Skip to content

CUDA-fication of CapsuleNetwork's Dynamic Routing - #3

Open
TennisGazelle wants to merge 37 commits into
masterfrom
CUDAify
Open

CUDA-fication of CapsuleNetwork's Dynamic Routing#3
TennisGazelle wants to merge 37 commits into
masterfrom
CUDAify

Conversation

@TennisGazelle

Copy link
Copy Markdown
Owner
  • CUDA enabled from cmake
  • UnifiedBlob classes are abstraction of arrays of vectors in many shapes
  • vector matrix multiplication for FP.

…cuda layer (for network) started; timing unit tests made (for simple timings); FP (and little BP) using cuda kernels done using seq conv. layer (bad timing);
…nd squashed bugs in par_kernels, fails over 3 tensor channels; removed unnecessary class, added Device Timer, added timing structs and methods in main
…fault sizes; added user side menu system; included host data for DATA; updated ProgressBar feel
…train); updated default GA and Caps parameter; added (redundant) mutex with (not working) threading; cuda Nan checking; updated chromosome for bounds (and size)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant