The MIT Categories Seminar is an informal teaching seminar in category theory
and its applications, with the occasional research talk. We (usually) meet each
**Thursday, 4.30pm to 5.30pm** in **MIT
2-255**.

Please email Brendan Fong (bfo (at) mit.edu) if you'd like to be on the mailing list.

** February 14 **: Brendan Fong

In this talk I'll introduce monads and their algebras, illustrating the
concepts using the finite probability distribution monad. In
particular, in addition to defining monads and algebras for monads,
I'll talk about free algebras and the Kleisli category for a monad.
Examples will include the power set monad, maybe monad, and the Giry
monad. We'll see how the notion of a stochastic matrix or conditional
probability distribution arises from this set up.

** February 21 **: Tobias Fritz (Perimeter Institute)

I will explain how monads often arise as Kan extensions of graded monads. For example, the set of lists over an alphabet is the disjoint union of the sets of lists of each length. I will then show how this leads to a construction of a probability monad similar to the Giry monad. The graded monad approach lets us replace the use of measure theory by the combinatorics of finite sets, and makes precise the idea that a probability measure is an idealized version of a finite sample.

Joint work with Paolo Perrone.
** February 28 **: Sam Tenka

Categorical Probability

** February 15 **: Brendan Fong

An overview of Coecke, Sadrzadeh, and Clark's *Mathematical foundations for a compositional distributional model of meaning*.

** February 22 **: David Spivak

Fuzzy simplicial sets I.

** March 1 **: David Spivak

Fuzzy simplicial sets II.

** March 8 **: Brendan Fong

An overview of McInnes and Healy's *UMAP: Uniform manifold approximation and projection for dimension reduction.*

** March 29 **: Brendan Fong

The equivalence of hypergraph categories and lax monoidal functors Cospan(FinSet) --> Set.

** April 5 **: Enrique Boroquez (UNAM)

The idea is to explore the concept "Point", to look at it from different perspectives and see the connections given by the Stone Duality and the Gelfand Duality. Starting from a topological perspective, a point as an element of a topological space, we'll see the correspondences with ultrafilters, global elements in the category of Stone Spaces, morphisms that have as codomain the initial algebra 2 and how this morphisms correspond to valuations of a language in classical logic. We'll then go to boolean valued models, see how the points there have a probabilistic behavior and see some interesting consequences given by doing the forcing method through principal ultrafilters vs non-principal ultrafilters.Then we'll see the equivalences given by the Gelfand Duality: global elements in the category of locally compact spaces, characters of commutative C* algebras, maximal ideals, pure states (as positive, normalized functionals) and irreducible representations.

** April 12 **: David Spivak

Videos: Part I, Part II, Part III, Teaser/Addendum

** April 19 **: Remy Tuyeras

How to do genetics with category theory I.

** April 26 **: David Spivak

** May 10 **: Remy Tuyeras

How to do genetics with category theory II: Recognition of biological mechanisms.

** May 17 **: Christoph Dorn (Oxford)

"Higher categories" are algebraic structures which generalise categories, in the same way that categories generalise sets. The theory of these structures is ubiquitous in many areas of application of category theory. In this talk, we will motivate higher structure from applications in logic, type theory and programming languages. Surprisingly, a formulation of higher categories which is amendable to computer implementation has been missing for decades — with homotopy type theory being a notable exception of the special case of “higher groupoids”. Starting from elementary observations about compositionality in higher dimensions, will sketch a combinatorial approach to higher categories that gives rise to a programming language which we call “Globular” and which comes with an elegant geometric model. We will see that programs in Globular can capture complex higher-dimensional geometric deformations, and learn how, for instance, they can capture higher homotopy groups of spheres and other interesting geometric facts.

** May 24 **: Tim Havel

(Slides)

System Dynamics is a moniker coined by Jay Forrester over 50 years ago at MIT, when computers were just beginning to be used to simulate the behaviors of complex models of social and economic processes. Today it has a substantial following in business schools, management consulting firms, and some multinational corporations, with the goal of teaching "Systems Thinking" to their respective stake holders (see "Business Dynamics" by John Sterman). While it has since branched out into more sophisticated agent-based and dynamic network models (see this short article), it was initially based on a diagrammatic language for nonlinear ODEs that continues to play a central role in its applications. This talk will introduce this language and give some examples, with the hope that the audience will discover some analogies, and new applications, for the categorical formalisms it has developed for modeling and analyzing systems in a broader sense.

