Amazon open machine learning system source code: Challenges Google TensorFlow

Amazon open machine learning system source code: Challenges Google TensorFlow
Amazon took a bigger step in the open-source technology field and announced the opening of the company's machine learning software DSSTNE source code. This latest project will compete with Google's TensorFlow, which was open-source last year. Amazon said that DSSTNE has excellent performance in the absence of a large amount of data to train machine learning systems, while TensorFlow is good at processing massive data.

According to Amazon, DSSTNE is faster than TensorFlow. When the data volume is small, the processing speed is 2.1 times that of the latter. Amazon developed this software to recommend products to users on its retail platform. To implement this function, you must use the help of neural network programs, but Amazon may not always be able to master a lot of data when making recommendations.

Amazon's system is able to achieve such a fast speed, partly because it leverages the performance of multiple GPUs. Unlike other open-source deep learning software, DSSTNE can allocate tasks among many GPUs without reducing the speed and accuracy.

The dsstne faq section states: "This means that you only need to provide recommendations with 10 million unique products, instead of using hundreds of millions of products. To solve this problem, other software needs to use the computing performance of the CPU to analyze sparse data, resulting in an order of magnitude reduction in performance ,"

The reason why Amazon opened the source code of this software is that machine learning is not limited to speech and language recognition, but can be expanded to search, recommendation, and other functions.

"We want researchers from around the world to improve it together," wrote DSSTNE's FAQ. "But more importantly, we want it to inspire innovation in more areas."

Developers can download DSSTNE from GitHub Based on Apache 2.0 open source license.

From: http://www.cnbeta.com/articles/501907.htm

Address: http://www.linuxprobe.com/amazon-google-tensorflow.html