Dynamical systems as temporal feature spaces
- Submitting institution
-
The University of Birmingham
- Unit of assessment
- 11 - Computer Science and Informatics
- Output identifier
- 91921657
- Type
- D - Journal article
- DOI
-
-
- Title of journal
- Journal of Machine Learning Research
- Article number
- 19-589
- First page
- 1
- Volume
- 21
- Issue
- 44
- ISSN
- 1532-4435
- Open access status
- Compliant
- Month of publication
- February
- Year of publication
- 2020
- URL
-
-
- Supplementary information
-
-
- Request cross-referral to
- -
- Output has been delayed by COVID-19
- No
- COVID-19 affected output statement
- -
- Forensic science
- No
- Criminology
- No
- Interdisciplinary
- No
- Number of additional authors
-
0
- Research group(s)
-
-
- Citation count
- 2
- Proposed double-weighted
- No
- Reserve for an output with double weighting
- No
- Additional information
- The paper presents a completely new framework for theoretical analysis of dynamical systems that form the core of many machine learning models designed to process time series data. This is important, because for the first time we are able to understand (sometimes counter-intuitive) phenomena previously observed empirically. The framework bridges the field of dynamical systems and the well studied and understood field of kernel machines in machine learning. This new viewpoint has the potential to generate a series of results bringing deep insight into how different architectures and parameterisations of recurrent neural networks affect their learning capabilities.
- Author contribution statement
- -
- Non-English
- No
- English abstract
- -