** May 31 **: Brendan Fong

Causal theories: a categorical approach to Bayesian networks.

** June 7 **: David Spivak

** June 14 **: Remy Tuyeras

How to do genetics with category theory III.

** June 21 **: Brendan Fong

I'll talk about an algebraic setting (in particular, a category) in which supervised learning algorithms can be composed, and then explain how this articulates the structure underlying backpropagation on a neural network. More details here.

** July 19 **: David Spivak

Decomposition spaces are combinatorial objects that allow one to decompose maps in a natural (co-associative and co-unital) way. All categories are decomposition spaces, but not vice versa. Decomposition spaces form a category, called Decomp, where the maps preserve the decomposition structure. I'll discuss these ideas and recent work (joint with J. Kock) where we show that every slice category of Decomp is a topos. One of them is the topos I discussed recently: that of "discrete temporal type theory".

** July 26 **: David Spivak

** August 2 **: Antwane Mason (Rensselaer Polytechnic Institute)

Formal verification is the process of proving that a system adheres to a specification of its intended behavior. Like programming languages, there exists a variety of specification languages, each suited to describing distinct types of behavior. Some systems may require a combination of specification types and reasoning strategies to verify the whole system. This scenario motivates the need for heterogeneous specification languages. In this talk, we shall explore Farrell et. al.'s 2017 paper entitled "Combining Event-B and CSP: An Institution Theoretic approach to Interoperability". We shall see how the authors use category theory to establish interoperability between two specification languages, Event-B and CSP, focusing particularly on how they develop semantic preserving translations between the two languages.

** August 9 **: David Spivak

In last week's seminar, we heard from Antwane Mason about Goguen and Burstall's notion of institution, which is a category-theoretic formalism used for software specification and verification. The definition is usually phrased in terms of a satisfaction relation between models and sentences of signatures, and a certain bivariant coherence condition on it.

In this talk I will discuss a few different category-theoretic perspectives on this notion, including from categorical database theory, posetal bifibrations, and topos theory. In the order "signatures, models, sentences, satisfaction", we can view

- databases as an institution via "schemas, instances, embedded dependency constraints, satisfaction",
- any posetal bifibration as an institution via: "base objects, fiber elements, fiber elements, ≤", and
- toposes as an institution via "toposes, points, truth values, satisfaction".

** August 16 **: Brendan Fong

A simple tutorial on Kan extensions: their definition, relationship with (co)limits and adjoints, and some basic examples. There will be plenty of time for questions.

** August 23 **: David Spivak

An introduction to double categories

** August 30 **: Brendan Fong

2-categories and adjunctions

** September 10 **: David Spivak

Adjunctions and Mates

** September 17 **: Remy Tuyeras

In this talk, we will use the concepts of left Kan extension and monomorphism to describe Genome-wide association studies used in genetics. This will be the opportunity to review the concept of Kan extension and see its real world applications.

** October 1 **: Reuben Cohn-Gordon (Stanford)

In cooperative conversation, speakers tend to prefer informative utterances, and listeners assume the speaker is being informative in order to draw inferences. For instance, if you hear me say "John liked half of the concert", you infer that he did not like to other half, since as an informative speaker, I would then have said "John liked the concert", were that true. Bayesian models can be used to formalize these inferences.

