June 1, 2000Neural Computation
Separating Style and Content with Bilinear Models
Authors
Discussion
Loading...
Member takes
Implication
Randomized trial demonstrates a new model for separating content from style in perceptual tasks, suggesting improved computational approaches.
Key Points
- Explore a computational model to effectively separate content from style in perceptual observations.
- Developed a bilinear model framework for two-factor tasks.
- Utilized singular value decomposition and expectation-maximization for efficient learning algorithms.
- Tested the model on spoken vowel classification, font extrapolation, and face translation under novel conditions.
- Achieved promising performance in spoken vowel classification with multi-speaker data.
- Successfully extrapolated fonts to unseen letters.
- Effectively translated face appearances under varying illumination conditions.
Cite This Study
Tenenbaum et al. (2000) studied this question.