PuzzleLib Neural network library

PuzzleLib is the software which allows to bring together neural network under a specific objective, to visualize its architecture, to start training of this neural network and to unload the trained network for further use.

2019: Entering into the Unified register of domestic software

On January 25, 2019 in Skolkovo Foundation reported that development of Ashmanov's Neuronets company, the resident Klaster of information technologies of Skolkovo Foundation, is entered in the Unified register of the Russian programs for electronic computers and databases.

"I consider emergence of domestic technology of work with the neural networks optimized as well under the Russian stack of the computing equipment, very important milestone of development of IT technologies in the country. This fact expands a circle of customers of this software and gives the chance to create neural network technologies of dedicated application". Sergey Voinov, head "IoT, Electronics, Digital Health" of an IT cluster of Skolkovo Foundation

As noted in Skolkovo, to the next analogs which are applied in the Russian companies, it is possible to carry or foreign libraries in open access (Google TensorFlow Facebook PyTorch etc.), or hardware and software systems of large foreign corporations (IBM Power AI SAP HANA etc.).

Features of PuzzleLib mentioned by the developer:

• high speed – only low-level utilities for video cards are used NVIDIA and AMD and resources are effectively involved processor;
• work for CPU and GPU (central processors and video cards) is supported;
• the library can work for all main operating systems (family Linux Windows Mac OS, iOS i Android);
• automatic optimization of calculations;
• more than 60 types of neural network modules which allow to build ultraprecise, recurrent neuronets and networks with any computation graphs;
• the input threshold in PuzzleLib for the user is lower, than for other libraries;
• completely domestic development.

"Data processing rate is very important characteristic. For example, there are cars with the built-in systems of detection of an obstacle which are developed on neural networks. The quicker this system will define danger, the probability to avoid collision is higher". Anton Drobyshev, specialist in machine learning of Ashmanov's Neuronets company

Also the Ashmanov's Neuronets company is going to provide support of the domestic processors Elbrus, Baikal and ELVIS ELISE which are used in production of the Russian computers, devices of industrial automation, calculators, multimedia devices, etc.