Emergence of the SVD as an interpretable factorization in deep learning for inverse problems.
Shashank SuleRichard G. SpencerWojciech CzajaPublished in: CoRR (2023)
Keyphrases
- deep learning
- inverse problems
- singular value decomposition
- image reconstruction
- global optimization
- convex optimization
- unsupervised learning
- optimization problems
- optimization methods
- early vision
- dimensionality reduction
- weakly supervised
- principal component analysis
- machine learning
- partial differential equations
- evolutionary algorithm
- mental models
- higher order
- parameter space
- super resolution
- feature space
- feature extraction