From Entertainment to Medicine: Programmers Presented their Projects at AI Hackathon
The AI Hackathon took place in the Higher School of Economics last weekend. Fifty seven teams presented ML, DL and AI technologies for solving medical, financial, social and many other problems. During the hackathon’s finals, 38 teams presented their projects in a three-minute pitch format.
The projects presented at the hackathon were devoted to socially important issues and those of entertainment. However, some projects couldn’t be referred to only one group; for instance, both a model who constantly keeps diet and a person with diabetes can use an application with a food recognition technology.
Lots of projects were devoted to medicine. The participants developed apps that support patients with diabetes and respiratory tract diseases, improve the quality of MRI images, detect tumors on X-ray images and so forth.
The authors of the "MedBot" project worked on the regulate self-treatment problem. According to them, many people in Russia prefer to look for appropriate treatment using the Internet rather than visiting a doctor. It has various reasons, but the main ones are that people don’t trust physicians and lack time.
The chat bot is based on Telegram. The developers used the Health.Mail.Ru service — its "question-answer" section has over 4000 presumptive diagnoses based on online applications. The developers taught the chat bot with this data.
One other medical service presented at the hackathon was "Onco Unite." The system systematizes gene-mutations. There are certain mutations that are associated with different diseases — some of them are even included in the so-called "mutation bank". The problem is that everybody, even healthy ones have the mutations presented in the database. In some cases one can’t find them in a sick person’s genome.
"As of now, we realize the following problem: there are genetic disorders like, for instance, epilepsy — in 30% of its cases, the causes are genetic. Physicians have to look for certain mutations in the database, which is often useless. There is no any service that would search for even normal mutations localized on particular DNA region," says a representative of the project.
OncoUnite's developers came to the point that a disease can be caused by a set of mutations, as opposed to one mutation presented in the database. A patient with "normal" mutations may fall in a high-risk group, and vice-versa. During the hackathon, the programmers analyzed the gene-mutation database, chose particular sets that consist of pathologic mutations, but are not related to any disease. In the future, they are planning to work on disease-related sets.
The authors of entertainment projects tried to satisfy the needs of so-called "digital-generation." Their apps can predict the success of movies, delete a person from a group photo, generate background music and so forth. The "Kickcity" project makes it possible to look for events and also find soulmates who have the same preferences as you. Authors of Shavermalizer developed a guide for kebab cafes in St. Petersburg. To solve a problem of one-sided love the programmers made the Perfect Wife service, which was awarded with the Audience Choice Prize. The team made two chat bots: the first one receives a photo of a girl and then sends an image with a porn actress who resembles her. The second one can detect a nationality by analyzing portraits of people. Other developers presented apps that generate sophisticated comments to photos on social networks, apply filters to images making people look like famous writers, and so on.
There were also serious projects. The programmers from "Sindicator" made a hybrid intelligence system, which helps make decisions in the context of high uncertainty in the financial market.
This hybrid intelligence comprises the collective mind of financial analysts who make forecasts every day, and an AI that uses machine learning methods. The system gatheres the forecasts made by specialists and generates its own ones using information on financial quotations and implications so as to make relevant investment bargains.
Some projects were related to social aspects and security. The Cline system gathers information about lines, detects how many people are in wait and measures various metrics. Cline’s team made a program, which identifies people in a line as separate objects and color them with different colors.
The "Data Fuel" service aims to control manipulation techniques used in social networks. This year, the audience of VKontakte became larger than those of the Channel One [The main TV channel in Russia -Ed.].It means that social networks become more and more popular nowadays.
AI Hackathon's winners
"It is easy to control minds using social networks. We created a tool that helps limit social interactions," say the developers.
The programmers of the Little Bro service also made a tool against virtual manipulations. It allows limiting propaganda during an election campaign by blocking profiles with political information.
Here are the top-three projects: HowPop placed 1st at the AI Hackathon. Its developers created an AI copywriter. The second prize was given to "DMPipeline," a service for modeling molecules of medications. The programmers who placed 3rd made the "Lung Diagnostics" service, which diagnoses respiratory tract diseases.