• With innovation, creators can embed unique "fingerprints" and protect their rights
• I studied computer vision and neural networks at the Technical University 35 years ago, says the Deputy Rector Prof. Milena Lazarova, who won numerous awards at global competitions organized by Microsoft and Intel with her students
We meet with the Deputy Rector of the Technical University - Sofia, Prof. Milena Lazarova, at a moment when a swarm of enthusiastic young people, eager to absorb the synthesized knowledge of one of our best specialists in the field of artificial intelligence and machine learning, "buzzes" around her.
The reason is not only that she participates in more than 20 research projects co-financed by the EU, and is the head of some of them. Not only has she been trusted to be a reviewer for articles in the publications of the Institute of Electrical and Electronics Engineers (IEEE) - the largest professional organization in the world of scientists and technologists focused on advanced technologies, and in Journal Neural Computing and Applications. She herself has won awards for the best paper at one of the elite international scientific conferences CompSysTech and Computer Science for years.
Young people are always around her because she walks with the halo of a top mentor, with whom they reach the finals of various global competitions of companies such as Intel and Microsoft. According to her, everything lies in the preparation she received 35 years ago at her beloved TU - Sofia. As early as 1990, her mentor and later scientific supervisor Prof. Gacho Gachev taught computer vision. It was he who ignited the spark in her, and even then she embarked on this innovative field. Although it may sound extremely strange from today's perspective, in 1996 Milena Lazarova created her first neural network, and it was the hard way - without ready-made templates, without programs developed by someone else that needed to be "assembled", without convenient languages like Python. She programmed everything in C++.
"At that time, there were no data, no publications, no possibility to read whatever you wanted on the internet - she recounts. - The work was hard, the current conveniences did not exist, but this is the truth about artificial intelligence at TU - Sofia, AI was taught here as early as the 90s. Even computer vision appears as a subject in my diploma. Of course, we used the traditional approach then, which is very different from the modern one, because only recently was there a breakthrough with supercomputers capable of processing large volumes of data, with graphics processors."
Despite the difficulties in the past, thanks to her scientific supervisor, Milena Lazarova began working on her doctoral thesis dedicated to artificial intelligence, and the first task was the analysis of data from satellite images.
"We were trying to recognize and classify land surface data," she recalls. After several months of training at the University of Twente, Netherlands, she devoted herself to working with students, and in those distant times, their first challenge was to develop a system for detecting emerging fires by collecting data from satellites. "This task is currently solved, but it was on the agenda back then," she recalls. "The students themselves approached me asking me to become their mentor. Parallel computing was developing at that time, our project was approved, and we reached the world finals, and Microsoft invited us to their headquarters in the Seattle area."
The success was such that the following year, a new group of ambitious students came to Prof. Milena Lazarova, firmly determined to show their capabilities on a global level. They received an award again, and then more and more.
Today, she and the team from the scientific group "Data Science and Cybersecurity" within the project BG-RRP-2.004-0005 "Improving Research Capacity and Quality for International Recognition and Sustainability of TU - Sofia" are inspired by a new major challenge. And it literally devastates democratic societies because it is related to disinformation, fake images, and videos.
With the help of AI, they are generated in an avalanche-like manner and become a threat not only to public figures, whose images can be abused, but also to ordinary people exposed to the risk of fraud, blackmail, and the spread of false information.
"The new methods being developed combine several types of analysis to distinguish real footage from artificially generated or software-manipulated ones," notes Prof. Milena Lazarova. "Moreover, generating unique 'fingerprints' for video data allows tracking the source of the manipulation and blocking unauthorized content."
How can all this be done, and is the end of AI theft coming?
"To understand whether an image or video is a fake generated by AI, we use another more advanced model to understand how this was done," Prof. Lazarova points out. "In one study, we explore how to create hybrid models, as it is not yet certain that AI can distinguish a fake image from a real one."
