Complexity Theory

Complexity theory investigates the inherent difficulty of computational problems, currently focusing on the limitations of deep learning models in handling complex reasoning and function composition, as well as exploring the computational complexity of reasoning about agent expectations and observations. Research employs various approaches, including analyzing the structure of neural networks, applying tools from arithmetic circuit complexity to design efficient structured matrices for deep learning, and using methods from time series analysis and network theory to study natural language. These investigations aim to improve the efficiency and capabilities of artificial intelligence systems and provide a deeper understanding of complex systems in general.