Uploaded October 2026 | Updated November 2026, 3 hours ago
Oxford Seminar, October 9 2026
You can view the listing for this talk online at https://topos.institute/events/oxford-seminar/talks/2026-10-09_corfield_sweedler.html
Speaker: David Corfield
Full Title: Sweedler Theory for Systems
Abstract: Starting from theories of interaction, or Application Programming Interfaces with laws, we ask how one such theory, T, can be implemented inside another, R, by a stateful mechanism. Described via monoids in a duoidal setting, Sweedler theory supplies a universal object, M(T, R), whose coalgebras are precisely such mechanisms. These are known as runners. Our central systems-theoretic proposal is to organise theories, direct interpretations, stateful runners, and transformations of runners into a double-categorical architecture Run(D), relative to a given duoidal Sweedler setting D. Tight maps say how operation languages are reinterpreted in each other. Loose maps say how one operation language is realised by a stateful mechanism in another. Run(D) affords us a rich structure to investigate systems composition and behaviour. Varying D allows us to cover a wide range of circumstances, such as systems with certificates, costed, smooth, probabilistic and higher-dimensional systems.
Oxford Seminar, October 9 2026
You can view the listing for this talk online at https://topos.institute/events/oxford-seminar/talks/2026-10-09_corfield_sweedler.html
Speaker: David Corfield
Full Title: Sweedler Theory for Systems
Abstract: Starting from theories of interaction, or Application Programming Interfaces with laws, we ask how one such theory, T, can be implemented inside another, R, by a stateful mechanism. Described via monoids in a duoidal setting, Sweedler theory supplies a universal object, M(T, R), whose coalgebras are precisely such mechanisms. These are known as runners. Our central systems-theoretic proposal is to organise theories, direct interpretations, stateful runners, and transformations of runners into a double-categorical architecture Run(D), relative to a given duoidal Sweedler setting D. Tight maps say how operation languages are reinterpreted in each other. Loose maps say how one operation language is realised by a stateful mechanism in another. Run(D) affords us a rich structure to investigate systems composition and behaviour. Varying D allows us to cover a wide range of circumstances, such as systems with certificates, costed, smooth, probabilistic and higher-dimensional systems.
![[2-torial] Tim tells Jason about Deformation Theory [3/3]
[2-torial] Tim tells Jason about Deformation Theory [3/3] [2-torial] Tim tells Jason about Deformation Theory [3/3]](https://i.ytimg.com/vi/vxceguk8DRY/mqdefault.jpg)
![Kate Fleming: Beyond Tech Solutionism: Challenges of unlocking under-served community knowledge
Topos Institute Colloquium, 10th of April 2025.
———
[Abstract TBA] Kate Fleming: Beyond Tech Solutionism: Challenges of unlocking under-served community knowledge](https://i.ytimg.com/vi/w-0NbzoX8-M/mqdefault.jpg)
![[2-torial] Tim tells Jason about Deformation Theory [2/3]
[2-torial] Tim tells Jason about Deformation Theory [2/3] [2-torial] Tim tells Jason about Deformation Theory [2/3]](https://i.ytimg.com/vi/w1tKHbphFlc/mqdefault.jpg)
![Inna Zakharevich: The category of schemes is abelian (and other obviously-false true things)
Topos Institute Colloquium, 29th of January 2026.
———
[Abstract TBA] Inna Zakharevich: The category of schemes is abelian (and other obviously-false true things)](https://i.ytimg.com/vi/w4foQ_0iMSI/mqdefault.jpg)

![[Oxford Seminar] David Corfield | Categorical systems theory: control and emergence
Oxford Seminar, March 5 2026
Speaker: David Corfield
Also appearing in this recording are David Jaz Myers and Matteo Capucci.
Full Title: Categorical systems theory: control and emergence
Abstract: This will be an informal session with plenty of discussion time, investigating a couple of concepts that arise from the category-theoretic treatment of systems.
*(1) Control*
How inputs to a system regulate its behaviour. Starting points: (a) In response to the active inference program, some recent articles (e.g., https://arxiv.org/abs/2406.07577 and https://arxiv.org/abs/2508.06326) have looked to understand autonomous systems as composed of agent and controller subsystems, equipped with dual interfaces; (b) Ordinary Lyapunov functions have been treated category-theoretically (https://arxiv.org/abs/2502.15276), work that should be extendable to variants. Where ISS (input-to-state stability) Lyapunov functions concern stability under any external perturbation, control Lyapunov functions concern stability under a chosen input.