Yes, there is software for checking various forms of plagiarism, but accuracy is still relatively low there, as billions are poured into this technology globally, and models are becoming increasingly inventive. "Therefore, we are trying to create stronger defenses," Prof. Lazarova adds. "The same applies to images. Models generate increasingly better images, meaning the algorithm that will recognize them must be even more advanced, but in addition, we must combine it with other approaches to be able to trust it."
However, this is not an easy task. On the one hand, artificial intelligence requires a lot of data, and training and analysis take time. "The solution may lie in traditional techniques for detecting suspicious areas of the image, to which machines can be directed," explains Prof. Milena Lazarova. Therefore, she and her colleagues focus on face detection, because their unauthorized use violates not only rights but also constitutes a gross intrusion into personal space. "At this stage, we are concentrating on them, although the technology can be applied to any part of the image, including manipulated data from video games," she points out. "To this end, we are creating a unique fingerprint, from which it becomes clear who the author is and when
he published it. That is, when this image is added to fake information or stolen for some other purpose, it will contain a unique mark, and the author will be able to claim his rights."
The idea is that this innovation will curb the avalanche-like increase in speculation with frames and videos, which violate not only copyrights but also the reputations of people whose faces are abused. But regardless of this, the technology will have a much wider application, as it can be applied to texts, data, sensitive information, etc.
When asked what provokes the enthusiasm that is visibly transferred to students and PhD students, Prof. Lazarova says that she loves her work and daily contact with young people infinitely. "I go into a lecture and forget everything - she smiles. "The fact that they are there, that they are so inquisitive and asking questions, makes me continue to develop and encourages them as well. They are what drives the world forward. I cannot fully express how proud I am of our students."
According to her, increasingly capable young people are coming, and more and more of them choose to stay as teachers and researchers at TU - Sofia. "Sometimes I look at them and think that once I could have chosen to go to a prosperous private company, but today I am so glad that my fate turned out exactly this way - she adds. "I would not exchange my job here for any other. I am happy to have such luck and to be in this academic environment among so many wonderful colleagues and young talents."
24chasa.bg
• I studied computer vision and neural networks at the Technical University 35 years ago, says the Deputy Rector Prof. Milena Lazarova, who won numerous awards at global competitions organized by Microsoft and Intel with her students
We meet with the Deputy Rector of the Technical University - Sofia, Prof. Milena Lazarova, at a moment when a swarm of enthusiastic young people, eager to absorb the synthesized knowledge of one of our best specialists in the field of artificial intelligence and machine learning, "buzzes" around her.
The reason is not only that she participates in more than 20 research projects co-financed by the EU, and is the head of some of them. Not only has she been trusted to be a reviewer for articles in the publications of the Institute of Electrical and Electronics Engineers (IEEE) - the largest professional organization in the world of scientists and technologists focused on advanced technologies, and in Journal Neural Computing and Applications. She herself has won awards for the best paper at one of the elite international scientific conferences CompSysTech and Computer Science for years.
Young people are always around her because she walks with the halo of a top mentor, with whom they reach the finals of various global competitions of companies such as Intel and Microsoft. According to her, everything lies in the preparation she received 35 years ago at her beloved TU - Sofia. As early as 1990, her mentor and later scientific supervisor Prof. Gacho Gachev taught computer vision. It was he who ignited the spark in her, and even then she embarked on this innovative field. Although it may sound extremely strange from today's perspective, in 1996 Milena Lazarova created her first neural network, and it was the hard way - without ready-made templates, without programs developed by someone else that needed to be "assembled", without convenient languages like Python. She programmed everything in C++.
"At that time, there were no data, no publications, no possibility to read whatever you wanted on the internet - she recounts. - The work was hard, the current conveniences did not exist, but this is the truth about artificial intelligence at TU - Sofia, AI was taught here as early as the 90s. Even computer vision appears as a subject in my diploma. Of course, we used the traditional approach then, which is very different from the modern one, because only recently was there a breakthrough with supercomputers capable of processing large volumes of data, with graphics processors."
