AI3SD Video: Inference from Medical Images: Subspaces for Low Data Regimes
AI3SD Video: Inference from Medical Images: Subspaces for Low Data Regimes
Unlike in the field of visual scene recognition, where tremendous advances have taken place due to the availability of very large datasets to train deep neural networks, inference from medical images is often hampered by the fact that only small amounts of data may be available. When working with very small dataset problems, of the order of a few hundred items of data, the power of deep learning may still be exploited by using a model pre-trained on natural images as a feature extractor and carrying out classic pattern recognition techniques in this feature space, the so-called few-shot learning problem. In regimes where the dimension of this feature space is comparable to or even larger than the number of items of data, dimensionality reduction is a necessity and is often achieved by principal component analysis or singular value decomposition (PCA/SVD). In this paper, noting the inappropriateness of using SVD for this setting we explore two alternatives based on non-negative matrix factorization (NMF) and discriminant analysis. Using 14 different datasets spanning 11 distinct disease types we demonstrate that at low dimensions, discriminant subspaces achieve significant improvements over SVD and the original feature space. We also show that at modest dimensions, NMF is a competitive alternative to SVD in this setting. Joint work with Jiahui Liu, Keqiang Fan and Xiaohao Cai.
Niranjan, Mahesan
5cbaeea8-7288-4b55-a89c-c43d212ddd4f
Frey, Jeremy G.
ba60c559-c4af-44f1-87e6-ce69819bf23f
Kanza, Samantha
b73bcf34-3ff8-4691-bd09-aa657dcff420
1 March 2022
Niranjan, Mahesan
5cbaeea8-7288-4b55-a89c-c43d212ddd4f
Frey, Jeremy G.
ba60c559-c4af-44f1-87e6-ce69819bf23f
Kanza, Samantha
b73bcf34-3ff8-4691-bd09-aa657dcff420
Niranjan, Mahesan
(2022)
AI3SD Video: Inference from Medical Images: Subspaces for Low Data Regimes.
Frey, Jeremy G. and Kanza, Samantha
(eds.)
AI4SD Network+ Conference, Chilworth Manor , Southampton, United Kingdom.
01 - 03 Mar 2022.
(doi:10.5258/SOTON/AI3SD0190).
Record type:
Conference or Workshop Item
(Other)
Abstract
Unlike in the field of visual scene recognition, where tremendous advances have taken place due to the availability of very large datasets to train deep neural networks, inference from medical images is often hampered by the fact that only small amounts of data may be available. When working with very small dataset problems, of the order of a few hundred items of data, the power of deep learning may still be exploited by using a model pre-trained on natural images as a feature extractor and carrying out classic pattern recognition techniques in this feature space, the so-called few-shot learning problem. In regimes where the dimension of this feature space is comparable to or even larger than the number of items of data, dimensionality reduction is a necessity and is often achieved by principal component analysis or singular value decomposition (PCA/SVD). In this paper, noting the inappropriateness of using SVD for this setting we explore two alternatives based on non-negative matrix factorization (NMF) and discriminant analysis. Using 14 different datasets spanning 11 distinct disease types we demonstrate that at low dimensions, discriminant subspaces achieve significant improvements over SVD and the original feature space. We also show that at modest dimensions, NMF is a competitive alternative to SVD in this setting. Joint work with Jiahui Liu, Keqiang Fan and Xiaohao Cai.
Video
ai4sd_march_2022_day_1_MahesanNiranjan
- Version of Record
More information
Published date: 1 March 2022
Additional Information:
Mahesan Niranjan is a Professor of Electronics and Computer Science at Southampton. Prior to this, he worked at the University of Cambridge as a Lecturer in Information Engineering. And the University of Sheffield as a Professor of Computer Science, where he also served as Head of Computer Science and Dean of Engineering. Mahesan works in the area of machine learning, and his research interests are in the algorithmic and applied aspects of the subject. He has worked on a range of applications of machine learning and neural networks including speech and language processing, computer vision and computational finance. Currently, the major focus of his research is in computational biology. Alongside his duties at Southampton University, he also often travels to other international universities to present his research and teach intense short courses in Machine Learning.
Venue - Dates:
AI4SD Network+ Conference, Chilworth Manor , Southampton, United Kingdom, 2022-03-01 - 2022-03-03
Identifiers
Local EPrints ID: 468640
URI: http://eprints.soton.ac.uk/id/eprint/468640
PURE UUID: d820c054-c5d5-440e-a7bb-e05c94950cae
Catalogue record
Date deposited: 19 Aug 2022 16:35
Last modified: 17 Mar 2024 03:51
Export record
Altmetrics
Contributors
Author:
Mahesan Niranjan
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics