Givi Meishvili, a research scientist at Microsoft, will be speaking at DataFest Tbilisi 2024. His research interests include novel view synthesis and 3D reconstruction of faces, computational photography in the context of human faces, and multimodal learning. As we eagerly await his insights at the event, take a look at his interview with On.ge for a preview of his innovative work and perspectives.
Q: Could you tell us more about your educational background and professional experience in Georgia and abroad? What led you to your current role at Microsoft?
A: My journey in computer science began in 2008 when I was admitted to the Faculty of Exact and Natural Sciences at Tbilisi State University (TSU). I received a double bachelor's degree in computer science from TSU and Université Paris 8 in France.
I then decided to pursue my master's degree at the University of Geneva, Switzerland. At the same time, I worked as a Java Software Developer at the United Nations Office in Geneva. I received my master's degree in 2017 and subsequently began a Ph.D. program at the University of Bern, Switzerland, focusing on AI and Neural Networks.
During my Ph.D. studies, I participated in various prestigious international conferences, presenting multiple scientific papers. I defended my Ph.D. thesis titled “Learning Representations for Controllable Image Restoration” in 2020. During my doctoral studies, I worked on 2D and 3D modeling of human faces, as well as developing methods to reconstruct faces based on audio and video signals. Concurrently, I also completed internships at Disney Research and Microsoft Research. After defending my thesis, I was offered a Research Scientist position at Microsoft two years ago.
Q: Could you describe your work at Microsoft and tell us why it interests you?
A: Currently at Microsoft, I am focused on developing 3D avatars of humans. This involves creating digital reconstructions of 3D faces, eyebrows, hairstyles, and accessories. My interest in this field initially sparked during my master’s studies and deepened significantly during my Ph.D.
During my doctoral research, I became fascinated by how humans acquire and deepen their knowledge at a neural level—a question that remains unanswered and is actively researched today. I am particularly intrigued by the development of neural networks in machine learning to solve various problems, such as object classification and segmentation in images, and beyond. Additionally, I am passionate about advancing machine learning to enable systems to autonomously gather, analyze information, and make decisions without human intervention. This pursuit aims to streamline processes and enhance efficiency across different domains.
Q: What makes avatars so important at Microsoft, and in which of their products are they featured?
A: "To empower every individual and every organization on the earth to accomplish more", - this is our slogan at Microsoft. This is exactly what motivates us to develop and refine our many products.
Microsoft Teams is one of Microsoft's most famous products that features AI. I believe, digital avatars are key to increasing our productivity while working remotely.
When creating digital avatars, authenticity is crucial, especially in capturing individual intricacies such as hairstyle and facial features. Furthermore, avatars should be as realistic as they can get. Another factor to consider is inclusivity, focusing on how well avatars can be generated to represent diverse ethnic groups.
Q: To what extent does AI streamline avatar creation and why?
A: Recent advancements across various branches of AI have been essential prerequisites for the automatic generation of avatars. Advancing neural networks, which by itself includes many technical nuances, is the precondition for creating avatar components.
Artificial neural networks' capability to analyze large datasets enables the creation of digital models of human faces that reflect people’s appearances with near-perfect accuracy.
Q: What technological obstacles do you encounter most often? What are the challenges of creating digital models of human faces?
A: In our work, we frequently encounter technological challenges. For instance, vast amounts of digital data are required to train neural networks. This data needs to encompass various characteristics of the human face, each contributing to our physical identity. Handling such sensitive information, that requires meticulous attention, for efficiently training neural networks is truly difficult.
Moreover, ensuring that our avatar models accurately represent individuals poses another challenge. For instance, individuals may have unique features such as rare hairstyles that the neural network must have encountered during its training to reconstruct accurately. Therefore, the diversity and richness of the training dataset are crucial for achieving comprehensive representation.
Another critical challenge relates to efficiency. While achieving high-quality avatar reconstruction is essential, the speed at which these models can generate avatars is equally important. This efficiency is essential to minimize waiting times for customers and enhance user experience.
Q: Could you share your thoughts regarding the Metaverse? Is this the potential peak of avatar technology? What should be our expectations for the future?
A: Even just a decade ago, discussions about the Metaverse were virtually nonexistent. Back then, the primary challenges in our work revolved around object classification.
However, significant advancements in various types of artificial intelligence in recent years have transformed our technological landscape. These advancements have provided us with the tools and capabilities to develop products like the Metaverse, which seemed almost unimaginable just a decade ago. Currently, our focus is on advancing Metaverse technologies, yet I believe that ongoing technological progress will continue to unveil many other intriguing products in the near future.
I consider myself fortunate to work in a field directly related to my Ph.D. thesis at Microsoft. Microsoft is a company that heavily invests in the development of artificial intelligence, actively integrating various AI systems into its products. Therefore, it is poised to make substantial contributions to the technologies that will shape our future.
Q: It has come to our attention that you will be one of the speakers at DataFest Tbilisi 2024. Could you elaborate on the topics you plan to discuss and what valuable insights attendees can expect to gain from your presentation?
A: At this year's DataFest Tbilisi, my presentation will delve into the realm of digital technologies applied to modeling, restoring, and processing the human face. Specifically, I will examine the advancements achieved in this field over the past 10 to 25 years. Additionally, I will provide insights into the current state of technological development in these areas.
Q: Why is organizing events like DataFest Tbilisi important for Georgia?
A: One of the main catalysts for the renaissance of artificial intelligence as a field over the last decade has been international conferences and the publication of scientific papers on the subject. These conferences provide a platform for researchers and practitioners to share cutting-edge developments, discuss emerging trends, and collaborate on advancing AI technologies. The availability of scientific papers allows anyone interested in the field to stay informed about its progress and contribute their own ideas and innovations.
Organizing events like DataFest Tbilisi is critically important because they provide a unique platform for sharing knowledge and accumulated experience in the use and development of artificial intelligence. I believe this event facilitates widespread exposure to significant achievements in the field, inspiring professionals and students alike to engage with AI and take steps toward its advancement in Georgia. Moreover, such festivals foster valuable exchanges of professional contacts and promote collaborations among companies. I am deeply appreciative of the organizers of this event for their role in fostering these opportunities and connections.