Prof. Tobias Fritz | Markov Categories

Prof. Tobias Fritz | Markov Categories

by Prof. Tobias Fritz

Formal & Physical Sciences Mathematics PBMathematicsPBFAlgebra
🎙 Prof. Tobias Fritz 👥 2K 📅 February 12, 2026 ⏱ 91 min 👁 178 📄 lecture 🧭 2026-08-15
Available in: English (current) Français

Keywords

Markov categoriescategorical probabilitycausal inferenceconditional independencestring diagrams

Summary

This is a recording of a seminar talk by Professor Tobias Fritz at the Isaac Newton Institute, part of a programme on causal inference. The talk is the second part of a series on Markov categories. Fritz begins by recalling examples of Markov categories from the previous talk, including FinStoch, Stoch, TimeStoch, Gaussian probability, and UnknownFun. He then discusses the concept of conditionals in Markov categories, explaining that a category has conditionals if every morphism with two outputs can be factored into a part generating the first output and a conditional for the second. He connects this to the factorization criterion in causal models. The main focus of the talk is on a property called the ‘causality axiom’, which states that if two morphisms are equivalent when a variable is marginalized, they remain equivalent when that variable is observed. Fritz proves that every Markov category with conditionals satisfies the causality axiom. He also discusses conditional independence and its relation to d-separation in DAGs, based on a master’s thesis of his student. The talk is highly technical and assumes a background in category theory and probability.

185 words

Critical Evaluation

Value of the Information & Strength of the Argument

The talk provides a deep and rigorous introduction to Markov categories and their applications to causal inference. The speaker clearly explains the definitions and motivations, and he proves the main theorem (conditionals imply causality) in a step-by-step manner using string diagrams. The argumentation is solid and well-structured. The discussion with the audience adds value by clarifying subtle points and connecting the abstract concepts to practical considerations in causal modeling.

Scientific Rigor, Source Quality, Title Accuracy

The talk is scientifically rigorous, with precise definitions and proofs. The speaker is an expert in the field, and the talk is part of a formal seminar series at the Isaac Newton Institute. The title accurately reflects the content. The description provides links to the institute’s website and the specific seminar page, which are relevant sources. No external sources are cited in the talk itself, but the mathematical content is self-contained.

155 words

Title / Content Match

The title accurately reflects the content: the lecture is about Markov categories, a mathematical concept.

Quality & Reliability

9/10

The lecture is given by a professor at a renowned research institute (INI), part of a formal seminar series. The content is mathematically rigorous, with proofs and precise definitions. The speaker is an expert in the field. The video is a recording of a live seminar, so there is no editing or post-production, but the mathematical content is reliable.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Cited Sources

  • INI Seminar page — The seminar page for this talk, part of the Causal Inference programme.
  • Isaac Newton Institute — The institute's main website, providing information about the research programme.

Concurring Sources

  • Markov Categories — The foundational paper on Markov categories by Fritz and others.

External References

Contribution & Novelties

This talk presents a categorical framework for probability and causality, unifying various probabilistic concepts. The main novelty is the introduction of the ‘causality axiom’ and the proof that it follows from the existence of conditionals. This provides a new perspective on the foundations of causal inference.

Pour aller plus loin :

  • Markov category — Wikipedia article on Markov categories, providing background and references.
  • String diagram — Wikipedia article on string diagrams, the graphical language used in the talk.
  • Causal inference — Wikipedia article on causal inference, relevant to the applications discussed.
  • D-separation — Section on d-separation in Bayesian networks, relevant to the discussion of conditional independence.

106 words

Radar Profile

The radar profile shows very high scores in all dimensions, with a particularly high level of technicality. This indicates a highly specialized and rigorous mathematical lecture, suitable for experts in the field.

Reliability 9/10