Python 1。 Theano is a Python class library that uses array vectors to define and calculate mathematical expressions. It makes it easy to write deep learning algorithms in Python environment. On the basis of it, many class libraries have been established.

1.Keras is a concise and highly modular neural network library. Its design is based on Torch, written in Python language, and supports the operation of calling GPU and CPU optimization.

2.Pylearn2 is a library, which integrates a large number of common deep learning models and training algorithms, such as random gradient descent. Its function library is based on Theano.

3.Lasagne is a lightweight package library, which is used to build and train neural networks, based on Theano. It follows the principles of simplicity, transparency, modularity, practicality and specialization.

4.Blocks is also a framework based on Theano, which is used to help build neural networks.

2.Caffe is a framework for deep learning, focusing on the expression form, running speed and modularity of the code. It was jointly developed by the Berkeley Visual and Learning Center (BVLC) and community members. Google's DeepDream project is based on Caffe framework. This framework is a C++ library using BSD license and provides a Python calling interface.

3.nolearn includes a large number of encapsulation and abstract interfaces of existing neural network function libraries, famous lasagna and some commonly used modules of machine learning.

4.Genism is also a deep learning gadget written in Python, which uses efficient algorithms to process large-scale text data.

5.Chainer has built a bridge between the theoretical algorithm and practical application of deep learning. Its characteristics are powerful, flexible and intuitive, and it is considered as a flexible framework for deep learning.

6.deepnet is a deep learning algorithm library based on GPU. Developed in Python language, algorithms such as feedforward neural network (FNN), restricted Boltzmann machine (RBM), depth belief network (DBN), self-encoder (AE), depth Boltzmann machine (DBM) and convolutional neural network (CNN) are realized.

7.Hebel is also a Python library for deep learning and neural networks. It controls GPU acceleration supporting CUDA through pyCUDA. It realizes the most important neural network model, and provides a variety of activation functions and model training methods, such as momentum, Nesterov momentum, dropping out of school and stopping early.

8.CXXNET is a fast and concise distributed deep learning framework based on MShadow. It is a lightweight and extensible C++/CUDA neural network toolbox, which provides a friendly Python/Matlab interface for training and prediction.

9.DeepPy is a deep learning framework based on NumPy.

10.DeepLearning is a deep learning function library developed by C++ and Python***.

1 1.Neon is a deep learning framework of Nervana system, which is developed in Python.

matlab

Convnet convolutional neural network is a deep learning classification algorithm, which can learn useful features from original data by adjusting weights.

2.DeepLearnToolBox is a Matlab/Octave toolbox for deep learning, including deep belief network (DBN), stacked AE and convolutional neural network (CNN).

3.cuda-convet is a set of convolutional neural network (CNN) codes, which is also suitable for feedforward neural networks, and uses C++/CUDA for operation. It can simulate multi-layer neural network with arbitrary depth. As long as it is the network structure of directed acyclic graph. Back propagation algorithm (BP algorithm) is used in the training process.

4.MatConvNet is a convolution neural network (CNN)Matlab toolbox for computer vision applications. It is simple and efficient, and can run and learn the most advanced machine learning algorithms.

Card Print Processor (abbreviation for card print processor)

1.eblearn is an open source C++ packaging library for machine learning, which was developed by the machine learning laboratory of new york University led by Yann LeCun. It uses energy-based model to realize convolutional neural network, and provides visual interactive interface (GUI), examples and demonstration courses.

2.SINGA is a project supported by Apache Software Foundation, and its design goal is to provide a general distributed model training algorithm on the existing system.

3.NVIDIA DIGITS is a new system for developing, training and visualizing deep neural networks. It presents the powerful function of deep learning with browser interface, which allows data scientists and researchers to visualize the behavior of neural networks in real time and quickly design deep neural networks that are most suitable for data.

4.intel? The deep learning framework provides Intel? Unified platform for accelerating deep convolution neural network.

Java language (a computer language, especially for creating websites)

1.n-dimensional array Java (ND4J) is the scientific computing function library of JVM platform. Mainly used in products, that is to say, the design requirements of functions are fast operation speed and minimum storage space.

2.Deeplearning4j is the first commercial open source distributed deep learning class library, written in Java and Scala. It is designed for business environment, not as a research tool.

3.Encog is an advanced framework of machine learning, covering support vector machines, artificial neural networks, genetic programming, Bayesian networks, hidden Markov models, etc., and also supporting genetic algorithms.

Java Script language

1.Convnet.js is written by JavaScript, which is a package library for training deep learning models (mainly neural networks) completely in the browser. No other software, no compiler, no installation package, no GPU, or even a breeze.

Left upper arm

1.Torch is a scientific computing framework, which is widely used in various machine learning algorithms. Very easy to use, developed with LuaJit, a fast scripting language, and implemented with C/CUDA at the bottom. Torch is based on Lua programming language.

Julia

1.Mocha is Julia's deep learning framework, inspired by the C++ framework Caffe. The efficient implementation of general stochastic gradient solver and general module in Mocha can be used to train deep/shallow (convolutional) neural networks, and can be completed by unsupervised pre-training (optional) (stacked) self-encoder. Its advantages include modular structure, providing upper interface, and perhaps speed, compatibility and so on.

Bite the tongue

1.lush (lisp universal shell) is an object-oriented programming language for researchers, experimenters and engineers interested in large-scale numerical and graphic applications. It has a function library of machine learning, including a rich deep learning library.

Haskell

1.DNNGraph is a domain-specific language (DSL) used by Haskell to generate deep neural network models.

. net

1.Accord.NET is a. NET machine learning framework written entirely in C#, including class libraries for audio and image processing. It is a complete product-level framework for computer vision, computer audio, signal processing and statistical applications.

rare

1.darch package can be used to generate multilayer neural network (deep structure). Training methods include pre-training of contrast divergence and fine-tuning of well-known training algorithms (such as back propagation method or yoke gradient method).

2.deepnet has implemented many deep learning frameworks and neural network algorithms, including back propagation (BP), restricted Boltzmann machine (RBM), deep belief network (DBP) and deep automatic encoder.