David S. Barbera

Project · 2018

Neural Inference

A light-weight header-mostly Neural Network framework for on-device Inference on CPU.

History

The original purpose of this framework was to provide inference for neural networks on mobile devices at at a time when there were close to none. It was implemented from scratch in C++ relying on Eigen’s Library. It is completely cross-platform and currently works on Linux/Mac/Windows/iOS/Android.

Philosophy

The idea is to keep the simplicity of the original approach while keeping efficiency and low-latency at its core.

  • Fast compilation
  • Mathematically easy to read
  • Completely cross-platform
  • Didactic: to help people learn all the ‘nuts and bolts’ of ML, inference, mobile devices, plugin interfaces and framework creation

← All projects