MultiAdam: Parameter-wise Scale-invariant Optimizer for Multiscale Training of Physics-informed Neural Networks.
Jiachen YaoChang SuZhongkai HaoSongming LiuHang SuJun ZhuPublished in: ICML (2023)
Keyphrases
- scale invariant
- multiscale
- neural network
- affine invariant
- scale space
- natural images
- keypoints
- scale invariance
- single scale
- local binary pattern
- wavelet transform
- interest point detectors
- feature transformation
- image segmentation
- pairwise
- training set
- image structure
- feature detectors
- image processing
- scale selection
- image representation
- edge detection
- back propagation
- pattern recognition
- translation and scale invariant
- texture images
- scale space theory
- scale invariant features