*(2) Emergence*
Phenomena where the composite behaviour of the parts does not equate to the behaviour of the composite. Starting points: (a) Elie Adams thesis, Systems, Generativity and Interactional Effects (https://elieadam.com/eadam_PhDThesis.pdf); (b) Puca et al. on Failures of compositionality (https://arxiv.org/abs/2307.14461) (c) Erik Hoel on causal emergence (e.g., https://arxiv.org/abs/2202.01854). Two relevant CT constructions appear to be laxness of functors and coarse-graining as epimorphisms, potentially fitting well with a double category-theoretic outlook. [Oxford Seminar] David Corfield | Categorical systems theory: control and emergence](https://i.ytimg.com/vi/wSWmHZNjpzg/mqdefault.jpg)
![[TopOx] Tom Leinster: The many faces of magnitude
16th of February 2026. Slides available at https://topos.institute/events/topox/
The magnitude of a square matrix is the sum of all the entries of its inverse. This strange definition, suitably used, enables us to define the magnitude of many objects in different contexts across mathematics. All of them can be seen as measures of size. For example, the magnitude of a metric space combines classical quantities such volume, surface area, and dimension. The magnitude of a category is closely related to Euler characteristic. The magnitude of a graph is an invariant sharing features with the Tutte polynomial (but not a specialization of it). Magnitude also appears in the difficult problem of quantifying biological diversity: under certain circumstances, the greatest possible diversity of an ecosystem is exactly its magnitude. And there is now a theory of magnitude homology, which has the same relationship to magnitude as ordinary homology does to Euler characteristic. I will give an aerial view of this landscape. [TopOx] Tom Leinster: The many faces of magnitude](https://i.ytimg.com/vi/wxqfxRCGoOE/mqdefault.jpg)


![[Berkeley Seminar] Bryce Goldman (Topos Institute) | ModalTT
Title: ModalTT: a type theory for modal virtual double theories
Abstract: CatColab’s DoubleTT module uses virtual double theories to parameterize logics, which determine a type theory on models. These theories are currently hard-coded in CatColab’s backend, and their specification is not exposed directly to the user. This talk introduces an implementation of a user-facing DSL for specifying custom (modal) virtual double theories. This language is backed by a type theory for virtual double categories (VDCs), inspired by prior work by Hayato Nasu on an internal language of VDCs. We propose a handful of modifications to Nasu’s type theory, which integrate support for double monads, expose the innate resource-sensitivity and order-sensitivity of the internal logic, and better facilitate a translation into software.
Date: August 18, 2026 [Berkeley Seminar] Bryce Goldman (Topos Institute) | ModalTT](https://i.ytimg.com/vi/xK1KIaVo0Uo/mqdefault.jpg)
![[Oxford Seminar] Matthew Daggit | Developing support for end-to-end verification of neural AI agents
Oxford Seminar, 1st of May 2025,
Full Title: Developing programming language support for end-to-end verification of neural AI agents
Over the past decade, there has been impressive progress in developing algorithms capable of formally verifying first-order, linear specifications about small-to-medium-sized neural networks. This has opened the door to formally verifying the correctness of neural AI agents operating in complex environments. However, due to the complexity of integrating these verification algorithms with other reasoning tools and the many moving parts involved, deploying them in practice has proven more challenging than anticipated.
In this talk, I will introduce Vehicle, a tool that my collaborators and I have been developing. Vehicle allows users to: (i) write specifications for AI agents in a high-level, domain-specific language; (ii) use these specifications to guide training; (iii) formally verify whether the resulting neural network satisfies the specification; and (iv) export verified specifications to interactive theorem provers where the agents can be reasoned in the context of their environment. In the talk I will highlight how we leverage type theory in unusual ways to improve the user experience, our use of real-valued logic semantics to interpret Boolean specifications, and finally some of the open theoretical challenges in the field.
Links:
https://github.com/vehicle-lang/vehicle [Oxford Seminar] Matthew Daggit | Developing support for end-to-end verification of neural AI agents](https://i.ytimg.com/vi/xK8Kw-g0mNs/mqdefault.jpg)