Skip to content

Latest commit

 

History

History
13 lines (13 loc) · 684 Bytes

README.md

File metadata and controls

13 lines (13 loc) · 684 Bytes

Breeze

A super light-weight CPU deep learning framework with no external dependency.

Implemented features

  1. 2/3D batched convolution & transposed convolution and their backprop calculations
  2. batch normalization and its backprop calculation
  3. 2/3D max ppoling & average pooling (including backprop)
  4. MLP (linear transform) layer
  5. simple activation layers (ReLU, LeakyReLU, softmax)
  6. cross entropy loss, Dice loss, mean square error (MSE) loss
  7. optimized convolution operations (AVX+OpenMP) and high performance matrix multiplication
  8. a simple builtin tensor library written from scratch
  9. a simple mechanic to save & load tensor data (serialize & deserialize)