Abbyy NeoML

Main article: Machine Learning

2021: NeoML 2.0 with Python support

On June 22, 2021, ABBYY announced the update of the NeoML framework. Now, users of Python, one of the most popular programming languages ​ ​ for data analysis and machine learning, including for research tasks, can work with the ABBYY open cross-platform library. With NeoML, representatives of business and the scientific community will be able to supplement applications with the capabilities of object identification, classification, regression, clustering, semantic segmentation, verification, using the most relevant platforms and architectures for these tasks. The library supports about 20 machine learning methods, including 10 additional network layers and additional optimization methods.

In this version of NeoML the speed of performance classical algorithms on different tasks grew to 10 times, and training neuronets became 30% faster. This optimization is especially useful for specialists and companies that train ML models in, and cloud services will also simplify development for mobile applications customers. In addition, this version supports automatic calculation of gradients, an important function for quick implementation of neural networks various architectures. NeoML 2.0 also supports work in the latest environments: on M1 processors Apple and graphics processors in the environment, Linux including on integrated models from, which Intel significantly expands the capabilities of application development for customers.

Open code is the main source of innovation in the development of modern software. That is why we are committed to making the NeoML library accessible to even more users. Python is a universal programming language, convenient for integration with various information systems. Not only ABBYY employees, but also specialists from different countries actively participated in the creation of the interface for Python, which indicates the interest of the OSS community in our library and its capabilities. So, Alexander Borgardt, head of the Open source community, participated in the development of the wrapper for Python, DuckStax comments Vladimir Yunev, chief architect of ABBYY.

In my opinion, there are many problems in the solution of which the ABBYY library NeoML provides more opportunities along with other popular frameworks. In the process of assisting in the development of the connector to Python, I discovered a number of possibilities for starting the interference in different modes and with different restriction systems, and the performance of the interference was very pleased in many modes. And thanks to the key features of the architecture, the library can be used in environments where there is a need to assemble under desktops, mobiles, servers, or even in wasm. NeoML allows you to run the interference for all Apple devices. Thanks to the high-quality integration with Python, not only sired C++ developers can see the potential of the NeoML library, but also cool DL engineers can feel its power, "comments Alexander Borgardt, community leader Duckstax.com.

NeoML developers and researchers from the USA, Canada, Germany, the Netherlands, Russia, Brazil, China, India, Vietnam, South Korea and other countries are already using in their projects. Library tools are also used in all ABBYY products, including mobile applications. For example, thanks to the NeoML in the ABBYY mobile scanner, it became possible to recognize 7 different types of documents, search for text on images and measure objects using the AR ruler.

You can access the source code from the official project repository on the GitHub. NeoML can be used on Windows, Linux, macOS, iOS and Android. The library supports CPU and GPU processors. Open-source framework code is provided under the Apache2.0License. ABBYY will continue to increase the number of available algorithms and architectures, as well as increase the speed of the library.

2020: Open Source Machine Learning Library Announcement

On June 16, 2020, ABBYY introduced NeoML, an open-source machine learning library available on the GitHub platform. The library supports modern methods of both deep and classical machine learning to solve various problems - for example, remote identification, predictive analytics, risk management and others. The main possibility of NeoML is the ability to create business applications that work equally efficiently in the cloud, on desktops and mobile devices.

According to the company, with the help of NeoML, companies will be able to supplement applications with the capabilities of object identification, classification, semantic segmentation, verification, forecasting, and so on. For example, banks can develop, train and apply models to assess credit risks and predict customer outflows, telecom operators - analyze the success of mass marketing campaigns, retail and FMCG - develop services for registering customers, for example, in loyalty programs.

Companies with high levels of digital intelligence are increasingly relying on open source software. Therefore, we decided to give the scientific and business community access to ABBYY development in the field of machine learning. This makes it possible to accelerate the development of products through feedback from developers, as well as expand the scope of application of company technologies in projects and industries. commented by Dmitry Shushkin, ABBYY Russia CEO

The library for June, 2020 supports languages programmirovaniyas ++, Java, Objective C, and in the nearest future ABBYY will add to them Python. With support for the open ONNX standard, the library can use models from other frameworks, allowing developers to use the optimal combination of tools. The standard is created and supported jointly by Microsoft, Facebook and other partners as an open source project. NeoML is being developed in Microsoft Visual Studio.

One of the main priorities for Microsoft is the development of an open source ecosystem, especially in the field of artificial intelligence. We are pleased to support our ABBYY partner at this important stage. NeoML will open up a number of possibilities for using neural networks in business scenarios. Supporting the ONNX standard and placing the project on the GitHub will make the library globally accessible and create a vast ecosystem of developers around it. said Konstantin Goldstein, leading expert on Microsoft strategic technologies in Russia

The source code can be accessed from the official project repository. Business representatives, developers and data scientists around the world can offer their ideas for improving the code. NeoML can be used on Windows, Linux, macOS, iOS and Android. The library supports CPU and GPU. The open source framework for June 2020 is provided under the Apache 2.0 License. in the future ABBYY plans to increase the number of available algorithms and architectures, as well as further improve the speed of the library.