Google has taken a giant step to enable further advancements of artificial intelligence.
The company announced on Monday that it is open-sourcing its second-generation machine-learning system to outside developers.
There is no denying that learning systems have made it possible to create and improve apps when it comes to speech and image recognition technologies. Just take Google Photos for example, which was built using the company's machine-learning system, called DisBelief.
Developed in 2011, DisBelief has helped Google build large neural networks, but it has its limitations, including difficult configurations and its inability to share code externally.
As a result, the company has open-sourced TensorFlow, which was designed to fix the shortcomings of DisBelief. However, it's important to note that it only allows for part of the AI engine to be open-sourced.
TensorFlow is built-in support for deep learning that is said to be twice as fast as Google's previous machine-learning system, and it is much easier to use. The software allows developers to express their ideas in a flow graph using the flexible Python and C++ interface so that they can try out their ideas, and if these ideas work, they can move them directly into products without having to rewrite code.
It serves as one centralized tool for researchers in machine-learning and people who develop real products, enabling collaboration and communication, so that a researcher can develop an idea to then send the code that can be applied for by engineers somewhere else.
Google is providing outside developers with tools, tutorials and examples under the Apache 2.0 license. The machine-learning system is targeted at researchers, developers, students, engineers, inventors and the like.
TensorFlow runs on CPUs or GPUs, desktop, servers and mobile platforms. While the system is not completed, Google will continue to make the system better with a little help from outsiders with the release of the source code.
Source: Google Research Blog
