下表比较了用于深度学习的著名软件框架、库和计算机程序。
Deep-learning software by name
|
Software |
Initial Release |
Software license[a] |
Open source |
Platform |
Written in |
Interface |
support |
support |
CUDA support |
Has pretrained models |
Parallel execution (multi node) |
Actively Developed |
||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| BigDL | 2016 | Apache 2.0 | Yes |
Apache Spark |
Scala |
Scala, Python |
No | Yes | Yes | Yes | ||||||
| Caffe | 2013 | BSD | Yes | C++ | Yes |
Under development[3] |
Yes | Yes | Yes[4] | Yes | Yes | No | ? | |||
| Chainer | 2015 | BSD | Yes | Python | Python | No | No | Yes | Yes | Yes | Yes | Yes | No | Yes | Yes | |
| Deeplearning4j | 2014 | Apache 2.0 | Yes |
C++, |
Java, (Keras), |
Yes | No[5] | Yes[6][7] | Computational Graph | Yes[8] | Yes | Yes | Yes | Yes[9] | ||
| Dlib | 2002 | Boost Software License | Yes | C++ | C++ | Yes | No | Yes | Yes | Yes | No | Yes | Yes | Yes | ||
| Intel Data Analytics Acceleration Library | 2015 | Apache License 2.0 | Yes |
on Intel |
C++, |
C++, |
Yes | No | No | Yes | No | Yes | Yes | |||
| Intel Math Kernel Library | Proprietary | No |
on Intel |
C[12] | Yes[13] | No | No | Yes | No | Yes[14] | Yes[14] | No | ||||
| Keras | 2015 | MIT license | Yes | Python | Only if using Theano as backend |
Can use Theano, Tensorflow or PlaidML as backends |
Yes | Yes | Yes[15] | Yes | Yes | No[16] | Yes[17] | Yes | ||
| MATLAB + Deep Learning Toolbox | Proprietary | No |
C, C++, Java, |
MATLAB | No | No | Train with Parallel Computing Toolbox and generate CUDA code with GPU Coder[18] | No | Yes[19][20] | Yes[19] | Yes[19] | No | With Parallel Computing Toolbox[21] | Yes | ||
| Microsoft Cognitive Toolkit (CNTK) | 2016 | MIT license[22] | Yes |
via Docker on roadmap) |
C++ |
C++, BrainScript[25] (.NET on roadmap[26]) |
Yes[27] | No | Yes | Yes | Yes[28] | Yes[29] | Yes[29] | No[30] | Yes[31] | No[32] |
| Apache MXNet | 2015 | Apache 2.0 | Yes |
AWS, iOS, |
Small core library |
C++, Go, R, |
Yes |
On roadmap[37] |
Yes | Yes[38] | Yes[39] | Yes | Yes | Yes | Yes[40] | Yes |
| Neural Designer | Proprietary | No | C++ | Yes | No | No | ? | ? | No | No | No | ? | ||||
| OpenNN | 2003 | GNU LGPL | Yes | C++ | C++ | Yes | No | Yes | ? | ? | No | No | No | ? | ||
| PlaidML | 2017 | AGPL | Yes |
C++, |
? |
Some OpenCL ICDs are not recognized |
No | Yes | Yes | Yes | Yes | Yes | Yes | |||
| PyTorch | 2016 | BSD | Yes |
C, C++, |
Yes |
Via separately maintained package |
Yes | Yes | Yes | Yes | Yes | Yes | Yes | |||
| Apache SINGA | 2015 | Apache 2.0 | Yes | C++ |
C++, |
No |
Supported in V1.0 |
Yes | ? | Yes | Yes | Yes | Yes | Yes | ||
| TensorFlow | 2015 | Apache 2.0 | Yes |
C++, |
(Keras), |
No |
On roadmap [45] but already with support |
Yes | Yes[47] | Yes[48] | Yes | Yes | Yes | Yes | Yes | |
| Theano | 2007 | BSD | Yes | Python |
(Keras) |
Yes |
Under develo pment[49] |
Yes | Yes[50][51] | Through Lasagne's model zoo[52] | Yes | Yes | Yes | Yes[53] | No | |
| Torch | 2002 | BSD | Yes | C, Lua |
Lua, utility library for C++/ |
Yes |
Third party implemen |
Yes[60][61] | Through Twitter's Autograd[62] | Yes[63] | Yes | Yes | Yes | Yes[54] | No | |
| Wolfram Mathematica | 1988 | Proprietary | No |
C++, |
Wolfram Language | Yes | No | Yes | Yes | Yes[64] | Yes | Yes | Yes | Yes[65] | Yes |
- ^ Licenses here are a summary, and are not taken to be complete statements of the licenses. Some libraries may use other libraries internally under different licenses
Related software
- Neural Engineering Object (NENGO) – A graphical and scripting software for simulating large-scale neural systems
- Numenta Platform for Intelligent Computing – Numenta's open source implementation of their hierarchical temporal memory model
See also
- Comparison of numerical-analysis software
- Comparison of statistical packages
- List of datasets for machine-learning research
- List of numerical-analysis software
本文：https://pub.intelligentx.net/wikipedia-comparison-deep-learning-software
讨论：请加入知识星球或者小红圈【首席架构师圈】