Top AI Hackaton 2017 projects at Belarus Hi-Tech Park

Top AI Hackaton 2017 projects at Belarus Hi-Tech Park

AI Hackathon 2017 was held last weekend in the business incubator of the High Technologies Park of Belarus. Startups of sometimes very young teams concerned with developments in the field of artificial intelligence - they used neural networks. The organizers wanted the developments to be of benefit to the community and of social significance. The prize fund of the forum was $10,000. Find out about top three projects of the 2017 Hackathon.

NeuroCrypto. Self-learning technology that will help encrypt data

Team NeuroCrypto in the opinion of the jury took first place and received the main prize - $4,000. Their project was at the crossroads of two areas: cryptography and machine learning.

Denis, a participant in NeuroCrypto, said: "Google has been using machine learning for quite some time to find successful neural network architectures. We tried to repeat the same thing in the field of cryptography. We developed a new approach, namely, we proposed a generalization of the known scheme of data encryption, part of which was replaced by a neural network and thereby we made the system self-learning."

"Our technology can be used to search for potentially crypto-resistant encryption algorithms in automatic mode, you do not need to be an expert in this field."

The idea of ​​the team is the development of its own unique algorithms that can be used to encrypt data. They want to sell it to those enterprises that do not have confidence in the existing schemes. Although within the Hackathon the team did not have time to find an algorithm that surpasses the existing analogs, they got several algorithms comparable with them in terms of cryptographic stability. The jury decided that the project is promising and deserves first place.

Kimchi. An application that will make you Batman (at least virtually)

Everyone remembers MSQRD and their animated masks. Team Kimchi (the Faculty of Applied Mathematics and Informatics of the BSU, as well as specialists of the Institute of Informatics Problems of the National Academy of Sciences) decided to go even further, and began work on three-dimensional costumes. With this idea, the team took second place, taking $3,000.

The project is inspired by Apple's technology Animated Emojis. Kimchi's idea is to create a system that would determine the skeleton of a person, their movements, and project it into a three-dimensional model of the prototype. Such technology can be used in a variety of spheres, for example, in the identification of a person by behavior. Members of the team admitted they always wanted to become superheroes, so the team began to develop the app based on entertainment: you can wear costumes, for example, the one of Iron Man.

Prior to Hackathon, these guys did not have a deep knowledge of neural networks, but they managed to master them in 48 hours. During this time, the team was able to teach the neural network to recognize the skeleton and even the person's posture.

LaineAssist. Neural network to assess driving style

Team LaineAssist took third place and received $2,000. The developers wanted to create a system for monitoring traffic and the quality of driving. At the 2016 Hackathon the jury really liked the project Olenemetr (Deer Meter, since "Olen", or "Deer" is a swear word in Russian, which means "careless person or driver"). The application read the car's license plate and rated each driver, thus some of them became "deer on the road". But that application worked on the basis of people's assessments of each other. Now machines should do everything themselves, without man.

"We had two tasks: the first is connected with the analysis of data, the second with computer vision," the participants said. "To solve the first, we collected data from mobile sensors that were in the car during its movement. On the basis of these data, the neural network was taught to determine when the car was traveling smoothly and when some sharp turns or sudden braking occurred. These data could be useful for assessing the driving style, for example, at companies such as Yandex and Uber."

The team taught the neural network to recognize the markings on the road and the distance to the car ahead. Although the jury noted that the program is somewhat "shortsighted," which can be dangerous, for example, at steep corners. But now the team has time and, most importantly, means to refine it. Perhaps soon people with a bad vestibular apparatus will not be scared to use taxis :)

MORE: OneSoil, Belarusian application which will feed 1 billion people


Belarusian programmers to develop alternate reality app

Belarusian programmers develop food ingredients’ scanning app