Learning Causal World Models from Acting and Seeing Using Score Functions

Explore a research lecture on learning causal world models through the integration of action and observation data using score functions. Delve into advanced methodologies for understanding causal relationships in complex systems by combining behavioral data with visual information. Examine how score-based approaches can be leveraged to construct robust causal models that capture the underlying dynamics of real-world environments. Learn about the theoretical foundations and practical applications of this approach in machine learning and causal inference. Discover how this methodology bridges the gap between passive observation and active intervention in causal discovery, offering insights into how agents can build comprehensive understanding of their environment through both seeing and acting. The presentation covers cutting-edge research at the intersection of causal inference, machine learning, and probabilistic modeling, providing valuable perspectives for researchers and practitioners working on causal discovery, reinforcement learning, and world model construction.