News
Latest news

Home > Latest news

Baidu why open source depth of machine learning platform?

On May 20, baidu open its depth on making machine learning platform. This release depth of machine learning open source platform belongs to the "au" open source organization, its core developers from baidu deep learning institute (IDL), Microsoft Asia research institute, university of Washington, New York university, the Hong Kong university of science and technology, Carnegie meilong university and other famous companies and universities.


Through the open source platform, developers can get free from all over the world better and easier to use the machine learning algorithm source code, thus significantly reducing development and deployment of distributed machine learning system and related application threshold. Including headlines now and the car home, several companies have benefited by the open platform.


As the layout of the older players in ai, baidu has the strength of the leading industry. Baidu deep learning institute (IDL) in 2013 to create and Andrew Ng joining in 2014, baidu DMLC distributed depth of machine learning open source projects (hereinafter referred to as "the union") has been done to explore in the field of multiple applications of deep learning, launched such as xgboost (speed Boosting model) with good effect, CXXNET (extreme deep learning c + + library), Minerva, highly efficient and flexible parallel deep learning engine) and the Parameter Server (one hour training data 600 t), and other products, in speech recognition, OCR recognition, face recognition and computing efficiency released more mature products.


With a series of lead baidu chose to open source the depth of machine learning platform, why clarificaiton will own the core technology?


Careful comparison, it is helpless in the face of the industry of foresight.


Embracing the world: the charm of open source


Open source, just as its name implies, is to open its source code to others access to and use, though it seems silly, but many historical facts tell us embracing the world also can let the world embrace you.


Counter attack is evidence of Android, iOS occupy the vast majority of the era of intelligent mobile phone operating system, Google's open source, having already no chance in front of the apple industry saw hope, since Android has more than half the market share, so that in the mobile Internet is popular today, Google can rely on the integration of it and its products to build their own powerful ecosystem. Born too, by contrast, the story of Windows mobile's hair awake enough, finally have to sponsor to Microsoft compatible with Android and iOS applications. In addition, as Microsoft's main application framework, on Microsoft. Net, hope it can open source to all platforms, and then repeat the JAVA of glorious history. It also proved that even by strong like Microsoft cannot let their products get the embrace of the world.


High flame: many materials with platform netting the power of the world


To realize the artificial intelligence, machine must have the ability to autonomous learning and deep learning as a major breakthrough in the field of machine learning, can be highly abstract characteristics of artificial intelligence task, such as natural language understanding complex scenes, and praised by the industry. Machine learning field, of course, not only have deep learning that the existence of an algorithm, deep learning, however, has obvious advantages: in the case of a data set is large enough, the depth of the study has the best prediction ability. Although still exists on the choice of algorithm "kill the goose that how to use" wheel ", but deep learning algorithm for artificial intelligence into strong ability is beyond compare with other algorithm; At the same time, with the mature of deep learning technology, a lot of traditional machine learning algorithms of elimination is almost a certainty. However, as the foregoing, in the demand of large amounts of data and its deep learning is still the development of the complexity of the biggest obstacles on the road, is also the helplessness of the industry.


Baidu power earlier in this field, and carried on the thorough research in many directions, using in-depth study combined with its own search engine big data let machine translation and natural language technology to realize a new leap forward. However, when available to a wider and more specific application scenarios, any company is difficult to meet all the requirements. Not only can draw lessons from history, open a win-win of doing things, also let the industry and healthy development of the ecological system.


Baidu, the open source is completely for a deal in Taiwan. Fully using c + + language to build the core, stable and efficient operation of the platform laid a foundation; Covering the three kinds of the most commonly used machine learning algorithms, including the sparse linear model is used to click, the decision tree model for sorting, and in-depth study, meet the needs of the masses; Key development "wormhole" project will automatically build deep union all projects, provide consistent data flow support for all of the components and provide including Amazon EC2, Microsoft Azure, Google Compute Engine, cloud computing platform compatibility support, reduce the barriers to entry platform.


The open source into in making further underscored its openness and confidence in the platform. A series of actions is wants to let developers will be able to obtain more deep learning algorithm source code quality and easier to use, to reduce the development and deployment into the threshold of the learning system and the related application, and then use the power of the world grow.


In the face of the outbreak of the predictable, mature technology gradually developed, to a certain extent, but also because of their respective gene and route technology, facing the great divide. An open mind to embrace the world, to build a international standards in the field of machine learning, baidu's move is not only for recycling in prophase research security, more friends hand in hand to win business, to ensure that their own voice.