Highly optimised open source machine learning library research & development
DeepTrainer is an open source project to implement heavily optimised deep learning methods for artificial neural networks.
DeepTrainer currently implements the following algorithms:
- Batch Backpropagation
- Distributed Batch Backpropagation
- Resilient Propagation
- Distributed Resilient Propagation
With the following activation functions selectable for each layer:
- Hyperbolic tangent
- Arcus tangent
- PReLU (lazy ReLU)
All algorithms are making use of a matrix implementation that uses automatic partitioning (4x4 or 8x8, single or double precision), and hardware accelerated matrix operations (dot product using SSE2, AVX, AVX512).
The algorithms are implemented using C++17 in Visual Studio on Windows OS, although a portable version is planned too. More details here: https://bulyaki.com/2018/04/02/c11-dll-library-for-the-matrix-rprop-algorithm/ https://bulyaki.com/2018/04/01/old-demo-for-the-rprop-algorithm-using-matrices/ https://bulyaki.com/2013/04/14/the-matrix-form-of-the-rprop-algorithm/ https://bulyaki.com/2013/04/14/the-matrix-form-of-the-backpropagation-algorithm/
Build notes: For CPU-only build select the "Intel x86" configuration. For Nvidia CUDA acceleration select the CUDA x64 configuration.
License information for third-party packages used in this solution:
Extended.Wpf.Toolkit https://github.com/xceedsoftware/wpftoolkit/blob/master/license.md OxyPlot.Core https://raw.githubusercontent.com/oxyplot/oxyplot/master/LICENSE