Top 5 domestic developments presented at the forum "Artificial Intelligence in Belarus"

Scientists and experts around the world argue whether artificial intelligence is good or evil. The truth is probably, as always, in the middle, because neural networks, when used rationally, can help a person and expand his capabilities. For example, personalize treatment, store information for millennia, recognize bird voices, or estimate housing costs. Our correspondents got acquainted with some developments presented at the V International IT-Academgrad Forum “Artificial Intelligence in Belarus” held in Minsk.

1. The system of personalized therapy of respiratory failure helps to select the patient the necessary amount of oxygen and adjust its supply depending on the condition of the person. The development is designed to make treatment more accurate, reduce the burden on medical personnel and reduce oxygen consumption.

The system includes three modules: diagnostic, control and regulation unit, and oxygen concentration module. They can be used both together and separately, depending on the tasks of the medical institution.

The compact diagnostic module allows measuring blood oxygen saturation and pulse with a wearable device, and respiratory rate with a special nasal cannula. The system has already received the necessary documents for use in healthcare institutions and is being launched into production.

Associate Professor BSUIR Oleg Zelmansky. The technology is planned to be used for early detection of diseases of the respiratory system, treatment of respiratory failure and rehabilitation. Another promising area is home therapy. The automated system will be able to select the necessary mode taking into account current indicators.

The results of clinical testing at the Republican Scientific and Practical Center of Pulmonology and Phthisiatrics showed that oxygen consumption was reduced to 60 percent. This is one of the most notable test results, although actual savings may depend on the patient's condition and the system's application conditions. 2. “5D-archive of Belarus”: memory for millennia The project’s goal is to ensure long-term preservation of information, designed for a longer perspective than the life of conventional electronic devices.

The technology involves recording information using a laser on a special glass. The estimated shelf life may exceed 10,000 years. Such a carrier does not need electricity to store recorded data and is resistant to a number of external influences, including temperature and mechanical.

One of the main goals of the project is to create a reliable infrastructure for the preservation of Belarusian cultural heritage. It is possible to record significant information on glass, which is important to pass on to the next generations. This is especially true for historical documents and other materials, the loss of which is irreparable.

The development involves practical application in museums: models are created that can become exhibition elements or souvenir products. In addition, experts are exploring the possibilities of using the technology for secure storage of digital data on glass and metals. The project uses laser systems and equipment of domestic development, glass mainly produced in Belarus. 3. How much is the apartment? Artificial intelligence helps to analyze the real estate market. The National Cadastre Agency of Belarus has developed a free online service that calculates the approximate market value of apartments. The user needs to specify only the address of the house, the number of rooms, area and type of finish. Within seconds, the service will generate a result and issue a report with additional information about the object and its surroundings.

The model is trained on information about more than 175 thousand real estate transactions registered since 2018. The algorithm takes into account more than 15 factors: location, floor, wall material, year of commissioning of the building, availability of public transport and subway, features of the area and other characteristics. The report may indicate, for example, the distance to the nearest park and public transport stops.

The absolute error of the model is up to 5.59 percent for Minsk and up to 6.53 percent for regional centers. The service is useful for those who are going to buy or sell an apartment or want to navigate the market prices. However, the calculation should be taken as a guideline: the features of a particular object and the terms of the transaction can affect the final cost.

Now the instrument covers Minsk and regional centers. In the future, it is planned to extend to district centers. There is a limit - no more than ten requests per day per user. 4. Externally, two signatures can be similar, but a person rarely reproduces the same stroke exactly the same. The speed of writing, the trajectory of the pen and the temporal characteristics may differ. It is these features that the software for checking dynamic signatures developed at the Joint Institute of Informatics Problems of the NAS of Belarus uses.

In the normal inspection, the signature image is compared. Dynamic analysis allows you to study the process of its implementation on the tablet: the program compares the coordinates of the stylus and the time spent on movement. Even if the fake looks like the original, differences in the sequence of movements can help identify the discrepancy.

To form an individual classifier, according to the developers, about 7-15 authentic signatures are enough.

The system does not use data on the strength of the stylus. The reason is practical: not every tablet supports this feature. Therefore, the program focuses on coordinates and time parameters, which allows you to use it on a wider range of devices.

Processing includes purification of data from interference, normalization, creation of a biometric sample and comparison with the verifiable signature. This approach can improve the reliability of electronic document verification and become an additional tool for biometric identification. At the same time, the technology does not exclude counterfeiting absolutely, but helps to detect differences through the analysis of dynamic features. 5. While walking through the forest, we hear many bird voices, but we can not always determine who they belong to. For ornithologists, sounds are an important source of information about species diversity and the state of ecosystems. The development of the Joint Institute of Informatics Problems of the NAS of Belarus helps to automate the analysis of such data.

The system is created by the Intelligent Services Department. The neural network was trained on about 2,300 audio recordings from open sources, posted with the participation of ornithologists. The training base includes the voices of 116 species of birds living in our country.

Audio for analysis is divided into two-second segments, and then converted into chalk spectrograms - a visual representation of the frequency characteristics of sound. The model learns to recognize the characteristic acoustic features and determine the likely species of bird.

The user will be able to record singing on a smartphone while walking and get the intended name of the bird with a brief description. The developers plan to expand the sound library and add new views.

However, the main value of the project is not limited to the ability of nature lovers to recognize birds by voice. Automatic recognition will allow analyzing large volumes of audio recordings and organizing long-term monitoring of selected forest areas. This will help scientists assess species diversity, observe rare and endangered species, and plan conservation efforts.