Detecting Anomalous Data Using Auto-Encoders
- Submitting institution
-
University College London
- Unit of assessment
- 11 - Computer Science and Informatics
- Output identifier
- 14547
- Type
- E - Conference contribution
- DOI
-
10.18178/ijmlc.2016.6.1.565
- Title of conference / published proceedings
- International Journal of Machine Learning and Computing
- First page
- 21
- Volume
- 6
- Issue
- 1
- ISSN
- 2010-3700
- Open access status
- Out of scope for open access requirements
- Month of publication
- February
- Year of publication
- 2016
- 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
-
2
- Research group(s)
-
-
- Citation count
- -
- Proposed double-weighted
- No
- Reserve for an output with double weighting
- No
- Additional information
- Anomaly Detection (AD) methods struggle with high-dimensional representations (every datum is unique). Effective low-dimensional representations can be learnt or engineered, but this is illegitimate if anomalous data is made use of (even implicitly). Here we make the novel proposal to use an auto-encoder network to learn a representation from the normal data alone by requiring it to be low-dimensional but adequate to reconstruct the raw data. Experiments on a range of problems (including real-world X-ray cargo container images) showed that this approach is effective for AD, and an improvement over the reconstruction error approach that has previously been proposed.
- Author contribution statement
- -
- Non-English
- No
- English abstract
- -