News classification
Contact us
- Add: No. 9, North Fourth Ring Road, Haidian District, Beijing. It mainly includes face recognition, living detection, ID card recognition, bank card recognition, business card recognition, license plate recognition, OCR recognition, and intelligent recognition technology.
- Tel: 13146317170 廖经理
- Fax:
- Email: 398017534@qq.com
A debate on the framework of deep learning
A debate on the framework of deep learning
Over the past two years, companies or research institutions have launched their own deep learning frameworks, such as Caffe, TensorFlow, etc., and the framework related to deep learning is changing over time. Theano is the first widely used deep learning framework, which is led by the Daniel Yoshua Bengio in the field of deep learning and created by MILA. However, in September this year, MILA announced that it would not continue to develop the framework after the final version of Theano was updated in 2018. This news was not very upset. In the past few years, some different open source Python deep learning frameworks have been introduced. These deep learning frameworks are usually developed by a large technology company or from a number of companies.
640? Wx_fmt=png&wxfrom=5&wx_lazy=1
.
At the same time, TensorFlow developed by Google Corporation seems to be the most commonly used deep learning framework. Some experts predict that Google's TensorFlow will lead the deep learning framwork market for many years. However, some other types of deep learning frameworks have gradually attracted more and more users, and the most notable is the PyTorch deep learning framework. PyTorch is a deep learning framework developed by Facebook in January 2017. The framework is implemented in C language and Lua language, and GPU is accelerated by Python language. In addition to GPU acceleration and efficient memory usage, the other reason why PyTorch is popular is the use of its dynamic computation diagram, which has been used by other non mainstream deep learning frameworks, such as Chainer. The advantage of using these dynamic computation diagrams is that the graph is defined by running, rather than the traditional "definition and operation" mode, especially in the case of variable input, such as textual, unstructured data.
0? Wx_fmt=gif
Other tech giants are not standing still, Microsoft Corp has developed a deep learning within the framework of CNTK, and in 2017 officially launched its first version 2 and renamed Microsoft Cognitive Toolkit. In the same year, Facebook also launched Caffe2, which is the heir to the Caffe framework of all known weeks. The original Caffe framework is developed by visual and Learning Center at Berkeley, is still widely used in the field of computer vision, and some mature model parameters can be found in the Model Zoo, these model parameters can be used to transfer learning, initialization of network parameters. At the moment, Caffe2 has not been able to keep up with Caffe's footsteps.
Another popular deep learning framework is MXNet, which is supported by the two giants of Microsoft and Amazon. MXNet has been launched for some time, but when it comes to the deep learning framework MXNet, it is often mistaken for the framework for the R language. In fact, MXNet supports a variety of languages, including not only the R language, but also other languages, such as C++, Python, JavaScript, and Go. The advantage of MXNet lies in its extensibility and its high performance.
These are just selected some widely applied deep learning frameworks, and some other open source deep learning frameworks, such as Deeplearning4j and Dlib (based on C++ language). In addition, Sonnet, a high - level object - oriented library based on TensorFlow, was released by Google's DeepMind in 2017. Other deep learning frameworks worth mentioning are H2o.ai and Spark.
0? Wx_fmt=png
In all of the deep learning frameworks, there are some framework interfaces that need to be described in detail. As the most widely used interface undoubtedly known as well as the application of Keras, Keras application program interface with a high level of deep learning written by Python (API), created by researchers at the Google Corporation Fran OIS Chollet C. In addition, Google Corporation announced in 2017 that Keras has been selected as the advanced API of TensorFlow, which means that Keras will be included in the next version of TensorFlow. In addition to TensorFlow, Keras can also be used in Theano or CNTK.
The power of Keras is that it can create a deep learning model simply by overlaying multiple layers. When using Keras, users do not need to do mathematical operations behind the layer. It looks like an ideal rapid prototyping machine, and Keras has also become a hot tool in the Kaggle competition.
Therefore, on the one hand, there is a high-level Keras API on the one hand, which allows you to build advanced deep learning models easily. On the other hand, there are low level TensorFlow frameworks that enable modeling to become more flexible. All two projects are supported by Google Corporation. As expected, other companies are also unwilling. Microsoft and Amazon jointly announced their Gluon API, Gluon is a high-level Python deep learning interface, currently supporting the MXNet framework, will soon support Microsoft's CNTK framework. Gluon is a direct competitor of Keras. Although AWS (Amazon Co's cloud computing service platform) claims that they strongly support all the deep learning frameworks, AWS will certainly support Gluon's competition in the AI field.
