A CRuby and Fluentd committer working at DeNA. Recently working on development of DNN framework.
Fast Numerical Computing and Deep Learning in Ruby with Cumo
Ruby is far behind than other languages such as Python in scientific computing. One reason is because there is no fast numerical library in Ruby.
I have made a high speed numerical library for Ruby using CUDA named "Cumo". In this talk, I describe:
- Overview of Scientific Computing in Ruby
- Basic knowledge of GPU programming
- Cumo inside, such as
- Writing fast CUDA kernel
- Implementing GPU memory pool
- JIT compiling user-defined kernel
- Performance comparison with Numo in an emerging DNN framework written in Ruby, red-chainer