The Modern Mathematics of Deep Learningexplained for non-mathematicians

Concepts

Learned features

A representation expresses data using selected features.

Lessons

  1. Count nonzero entries
  2. Dictionary representation
  3. Thresholding coefficients
  4. Translation and Fourier magnitude
  5. Convolution and modulus cascade
  6. Convolutional dictionary
  7. Local sparsity
  8. Sparse hierarchy
  9. Composed threshold maps
  10. Scattering feature norm
  11. Shift and Fourier phase
  12. Fourier energy preservation
  13. Dictionary projection
  14. Deforming an input
  15. Thresholded coefficient approximation
  16. Check the printed threshold condition
  17. A multichannel signal window
  18. Local kernel support
  19. A zero-deformation comparison