Critical Points Of An Autoencoder Can Provably Recover Sparsely Used Overcomplete Dictionaries.
Akshay RangamaniAnirbit MukherjeeAshish AroraTejaswini GanapathyAmitabh BasuSang Peter ChinTrac D. TranPublished in: CoRR (2017)
Keyphrases
- critical points
- sparse representation
- scale space
- vector field
- sparse coding
- face recognition
- image classification
- gaussian scale space
- dimensionality reduction
- signal processing
- morse theory
- saddle points
- deep structure
- worst case
- image representation
- reference points
- high dimensional data
- gaussian filter
- natural images
- feature points
- multiscale