Deep learning is a machine learning method based on empirical learning of data that has grown and gained popularity in recent years.
As a relatively new concept, there are plenty of learning resources at your fingertips for both beginners looking to get into the field, and veterans who already know the method.
In order not to be eliminated by the ever-changing technologies and trends, it is a good way to actively participate in the study and interaction of open source projects in the deep learning community.
In this article, Digest Bacteria will introduce the 16 most popular deep learning open source platforms and open source libraries in GitHub in detail. In addition, there are some relatively good platforms and frameworks. Although they are not on the list, Digest Bacteria is also listed. out for your reference.
The 16 open source deep learning frameworks with the highest collection and contribution rates on GitHub. The more green the circle is, the newer the framework is, and the more blue the color is, the earlier the frame is.
As can be seen from the above figure, TensorFlow tops the list, followed by Keras and Caffe respectively. The following abstract bacteria will share these resources with you.
16 Best Open Source Frameworks and Platforms for Deep Learning
TensorFlow
TensorFlow was originally developed by researchers and engineers on the Google Brain Team in Google's Machine Intelligence research organization. This framework is designed to facilitate researchers in machine learning research and to simplify the process of migrating from research models to real production.
Collection: 96655, Number of Contributors: 1432, Number of Program Submissions: 31714, Date of Establishment: November 1, 2015.
Link:
https://github.com/tensorflow/tensorflow
Hard
Keras is an API for high-level neural networks written in Python that can be used with TensorFlow, CNTK or Theano.
Collection: 28385, Number of Contributors: 653, Number of Program Submissions: 4468, Date of Establishment: March 22, 2015.
Link:
https://github.com/keras-team/keras
Caffe
Caffe is a deep learning framework focused on expressiveness, speed, and modularity, developed by the Berkeley Vision and Learning Center and community contributors.
Collection: 23750, Number of Contributors: 267, Number of Program Submissions: 4128, Date of Establishment: September 8, 2015.
Link:
https://github.com/BVLC/caffe
Microsoft Cognitive Toolkit
The Microsoft Cognitive Toolkit (formerly CNTK) is a unified deep learning toolset that describes a neural network as a series of computational steps represented by a directed graph.
Collection: 14243, Number of Contributors: 174, Number of Program Submissions: 15613, Date of Establishment: July 27, 2014.
Link:
https://github.com/Microsoft/CNTK
PyTorch
PyTorch is a framework for tensor computing and dynamic neural networks with powerful GPU support fused with Python.
Collection: 14101, Number of Contributors: 601, Number of Program Submissions: 10733, Date of Establishment: January 22, 2012.
Link:
https://github.com/pytorch/pytorch
Apache MXnet
Apache MXnet is a deep learning framework designed for efficiency and flexibility. It allows users to mix symbolic and imperative programming to maximize efficiency and productivity.
Collection: 13699, Number of Contributors: 516, Number of Program Submissions: 6953, Date of Establishment: April 26, 2015.
Link:
https://github.com/apache/incubator-mxnet
DeepLearning4J
DeepLearning4J, like ND4J, DataVec, Arbiter, and RL4J, is part of the Skymind Intelligence Layer. It is an open source distributed neural network library written in Java and Scala and certified for Apache 2.0.
Collection: 8725, Number of Contributors: 141, Number of Program Submissions: 9647, Date of Establishment: November 24, 2013.
Link:
https://github.com/deeplearning4j/deeplearning4j
Theano
Theano can efficiently handle user-defined, optimized, and computed mathematical expressions on multidimensional arrays. But in September 2017, Theano announced that there would be no further major progress after the release of version 1.0. Don't despair though, Theano is still a very powerful library for your deep learning research.
Collection: 8141, Number of Contributors: 329, Number of Program Submissions: 27974, Date of Establishment: January 6, 2008.
Link:
https://github.com/Theano/Theano
TFLearn
TFLearn is a modular and transparent deep learning library built on top of TensorFlow, designed to provide TensorFlow with a higher-level API to facilitate and speed up experimental research while maintaining complete transparency and compatibility.
Collection: 7933, Number of Contributors: 111, Number of Program Submissions: 589, Date of Establishment: March 27, 2016.
Link:
https://github.com/tflearn/tflearn
Torch
Torch is the main package in Torch7, which defines data structures and mathematical operations for multidimensional tensors. Additionally, it provides a lot of utility software for accessing files, serializing objects of arbitrary types, and more.
Collection: 7834, Number of Contributors: 133, Number of Program Submissions: 1335, Date of Establishment: January 22, 2012.
Link:
https://github.com/torch/torch7
Caffe2
Caffe2 is a lightweight deep learning framework with features such as modularity and extensibility. It improves upon the original Caffe, improving its expressiveness, speed and modularity.
Collection: 7813, Number of Contributors: 187, Number of Program Submissions: 3678, Date of Establishment: January 21, 2015.
Link:
https://github.com/caffe2/caffe2
PaddlePaddle
PaddlePaddle (Parallel Distributed Deep Learning) is an easy-to-use efficient, flexible, and scalable deep learning platform. It was originally developed by Baidu scientists and engineers to apply deep learning to many of Baidu's products.
Collection: 6726, Number of Contributors: 120, Number of Program Submissions: 13733, Date of Establishment: August 28, 2016.
Link:
https://github.com/PaddlePaddle/Paddle
DLib
DLib is a modern C++ toolkit containing machine learning algorithms and tools for developing complex software based on C++ to solve practical problems.
Collection: 4676, Number of Contributors: 107, Number of Program Submissions: 7276, Date of Establishment: April 27, 2008.
Link:
https://github.com/davisking/dlib
Chainer
Chainer is an independent open source framework for deep learning models based on python. It provides flexible, intuitive, and high-performance means to implement comprehensive deep learning models, including the latest emerging recurrent neural networks and variational automation. Encoders (variational auto-encoders).
Collection: 3685, Number of Contributors: 160, Number of Program Submissions: 13700, Date Established: April 12, 2015.
Link:
https://github.com/chainer/chainer
Neon
Neon is a Python-based deep learning library developed by Nervana. It's easy to use, while performance is at the highest level.
Collections: 3466, Number of Contributors: 77, Number of Program Submissions: 1112, Date of Establishment: May 3, 2015.
Link:
https://github.com/NervanaSystems/neon
Lasagne
Lasagne is a lightweight library that can be used to build and train neural networks on Theano.
Collection: 3417, Number of Contributors: 64, Number of Program Submissions: 1150, Date of Establishment: September 7, 2014.
Link:
https://github.com/Lasagne/Lasagne
Other options
-
H2O.ai
https://github.com/h2oai/h2o-3
-
PyLearn
https://github.com/lisa-lab/pylearn2
-
BigDL
https://github.com/intel-analytics/BigDL
-
Shogun
https://github.com/shogun-toolbox/shogun
-
Apache SINGA
https://github.com/apache/incubator-singa
-
Blocks
https://github.com/mila-udem/blocks
-
Mocha
https://github.com/pluskid/Mocha.jl