wicwiu/wicwiu ↗
Created Dec 24, 2020 · View the wicwiu/wicwiu repository page
WICWIU(What I can Create is What I Understand)
Want this for your repo?
Render a free sample of any GitHub repo in seconds.
Contributors
24
Lines of Code
2,143
From
May 11, 2018
To
Dec 21, 2020
About wicwiu/wicwiu
WICWIU is a deep learning framework written entirely in C++ and released as the first open-source deep learning framework from a South Korean university. The name draws inspiration from Richard Feynman's philosophy, representing the idea that "what we create ourselves, we understand." By implementing all APIs in C++, the framework prioritizes memory efficiency and performance optimization, and allows users to modify and extend the framework itself for specialized environments.
The framework is composed of several core components including Tensor and Shape classes for data storage and management up to five dimensions, Operator classes for forward and backward propagation, Module classes for building complex neural networks through combinations of operators, and Loss Functions and Optimizers for training. It supports various neural network layers, graph-structured network architectures, readable low-level CPU code, high-performance GPU parallel computation through cuDNN, and includes tutorials for MNIST, CIFAR-10, and ImageNet datasets. The project supports Linux (both CPU and GPU), and Windows CPU environments.
WICWIU is released under the Apache 2.0 license, making it freely available for both research and commercial use. The framework was developed by students at Handong University under the guidance of Professor Kim In-joong, with the project now in its seventh iteration involving multiple student teams. Documentation in Korean is included as comments throughout the code, with more comprehensive Korean documentation planned for future releases.