projUNN: efficient method for training deep networks with unitary matrices.
Bobak Toussi KianiRandall BalestrieroYann LeCunSeth LloydPublished in: NeurIPS (2022)
Keyphrases
- high precision
- cost function
- high accuracy
- synthetic data
- experimental evaluation
- highly efficient
- training process
- detection method
- clustering method
- training samples
- artificial neural networks
- support vector machine
- dynamic programming
- neural network
- feature set
- computationally efficient
- significant improvement
- pairwise
- preprocessing
- training algorithm
- high efficiency
- edge detection
- singular value decomposition
- objective function
- image sequences