Visualizing Deep Neural Network Decisions: Prediction Difference Analysis
- Submitting institution
-
University of Glasgow
- Unit of assessment
- 11 - Computer Science and Informatics
- Output identifier
- 11-09960
- Type
- E - Conference contribution
- DOI
-
-
- Title of conference / published proceedings
- ICLR 2017 5th International Conference on Learning Representations
- First page
- 1
- Volume
- -
- Issue
- -
- ISSN
- -
- Open access status
- -
- Month of publication
- April
- Year of publication
- 2017
- URL
-
http://eprints.gla.ac.uk/214152/
- 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
-
3
- Research group(s)
-
-
- Citation count
- -
- Proposed double-weighted
- No
- Reserve for an output with double weighting
- No
- Additional information
- ORIGINALITY: presents a novel efficient method for visualizing the response of a deep neural network to a specific input, improving on previous methods with a more powerful conditional, multivariate model. SIGNIFICANCE: This is a leading method in explainable AI, explaining deep network classification decisions, and is highly cited. ICLR is a premier machine learning conference. RIGOUR: New method based on probabilistic approximations. Pixel relevance is efficiently evaluated by visualising the change in classification, had this particular pixel not been in the image.
- Author contribution statement
- -
- Non-English
- No
- English abstract
- -