The following table compares notable software frameworks, libraries and computer programs for deep learning.
Software | Creator | Initial release | Software license[a] | Open source | Platform | Written in | Interface | OpenMP support | OpenCL support | CUDA support | ROCm support[1] | Automatic differentiation[2] | Has pretrained models | Recurrent nets | Convolutional nets | RBM/DBNs | Parallel execution (multi node) | Actively developed |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
BigDL | Jason Dai (Intel) | 2016 | Apache 2.0 | Yes | Apache Spark | Scala | Scala, Python | No | No | Yes | Yes | Yes | ||||||
Caffe | Berkeley Vision and Learning Center | 2013 | BSD | Yes | Linux, macOS, Windows[3] | C++ | Python, MATLAB, C++ | Yes | Under development[4] | Yes | No | Yes | Yes[5] | Yes | Yes | No | ? | No[6] |
Chainer | Preferred Networks | 2015 | BSD | Yes | Linux, macOS | Python | Python | No | No | Yes | No | Yes | Yes | Yes | Yes | No | Yes | No[7] |
Deeplearning4j | Skymind engineering team; Deeplearning4j community; originally Adam Gibson | 2014 | Apache 2.0 | Yes | Linux, macOS, Windows, Android (Cross-platform) | C++, Java | Java, Scala, Clojure, Python (Keras), Kotlin | Yes | No[8] | Yes[9][10] | No | Computational Graph | Yes[11] | Yes | Yes | Yes | Yes[12] | Yes |
Dlib | Davis King | 2002 | Boost Software License | Yes | Cross-platform | C++ | C++, Python | Yes | No | Yes | No | Yes | Yes | No | Yes | Yes | Yes | |
Flux | Mike Innes | 2017 | MIT license | Yes | Linux, MacOS, Windows (Cross-platform) | Julia | Julia | Yes | No | Yes | Yes[13] | Yes | Yes | No | Yes | Yes | ||
Intel Data Analytics Acceleration Library | Intel | 2015 | Apache License 2.0 | Yes | Linux, macOS, Windows on Intel CPU[14] | C++, Python, Java | C++, Python, Java[14] | Yes | No | No | No | Yes | No | Yes | Yes | |||
Intel Math Kernel Library 2017 [15] and later | Intel | 2017 | Proprietary | No | Linux, macOS, Windows on Intel CPU[16] | C[17] | Yes[18] | No | No | No | Yes | No | Yes[19] | Yes[19] | No | |||
Google JAX | 2018 | Apache License 2.0 | Yes | Linux, macOS, Windows | Python | Python | Only on Linux | No | Yes | No | Yes | Yes | ||||||
Keras | François Chollet | 2015 | MIT license | Yes | Linux, macOS, Windows | Python | Python, R | Only if using Theano as backend | Can use Theano, Tensorflow or PlaidML as backends | Yes | No | Yes | Yes[20] | Yes | Yes | No[21] | Yes[22] | Yes |
MATLAB + Deep Learning Toolbox (formally Neural Network Toolbox) | MathWorks | 1992 | Proprietary | No | Linux, macOS, Windows | C, C++, Java, MATLAB | MATLAB | No | No | Train with Parallel Computing Toolbox and generate CUDA code with GPU Coder[23] | No | Yes[24] | Yes[25][26] | Yes[25] | Yes[25] | Yes | With Parallel Computing Toolbox[27] | Yes |
Microsoft Cognitive Toolkit (CNTK) | Microsoft Research | 2016 | MIT license[28] | Yes | Windows, Linux[29] (macOS via Docker on roadmap) | C++ | Python (Keras), C++, Command line,[30] BrainScript[31] (.NET on roadmap[32]) | Yes[33] | No | Yes | No | Yes | Yes[34] | Yes[35] | Yes[35] | No[36] | Yes[37] | No[38] |
ML.NET | Microsoft | Yes | Windows, Linux, macOS | C#, F# | Yes | |||||||||||||
