A Relational Model for One-Shot Classification

Arturs Polis; Alexander Ilin

doi:10.14428/esann/2021.es2021-75

A Relational Model for One-Shot Classification

Research output: Chapter in Book/Report/Conference proceeding › Conference article in proceedings › Scientific › peer-review

Abstract

We show that a deep learning model with built-in relational inductive bias can bring benefits to sample-efficient learning, without rely-ing on extensive data augmentation. The proposed one-shot classification model performs relational matching of a pair of inputs in the form of local and pairwise attention. Our approach solves perfectly the one-shot image classification Omniglot challenge. Our model exceeds human level accuracy, as well as the previous state of the art, with no data augmentation.

Original language	English
Title of host publication	ESANN 2021 proceedings
Publisher	i6doc.com
Pages	647-652
ISBN (Electronic)	9782875870827
DOIs	https://doi.org/10.14428/esann/2021.es2021-75
Publication status	Published - 2021
MoE publication type	A4 Conference publication
Event	European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning - Virtual, Online, Bruges, Belgium Duration: 6 Oct 2021 → 8 Oct 2021 Conference number: 29

Conference

Conference	European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning
Abbreviated title	ESANN
Country/Territory	Belgium
City	Bruges
Period	06/10/2021 → 08/10/2021

Access to Document

10.14428/esann/2021.es2021-75

https://arxiv.org/abs/2111.04313Licence: CC BY

Cite this

@inproceedings{c8c85ae777c646b5bc995eca3e0d86e1,

title = "A Relational Model for One-Shot Classification",

abstract = "We show that a deep learning model with built-in relational inductive bias can bring benefits to sample-efficient learning, without rely-ing on extensive data augmentation. The proposed one-shot classification model performs relational matching of a pair of inputs in the form of local and pairwise attention. Our approach solves perfectly the one-shot image classification Omniglot challenge. Our model exceeds human level accuracy, as well as the previous state of the art, with no data augmentation.",

author = "Arturs Polis and Alexander Ilin",

year = "2021",

doi = "10.14428/esann/2021.es2021-75",

language = "English",

pages = "647--652",

booktitle = "ESANN 2021 proceedings",

publisher = "i6doc.com",

address = "Belgium",

note = "European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning, ESANN ; Conference date: 06-10-2021 Through 08-10-2021",

}

TY - GEN

T1 - A Relational Model for One-Shot Classification

AU - Polis, Arturs

AU - Ilin, Alexander

N1 - Conference code: 29

PY - 2021

Y1 - 2021

N2 - We show that a deep learning model with built-in relational inductive bias can bring benefits to sample-efficient learning, without rely-ing on extensive data augmentation. The proposed one-shot classification model performs relational matching of a pair of inputs in the form of local and pairwise attention. Our approach solves perfectly the one-shot image classification Omniglot challenge. Our model exceeds human level accuracy, as well as the previous state of the art, with no data augmentation.

AB - We show that a deep learning model with built-in relational inductive bias can bring benefits to sample-efficient learning, without rely-ing on extensive data augmentation. The proposed one-shot classification model performs relational matching of a pair of inputs in the form of local and pairwise attention. Our approach solves perfectly the one-shot image classification Omniglot challenge. Our model exceeds human level accuracy, as well as the previous state of the art, with no data augmentation.

UR - http://dx.doi.org/10.14428/esann/2021.es2021-75

UR - https://arxiv.org/abs/2111.04313

U2 - 10.14428/esann/2021.es2021-75

DO - 10.14428/esann/2021.es2021-75

M3 - Conference article in proceedings

SP - 647

EP - 652

BT - ESANN 2021 proceedings

PB - i6doc.com

T2 - European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning

Y2 - 6 October 2021 through 8 October 2021

ER -

A Relational Model for One-Shot Classification

Abstract

Conference

Access to Document

Other files and links

Fingerprint

Science-IT

Cite this

A Relational Model for One-Shot Classification

Abstract

Conference

Access to Document

Other files and links

Fingerprint

Equipment

Science-IT

Cite this