Baidu peddles PaddlesPaddles, floats open source AI tech
Deep learning suite with Python front end
Since Amazon, Google, Microsoft, IBM, and World+Dog have one, it's no surprise that Chinese giant Baidu would pitch an artificial intelligence offering at the waiting world.
Right now, it describes PaddlePaddle – Parallel Distributed Deep Learning with a few extra capitals added – as a pre-release version. The real thing will land September 30, with “full document and installation support”, the company says at its GitHub repo.
PaddlePaddle's particular pitch is Python, which it uses to expose its API. This, the company hopes, will make the deep learning system easier to use and train than other AIs.
Internally, Baidu says it's already used the framework for various operations including training a Chinese language speech-to-text system for broadcast transcription (and also for smartphone speech interfaces).
Other in-house uses have included “ad click-through rate (CTR) prediction, large-scale image classification, optical character recognition(OCR), search ranking, computer virus detection, [and] recommendation”.
The core system is C++ for speed, with SWIG providing the Python wrapper. Users can program directly in C++, but if that's outside the skill set, there's the PyDataProvider2 library to help ease the training process.
The system supports multiple CPUs or GPUs (the GitHub readme doesn't stipulate how many).
PaddlePaddle's promo site is here. ®