Apache MXNet | Apache Software Foundation | 2015 | Apache 2.0 | Yes | Linux, macOS, Windows,[39][40] AWS, Android,[41] iOS, JavaScript[42] | Small C++ core library | C++, Python, Julia, Matlab, JavaScript, Go, R, Scala, Perl, Clojure | Yes | No | Yes | No | Yes[43] | Yes[44] | Yes | Yes | Yes | Yes[45] | No |
Neural Designer | Artelnics | 2014 | Proprietary | No | Linux, macOS, Windows | C++ | Graphical user interface | Yes | No | Yes | No | Analytical differentiation | No | No | No | No | Yes | Yes |
OpenNN | Artelnics | 2003 | GNU LGPL | Yes | Cross-platform | C++ | C++ | Yes | No | Yes | No | ? | ? | No | No | No | ? | |
PlaidML | Vertex.AI, Intel | 2017 | Apache 2.0 | Yes | Linux, macOS, Windows | Python, C++, OpenCL | Python, C++ | ? | Some OpenCL ICDs are not recognized | No | No | Yes | Yes | Yes | Yes | Yes | Yes | |
PyTorch | Adam Paszke, Sam Gross, Soumith Chintala, Gregory Chanan (Facebook) | 2016 | BSD | Yes | Linux, macOS, Windows, Android[46] | Python, C, C++, CUDA | Python, C++, Julia, R[47] | Yes | Via separately maintained package[48][49][50] | Yes | Yes | Yes | Yes | Yes | Yes | Yes[51] | Yes | Yes |
Apache SINGA | Apache Software Foundation | 2015 | Apache 2.0 | Yes | Linux, macOS, Windows | C++ | Python, C++, Java | No | Supported in V1.0 | Yes | No | ? | Yes | Yes | Yes | Yes | Yes | |
TensorFlow | Google Brain | 2015 | Apache 2.0 | Yes | Linux, macOS, Windows,[52][53] Android | C++, Python, CUDA | Python (Keras), C/C++, Java, Go, JavaScript, R,[54] Julia, Swift | No | On roadmap[55] but already with SYCL[56] support | Yes | Yes | Yes[57] | Yes[58] | Yes | Yes | Yes | Yes | Yes |
Theano | Université de Montréal | 2007 | BSD | Yes | Cross-platform | Python | Python (Keras) | Yes | Under development[59] | Yes | No | Yes[60][61] | Through Lasagne's model zoo[62] | Yes | Yes | Yes | Yes[63] | No |
Torch | Ronan Collobert, Koray Kavukcuoglu, Clement Farabet | 2002 | BSD | Yes | Linux, macOS, Windows,[64] Android,[65] iOS | C, Lua | Lua, LuaJIT,[66] C, utility library for C++/OpenCL[67] | Yes | Third party implementations[68][69] | Yes[70][71] | No | Through Twitter's Autograd[72] | Yes[73] | Yes | Yes | Yes | Yes[64] | No |
Wolfram Mathematica 10[74] and later | Wolfram Research | 2014 | Proprietary | No | Windows, macOS, Linux, Cloud computing | C++, Wolfram Language, CUDA | Wolfram Language | Yes | No | Yes | No | Yes | Yes[75] | Yes | Yes | Yes | Yes[76] | Yes |
Software | Creator | Initial release | Software license[a] | Open source | Platform | Written in | Interface | OpenMP support | OpenCL support | CUDA support | ROCm support[77] | Automatic differentiation[2] | Has pretrained models | Recurrent nets | Convolutional nets | RBM/DBNs | Parallel execution (multi node) | Actively developed |
[further explanation needed]
Format name | Design goal | Compatible with other formats | Self-contained DNN Model | Pre-processing and Post-processing | Run-time configuration for tuning & calibration | DNN model interconnect | Common platform |
---|---|---|---|---|---|---|---|
TensorFlow, Keras, Caffe, Torch, | Algorithm training | No | No / Separate files in most formats | No | No | No | Yes |
ONNX | Algorithm training | Yes | No / Separate files in most formats | No | No | No | Yes |
{{cite web}}
: CS1 maint: numeric names: authors list (link)