September 26, 2024, Bangalore
During the AI Journey session of Cypher 2024 in India, professionals from Sber, AIRI – Artificial Intelligence Research Institute, and Skoltech spoke about the evolution of Gen AI, how engineering AI is taking us closer to artificial general intelligence, the use of AI in 3D content generation, data management when creating new materials, and other key tracks of the current AI evolution.
In his report called “Gen AI: From Digital Modalities to the Physical World”, Semyon Budyonny, PhD in math and physics, managing director and head of the Advanced AI Technologies Development at Sber, and scientific advisor at AIRI, spoke about the main Gen AI trends and the latest research results regarding AI-driven generation of functional materials, which are being used in a wide range of tasks: from energy storage technology to the creation of superstrong compounds. He noted that the emergence of Gen AI marked the beginning of the M3 era, which means multitasking, multivalence, and multimodality. The world of modalities is not limited to numbers, as physical modalities, including representations of molecules and materials, are increasingly grabbing the attention of researchers.
Yevgeny Burnayev, PhD in math and physics, director of the Skoltech Applied AI Center, head of the Learned Intelligence Group at AIRI, spoke about the present and future of engineering AI. It is not yet AGI, but it is no longer merely a tool to address highly specialized tasks with the help of neural AI. Today, it is a multi-agent system where symbolic AI is being used to organize the process of solving engineering problems and extracting relevant industry knowledge, while the neural AI it controls is used to automatically generate computational processes for solving the task at hand. Mr. Burnayev also presented several examples of using the concept of engineering AI in real life, including for generative design tasks.
Yana Charuiskaya, executive director of Experimental Machine Learning Systems at Sberbank, spoke about global experience in applying an innovative two-stage approach to 3D content generation using AI. The method includes a pre-trained 2D diffusion model, such as Kandinsky AI, to create consistent images of an object and universal reconstruction models to generate 3D representations. Streamlined training converted into top global quality at lower computational cost, while the use of advanced rendering techniques ensured high-quality geometry and textures of 3D models. According to her, this approach democratizes the creation of 3D content, making it accessible to mass users, and may be used in various verticals, from the gaming industry to virtual reality.
Sergey Palamarchuk, director of Sberbank’s Digital Customer Profile Development Division, described how one works with a customer profile using artificial intelligence. To date, Sber stores, updates, and protects 108 million profiles of its customers and manages the data efficiently. To identify and anticipate customer needs, the company uses classic methods: features, hashtags, sequences, graphs, and data reconstruction. Sber is also experimenting with new methods. Using GigaChat as an example, the speaker explained how AI helps automate data collection and create automatic customer hashtags.
AI Journey is one of the world’s biggest international conferences on artificial intelligence. It brings together top experts, researchers, and businesses from all over the world to discuss the most trending AI topics. This year, the AIJ track is included in the agenda of leading international AI conferences in India, Uzbekistan, and the UAE.
Cypher 2024 is the biggest AI summit in India and a flagship conference organized by Active Interest Media, one of the leading media companies focused on AI.