A Deep and Tractable Density Estimator
- Submitting institution
-
University of Edinburgh
- Unit of assessment
- 11 - Computer Science and Informatics
- Output identifier
- 58749726
- Type
- E - Conference contribution
- DOI
-
-
- Title of conference / published proceedings
- Proceedings of the 31st International Conference on Machine Learning
- First page
- 467
- Volume
- 32
- Issue
- 1
- ISSN
- 1938-7228
- Open access status
- Out of scope for open access requirements
- Month of publication
- June
- Year of publication
- 2014
- 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)
-
B - Data Science and Artificial Intelligence
- Citation count
- -
- Proposed double-weighted
- No
- Reserve for an output with double weighting
- No
- Additional information
- This was the first deep neural network that can evaluate normalized high-dimensional probability distributions, where training-time updates only require one GPU-friendly network pass. The training method was adapted for topic modelling and multimodal data (Zheng et al. TPAMI 2015, JMLR 2017), and Coconet, which generated Bach harmonizations for the first ever AI-powered "Google doodle". It is frequently compared against; Oord and Schrauwen (JMLR 15, 2014): "It is currently one of the few deep learning methods with good density estimation results on real-valued data and is the current state of the art on image patch modeling."
- Author contribution statement
- -
- Non-English
- No
- English abstract
- -