Explore a comprehensive lecture from the Topos Institute Colloquium that delves into the intersection of categorical causality and systems theory, examining how systems evolve based on inputs and produce observable outputs. Learn about the framework's application to classical physical systems and its relationship to Partially Observed Markov Decision Processes in AI. Gain insights into causal models using string diagrams in CD-categories, understanding their limitations and applications. Discover key concepts including structural causal models, intervention vs conditioning, causal representation learning, group theory, Markov categories, and open dynamical systems. Through examples like language modeling and domino demonstrations, understand why certain systems cannot be accurately described by causal models, highlighting important limitations in current causal modeling frameworks. Examine the future of AI through the lens of causal abstractions and explore the complexities of ambiguous interventions in system theory.
Categorical Causality and Systems Theory - From Structural Models to Dynamical Systems
Overview
Syllabus
Introduction
About Taco Cohen
Why study globality
Why study category theory
Outline
Structural causal model
Causal base net
Intervention vs Conditioning
Invariance
Demo
Language Modeling
Potential Outcome Function
Causal Representation Learning
Causal Models
Group Theory
Markov Categories
Free CDU Categories
Comp Factorization
Future of AI
Causal Models are Abstractions
Ambiguous Interventions
Other Issues
Domino Example
Pearl Galas Halpern
Open dynamical systems
Systems Theory
Taught by
Topos Institute
