Publisher Copyright: {\textcopyright} 2020 IEEE. Copyright: Copyright 2020 Elsevier B.V., All rights reserved.; 2020 International Joint Conference on Neural Networks, IJCNN 2020 ; Conference date: 19-07-2020 Through 24-07-2020",

year = "2020",

month = jul,

doi = "10.1109/IJCNN48605.2020.9207319",

language = "English",

isbn = "978-1-7281-6927-9",

series = "Proceedings of the International Joint Conference on Neural Networks",

publisher = "IEEE",

booktitle = "2020 International Joint Conference on Neural Networks (IJCNN)",

}

Jahrens, M & Martinetz, T 2020, Solving Raven's Progressive Matrices with Multi-Layer Relation Networks. in 2020 International Joint Conference on Neural Networks (IJCNN)., 9207319, Proceedings of the International Joint Conference on Neural Networks, IEEE, 2020 International Joint Conference on Neural Networks, Glasgow, United Kingdom, 19.07.20. https://doi.org/10.1109/IJCNN48605.2020.9207319

Solving Raven's Progressive Matrices with Multi-Layer Relation Networks. / Jahrens, Marius; Martinetz, Thomas.

2020 International Joint Conference on Neural Networks (IJCNN). IEEE, 2020. 9207319 (Proceedings of the International Joint Conference on Neural Networks).

Research output: Chapters in Books/Reports/Conference Proceedings › Conference contribution › peer-review

TY - GEN

T1 - Solving Raven's Progressive Matrices with Multi-Layer Relation Networks

AU - Jahrens, Marius

AU - Martinetz, Thomas

N1 - Conference code: 163566

PY - 2020/7

Y1 - 2020/7

N2 - Raven's Progressive Matrices are a benchmark originally designed to test the cognitive abilities of humans. It has recently been adapted to test relational reasoning in machine learning systems. For this purpose the so-called Procedurally Generated Matrices dataset was set up, which is so far one of the most difficult relational reasoning benchmarks. Here we show that deep neural networks are capable of solving this benchmark, reaching an accuracy of 98.0 percent over the previous state-of-the-art of 62.6 percent by combining Wild Relation Networks with Multi-Layer Relation Networks and introducing Magnitude Encoding, an encoding scheme designed for late fusion architectures.

AB - Raven's Progressive Matrices are a benchmark originally designed to test the cognitive abilities of humans. It has recently been adapted to test relational reasoning in machine learning systems. For this purpose the so-called Procedurally Generated Matrices dataset was set up, which is so far one of the most difficult relational reasoning benchmarks. Here we show that deep neural networks are capable of solving this benchmark, reaching an accuracy of 98.0 percent over the previous state-of-the-art of 62.6 percent by combining Wild Relation Networks with Multi-Layer Relation Networks and introducing Magnitude Encoding, an encoding scheme designed for late fusion architectures.

UR - https://www.scopus.com/record/display.uri?eid=2-s2.0-85093830225&origin=inward&txGid=a8da8a941bf9a97db453332478f0a07a

U2 - 10.1109/IJCNN48605.2020.9207319

DO - 10.1109/IJCNN48605.2020.9207319

M3 - Conference contribution

SN - 978-1-7281-6927-9

T3 - Proceedings of the International Joint Conference on Neural Networks

BT - 2020 International Joint Conference on Neural Networks (IJCNN)

PB - IEEE

T2 - 2020 International Joint Conference on Neural Networks

Y2 - 19 July 2020 through 24 July 2020

ER -

Solving Raven's Progressive Matrices with Multi-Layer Relation Networks

Abstract

Research Areas and Centers

Access to Document

Other files and links

Fingerprint

Cite this