Neural Network Trained in St. Petersburg to Identify Vulnerabilities in Text-Based CAPTCHAs

Development will enhance the security of Internet resources

Specialists from St. Petersburg State University (SPbSU) and the St. Petersburg Federal Research Center of the Russian Academy of Sciences (SPb FRC RAS) have trained a neural network to identify vulnerabilities in computer tests (CAPTCHA). These texts help determine whether a user or a computer bot is trying to access a website. This was reported by the press service of SPbSU.

The neural network was trained on a special dataset — a set of data for machine learning.

The main difference between our approach and its analogues is that it allows recognizing much more complex text tests, which contain more than 20 distortions at once.
Anastasia Korepanova, lecturer at the Department of Informatics at SPbSU 

The developers emphasized that their method will help improve the security of Internet resources.

Information security specialists can refine the methods of recognizing and bypassing CAPTCHAs, and on this basis develop improved algorithms for protection against automated attacks.
Maxim Abramov, Head of the Laboratory of Theoretical and Interdisciplinary Problems of Informatics at SPb FRC RAS

The new method made it possible to recognize 63% of distorted CAPTCHA images. This indicates the insecurity of sites using this type of computer texts, the St. Petersburg specialists summarized.

Earlier www1.ru reported that Yandex Cloud presented a new neural network-empath that recognizes emotions in human speech.

Read materials on the topic:

Russian AI will start making voice comments to scooter riders and smokers

A neural network has been developed in St. Petersburg to detect brain hematomas on CT scans

Russian neural network LAPUSKA: the development of NUST MISIS and HSE University promises to improve the quality of photos twice as fast

Sources
TASS
2024-05-28 15:10:55 Neural networks SPb FIC RAN SPbGU
Telegram Дзен

Now on home

All news