Using Generative Models to Improve Generative Models - Data-Centric Efficient AI

Watch a 20-minute conference talk from UC Davis Assistant Professor Yubei Chen exploring how multiple generative models can be integrated to create sophisticated training data and address data scarcity challenges in edge AI applications. Learn about the critical role of training data in defining model knowledge, the concept of a versatile "data simulator" for task-specific knowledge generation, and the strategic combination of different generative models that each replicate distinct aspects of reality. Discover a novel method for leveraging the fundamental strengths of multiple models working in concert, leading to both improved generative capabilities and potential solutions for data-limited edge AI scenarios. Gain insights into this data-centric approach as an important stepping stone toward developing comprehensive world models and scaling efficient AI systems.