Open Source Neural Network Library

Question

I am looking for an open source neural network library. So far I have watched FANN, WEKA and OpenNN. The rest, what should I look at? The criteria, of course, are documentation, examples, and ease of use.

Answer 1

Last updated: 2017/12/17 (from time to time I will update this answer ...)

Standard implementations of neural networks

• FANN is a very popular implementation in C / C ++ and has bindings for many other languages.
• I think WEKA does not have a very good implementation for neural networks. There is a better library for Java (and C #): Encog .
• In scikit-learn (Python) 0.18 (current development version), implementation of feedback neural networks ( API documentation ) will appear.
• PyBrain (Python) contains various types of neural networks and training methods.
• And I have to mention my own project called OpenANN ( Documentation ). It is written in C ++ and has Python bindings.

Deep learning

As there is huge hype around neural networks (“deep learning”), there are many research libraries available that may not be easy to set up, integrate, and use. On the other hand, they provide advanced functionality and high performance (with GPUs, etc.).

• Keras is the best of this category, in my opinion: useful, powerful and actively developed. As an alternative to Theano, he can use TensorFlow , Theano and CNTK as a backend.
• TensorFlow from Google (C ++ / Python)
• Microsoft CNTK (Python Learning / C ++ / C # / Java / Python Evaluation)
• Caffe from Berkeley Vision and Learning Center in C ++ with Python bindings
• Caffe2 from Facebook in C ++ with Python bindings
• Facebook's PyTorch , in Python, can be expanded with C / C ++
• mxnet (C ++, Python, R, Scala, Julia, Matlab, Javascript)
• Theagne based Lasagne (Python)
• Theano Blocks (Python)
• Neon provides very efficient implementations (Python)
• Neural Networks for Torch 7 (Lua, Torch 7 is a "Matlab-like environment", an overview of machine learning algorithms in the torch )
• Deeplearning4j (Java)
• Chainer (Python)
• MatConvNet (Matlab)
• PaddlePaddle , CUDA / C ++ with Python bindings
• NNabla in Cuda / C ++ 11 with Python bindings

A performance comparison for libraries with GPU acceleration can be found here (unfortunately outdated). A comparison of GPUs and library versions can be found here .

Inactive:

• Theano (Python) and its high-level APIs:
  • Pylearn 2
  • Theanets
  • scikit-neuralnetwork
• cuda-convnet2 in CUDA / C ++ with Python bindings
• Hebel (Python)

Answer 2

If you need the flexibility to define network configurations, such as sharing parameters or creating different types of convolutional architectures, then you should look at the Torch library family: http://www.torch.ch/ .

I haven't looked at the documentation for Torch 7 yet, but the documentation for other versions was pretty decent, and the code is very readable (in Lua and C ++).

Answer 3

You can use accord.net environment. http://accord-framework.net/

It contains neural learning algorithms such as Levenberg-Marquardt, Parallel Resilient Backpropagation, Nguyen-Widrow initialization algorithm, Deep Belief Networks and limited Boltzmann machines and many other neural networks associated with it.

Answer 4

Netlab is a commonly used Matlab library. (free and open source)

The Netlab toolkit is designed to provide the necessary tools for modeling a theoretically based neural network of algorithms and related models for use in teaching, research, and application development. It is widely used in MSc through Research in the Mathematics of Complex Systems.

Netlab library includes software implementations of a wide range of data analysis methods, many of which are not yet available in standard neural network modeling packages. Netlab works with Matlab version 5.0 and higher, but only the main Matlab is needed (i.e. there is no other toolbar). It is not compatible with earlier versions of Matlab.