0? Wx_fmt=png
Is the biggest competitor beyond all expectations, the TensorFlow framework actually seems to be PyTorch, the growing research interest in the framework of the PyTorch community, for example, in a recent Kaggle contest, users often choose to use the PyTorch framework as part of the solution, but also be used in the research paper in the latest. In the latest TensorFlow version of the Google Corporation in October 2017, a "run after definition" interface was introduced. The Google Corporation hopes the release will win back.
640? Wx_fmt=png&wxfrom=5&wx_lazy=1
.
At the same time, TensorFlow developed by Google Corporation seems to be the most commonly used deep learning framework. Some experts predict that Google's TensorFlow will lead the deep learning framwork market for many years. However, some other types of deep learning frameworks have gradually attracted more and more users, and the most notable is the PyTorch deep learning framework. PyTorch is a deep learning framework developed by Facebook in January 2017. The framework is implemented in C language and Lua language, and GPU is accelerated by Python language. In addition to GPU acceleration and efficient memory usage, the other reason why PyTorch is popular is the use of its dynamic computation diagram, which has been used by other non mainstream deep learning frameworks, such as Chainer. The advantage of using these dynamic computation diagrams is that the graph is defined by running, rather than the traditional "definition and operation" mode, especially in the case of variable input, such as textual, unstructured data.
0? Wx_fmt=gif
Other tech giants are not standing still, Microsoft Corp has developed a deep learning within the framework of CNTK, and in 2017 officially launched its first version 2 and renamed Microsoft Cognitive Toolkit. In the same year, Facebook also launched Caffe2, which is the heir to the Caffe framework of all known weeks. The original Caffe framework is developed by visual and Learning Center at Berkeley, is still widely used in the field of computer vision, and some mature model parameters can be found in the Model Zoo, these model parameters can be used to transfer learning, initialization of network parameters. At the moment, Caffe2 has not been able to keep up with Caffe's footsteps.
Another popular deep learning framework is MXNet, which is supported by the two giants of Microsoft and Amazon. MXNet has been launched for some time, but when it comes to the deep learning framework MXNet, it is often mistaken for the framework for the R language. In fact, MXNet supports a variety of languages, including not only the R language, but also other languages, such as C++, Python, JavaScript, and Go. The advantage of MXNet lies in its extensibility and its high performance.
These are just selected some widely applied deep learning frameworks, and some other open source deep learning frameworks, such as Deeplearning4j and Dlib (based on C++ language). In addition, Sonnet, a high - level object - oriented library based on TensorFlow, was released by Google's DeepMind in 2017. Other deep learning frameworks worth mentioning are H2o.ai and Spark.
0? Wx_fmt=png
In all of the deep learning frameworks, there are some framework interfaces that need to be described in detail. As the most widely used interface undoubtedly known as well as the application of Keras, Keras application program interface with a high level of deep learning written by Python (API), created by researchers at the Google Corporation Fran OIS Chollet C. In addition, Google Corporation announced in 2017 that Keras has been selected as the advanced API of TensorFlow, which means that Keras will be included in the next version of TensorFlow. In addition to TensorFlow, Keras can also be used in Theano or CNTK.
The power of Keras is that it can create a deep learning model simply by overlaying multiple layers. When using Keras, users do not need to do mathematical operations behind the layer. It looks like an ideal rapid prototyping machine, and Keras has also become a hot tool in the Kaggle competition.
Therefore, on the one hand, there is a high-level Keras API on the one hand, which allows you to build advanced deep learning models easily. On the other hand, there are low level TensorFlow frameworks that enable modeling to become more flexible. All two projects are supported by Google Corporation. As expected, other companies are also unwilling. Microsoft and Amazon jointly announced their Gluon API, Gluon is a high-level Python deep learning interface, currently supporting the MXNet framework, will soon support Microsoft's CNTK framework. Gluon is a direct competitor of Keras. Although AWS (Amazon Co's cloud computing service platform) claims that they strongly support all the deep learning frameworks, AWS will certainly support Gluon's competition in the AI field.
0? Wx_fmt=png
Is the biggest competitor beyond all expectations, the TensorFlow framework actually seems to be PyTorch, the growing research interest in the framework of the PyTorch community, for example, in a recent Kaggle contest, users often choose to use the PyTorch framework as part of the solution, but also be used in the research paper in the latest. In the latest TensorFlow version of the Google Corporation in October 2017, a "run after definition" interface was introduced. The Google Corporation hopes the release will win back.