Scientists at the School of Economics at the National Research University of Russia claim to have succeeded in creating an algorithm that would help neural networks to identify individuals through the content of their galleries.
Neural networks are computer models capable of performing machine learning as well as pattern recognition, inspired by an animal’s central nervous system and have several practical applications, such as helping virtual assistants to sound more human.
Without the presence of technology, the gender of a person is correctly identified in 90% of cases, and in the age issue, the process is much more uncertain. In the classification, there is no age-specific, but age groups and, analyzing each individual frame, the network concludes with percentage probabilities for each group.
Russian researchers, led by Professor Andrey Savchenko, have implemented a feature in the new algorithm that would aggregate percentages of age brackets into statistics and the Dempster-Shafer theory. The theory of evidence allows us to combine evidence from a variety of sources and reach a degree of credibility that takes into account all available evidence. That is, it would turn the percentages into statistics and use all the elements to reach the conclusion of the age of the person in question.
The method does not only include one neural network, but several, for each factor to be defined; one for age, one for gender, and so on. But when combined, they not only define the factors individually but also assign a unique vector to each person who passes through the process.
The algorithm has arrived in a new phase of tests and has already begun to be applied in some applications for Android. Unlike some existing systems, such as Instagram, it works even without an internet connection, processing the contents of the gallery to perform the data collection; analyzing and collecting information from intimate circles, close friends, family members, and initiating and conducting the process of assigning the unique vector to all circle members.
The new algorithm can and should be used for smartphone manufacturers to create new recommendation systems with greater ease and accuracy. If, for example, a user has too much content in his gallery with a child, as would tend to notify him of products related to toy stores and games.
Savchenko warns that system servers do not have access to user photos and videos, only to reports generated by the algorithm, meaning that they will not share data, as is common in recent news. The information that would be acquired depends on the user. It may be that the main circle of the user has 3 women and 2 men, or that he frequents a lot of fast food networks, etc.
The interest in technology is wide, with one of the largest smartphone manufacturers already eyeing the development and use of the algorithm. And although it does not name companies, it is very likely that the Samsung, Huawei or Apple giants will use the system on their handsets in the coming years.