Recommended 10 open-source deep learning frameworks on GitHub

Recommended 10 open-source deep learning frameworks on GitHub

Recently, Google Open source TensorFlow (GitHub), the move in the field of deep learning impact, because Google in the field of artificial intelligence research achievements, has a strong talent pool, and Google's own Gmail and search engines are using a self-developed deep learning tool.

Undoubtedly, the TensorFlow from the Google arsenal is necessarily the star of the open source deep learning software, landing on GitHub that day became the most watched project, when the week to get the number of stars easily more than 10,000.

For developers who want to integrate deep learning capabilities into their applications, there are many good open source projects on GitHub, and first we recommend the TOP3 with the highest popularity today:

First, Caffe. Caffe from the University of California, Berkeley, is widely used, including Pinterest, a big web. Like TensorFlow, Caffe is also developed by C + +, and Caffe is also the basis for the Deepdream project that Google released earlier this year, which identifies the artificial AI neural network of the MEW.

Second, Theano. Born in 2008 at the Montreal Institute of Technology, Theano derived a great deal of deep learning Python software packages, most notably Blocks and Keras.

Third, Torch. Torch has been born for ten years, but the real benefit of Facebook was that last year a lot of Torch's deep learning modules and extensions were open source. Another special feature of Torch is the use of the less popular programming language Lua (which was used to develop video games).

In addition to the above three more mature and well-known projects, there are many features of deep learning open source framework is also worth attention:

Four, Brainstorm. A very promising deep learning package from the Swiss AI lab Idsia, Brainstorm can handle hundreds of super deep neural networks-the so-called road network Highway Networks.

Five, Chainer. A Python framework from a Japanese deep learning startup, Preferred Networks, released this June. Chainer's design is based on the define by run principle, which means that the network is dynamically defined in the run, not defined at startup, and there are chainer detailed documentation.

Six, deeplearning4j. As the name implies, Deeplearning4j is the "for Java" deep learning framework and the first commercial-level deep-learning open Source Library. Deeplearning4j, launched by Skymind in June 2014, uses DEEPLEARNING4J's many star companies such as Accenture, Chevrolet, and Bo's consulting and IBM.

Deeplearning4j is a high-maturity, deep-learning open Source library for production and commercial applications that integrates with Hadoop and Spark, Plug and Play, and enables developers to quickly integrate deep learning capabilities into the app for the following deep learning areas:

· Human Face/image recognition

· Voice Search

· Voice-to-text (Speech to text)

· Spam filtering (anomaly detection)

· E-commerce fraud detection

Seven, Marvin. is the new C + + framework of Princeton University's visual working Group. The team also provided a file for converting the Caffe model into idiom Marvin compatible mode.

Eight, Convnetjs. This is Dr. Andrej Karpathy, a PhD student at Stanford University, who develops a browser plugin that can train a neural network in your navigator based on Universal JavaScript. Karpathy also wrote a Convnetjs introductory tutorial, as well as a concise browser demo project.

Nine, MXNet. From Cxxnet, Minerva, purine and other projects of the developer's hand, mainly in C + + written. MXNet emphasizes the efficiency of memory usage and even the task of running image recognition on smartphones.

Ten, Neon. Nervana Systems, a startup company, open source this May, and in some benchmarks, Neon, developed by Python and Sass, is even better than caffeine, Torch, and Google TensorFlow.

Recommended 10 open-source deep learning frameworks on GitHub