Despite the difficulties in the past, thanks to her scientific supervisor, Milena Lazarova began working on her doctoral thesis dedicated to artificial intelligence, and the first task was the analysis of data from satellite images.
"We were trying to recognize and classify land surface data," she recalls. After several months of training at the University of Twente, Netherlands, she devoted herself to working with students, and in those distant times, their first challenge was to develop a system for detecting emerging fires by collecting data from satellites. "This task is currently solved, but it was on the agenda back then," she recalls. "The students themselves approached me asking me to become their mentor. Parallel computing was developing at that time, our project was approved, and we reached the world finals, and Microsoft invited us to their headquarters in the Seattle area."
The success was such that the following year, a new group of ambitious students came to Prof. Milena Lazarova, firmly determined to show their capabilities on a global level. They received an award again, and then more and more.
Today, she and the team from the scientific group "Data Science and Cybersecurity" within the project BG-RRP-2.004-0005 "Improving Research Capacity and Quality for International Recognition and Sustainability of TU - Sofia" are inspired by a new major challenge. And it literally devastates democratic societies because it is related to disinformation, fake images, and videos.
With the help of AI, they are generated in an avalanche-like manner and become a threat not only to public figures, whose images can be abused, but also to ordinary people exposed to the risk of fraud, blackmail, and the spread of false information.
"The new methods being developed combine several types of analysis to distinguish real footage from artificially generated or software-manipulated ones," notes Prof. Milena Lazarova. "Moreover, generating unique 'fingerprints' for video data allows tracking the source of the manipulation and blocking unauthorized content."
How can all this be done, and is the end of AI theft coming?
"To understand whether an image or video is a fake generated by AI, we use another more advanced model to understand how this was done," Prof. Lazarova points out. "In one study, we explore how to create hybrid models, as it is not yet certain that AI can distinguish a fake image from a real one."
Yes, there is software for checking various forms of plagiarism, but accuracy is still relatively low there, as billions are poured into this technology globally, and models are becoming increasingly inventive. "Therefore, we are trying to create stronger defenses," Prof. Lazarova adds. "The same applies to images. Models generate increasingly better images, meaning the algorithm that will recognize them must be even more advanced, but in addition, we must combine it with other approaches to be able to trust it."
However, this is not an easy task. On the one hand, artificial intelligence requires a lot of data, and training and analysis take time. "The solution may lie in traditional techniques for detecting suspicious areas of the image, to which machines can be directed," explains Prof. Milena Lazarova. Therefore, she and her colleagues focus on face detection, because their unauthorized use violates not only rights but also constitutes a gross intrusion into personal space. "At this stage, we are concentrating on them, although the technology can be applied to any part of the image, including manipulated data from video games," she points out. "To this end, we are creating a unique fingerprint, from which it becomes clear who the author is and when
he published it. That is, when this image is added to fake information or stolen for some other purpose, it will contain a unique mark, and the author will be able to claim his rights."
The idea is that this innovation will curb the avalanche-like increase in speculation with frames and videos, which violate not only copyrights but also the reputations of people whose faces are abused. But regardless of this, the technology will have a much wider application, as it can be applied to texts, data, sensitive information, etc.
When asked what provokes the enthusiasm that is visibly transferred to students and PhD students, Prof. Lazarova says that she loves her work and daily contact with young people infinitely. "I go into a lecture and forget everything - she smiles. "The fact that they are there, that they are so inquisitive and asking questions, makes me continue to develop and encourages them as well. They are what drives the world forward. I cannot fully express how proud I am of our students."
According to her, increasingly capable young people are coming, and more and more of them choose to stay as teachers and researchers at TU - Sofia. "Sometimes I look at them and think that once I could have chosen to go to a prosperous private company, but today I am so glad that my fate turned out exactly this way - she adds. "I would not exchange my job here for any other. I am happy to have such luck and to be in this academic environment among so many wonderful colleagues and young talents."
24chasa.bg