I show that a model of an informative speaker (that says as much as is possible) can be defined as a right Galois connection (right adjoint between posets) to a literal listener, understood as a monotone function (functor between posets) from utterances to possible worlds. Dually, a pragmatic listener (that deduces that no more is meant that is said) is a left adjoint of a literal speaker, understood as a monotone function from possible worlds to utterances. While elementary, this points to a deeper category-theoretic formulation of Gricean pragmatics and more generally, cooperative games.
** October 11 **: Paolo Perrone (Max Planck Institute for Mathematics in the
Sciences)

Monads are a categorical concept which can be interpreted as encoding formal expressions, or formal operations in the sense of universal algebra. We give a construction which formalizes the idea of “evaluating an expression partially”: for example, “2+3” can be obtained as a partial evaluation of “2+2+1”, and conversely, we can view “2+2+1” as a partial decomposition of “2+3”. This construction can be given for all monads on a concrete category, and it is linked to a simplicial object called the bar construction, of which it gives an operational interpretation: the bar construction is a simplicial set, and its 1-cells are partial evaluations.

We study the properties of partial evaluations for general monads on concrete categories. We prove that whenever the multiplication of the monad is weakly cartesian, partial evaluations can be composed via the usual Kan filler property of simplicial sets, of which we give an interpretation in terms of substitution of terms.

For the case of probability monads, partial evaluations correspond to what probabilists call conditional expectation of random variables, which is used to define martingales. It follows that a martingale is equivalently described as a chain of partial decompositions.

This talk is part of a work in progress on a general operational interpretation of the bar construction.

This is joint work with Tobias Fritz.
** October 15 **: David Myers (Johns Hopkins University)

In order to make a model of some system, we have in mind the sort of things in that system we intend to model; we have an ontology of our model, and we hope the things in this ontology correspond to the things in the system. But coming up with such an ontology is more of an art than a science. A model's success is judged on its ability to predict, not how much it matches our intuitive ontology. What things do our best models actually model?

Can we find a "natural ontology" of a model, one that comes from the model
itself and contains only the things the model is modelling? What is a thing,
anyway? In this talk, we will look at a categorical approach to these problems
using the yoga of *behavior* types.

** October 29 **: Lee Mondshein

How can category theory be productively applied to extend concepts and techniques of logic and linguistics, so as to elucidate the behavior of biological networks?

I will discuss some concrete initial steps, based upon current categorical constructs in logic and topology, and will explain the relevance of emerging semiotic ideas concerning meaning-making and meaning transformation in biological networks.

** November 5 **: Remy Tuyeras

Linkage disequilibrium is the existence of a non-random relationship between two loci in the genome. I will show how one can talk about this type of relationship using a certain type of pedigrads in the category of idempotent commutative monoids.

** November 19 **: Ed Wike

(Slides)

A first look at a software development effort to implement the causal theory
methodology in Brendan Fong's masters thesis, "Causal Theories: A Categorical
Perspective on Bayesian Networks" (presented by Brendan at a seminar in May).
This talk, which includes a demo of the prototype version, will have a brief
overview of causal modeling, the categorical framework for causal theory models,
the Python software used, how the users can draw causal graphs with the program,
and plans for ongoing development. We will discuss the applicability of the
causal theory model program and the potential benefits of applying category
theory to analytical modeling.

** November 26 **: David Spivak

Brendan and I have been developing a new categorical viewpoint—and graphical calculus for—regular logic. I've recently been thinking about how some of the surrounding ideas might apply to learning and adaptation via something I might call a "theory building" adjunction. Some of these ideas seem to connect to Reuben Cohn-Gordon's recent seminar talk on the "Gricean adjunction" from pragmatics.

The ideas I'll present are at an early stage in terms of maturity, and the goal of the talk is to be fun and pictorial. Still, I will certainly make connections to the underlying mathematical abstractions.
** December 3 **: Alex Kavvos (Wesleyan)

It is informally understood that the purpose of modal (i.e. unary) type constructors in programming calculi is to control the flow of information between types. I will introduce a number of such constructors that are useful for security-aware functional programming, i.e. for writing programs without accidental high-level leaks of information flow. Following that, I will prove that the well-typed programs in calculi of this sort preserve confidentiality and integrity by design, by using some categorical algebra and a few background ideas from topology. (This talk is based on arXiv:1809.07897.)

** December 10 **: Brendan Fong

I'll present some foundational ideas behind the work David and I have been doing on graphical regular logic. In particular, I'll give a number of characterisations of the notion of meet semilattice. The aim will be to not only give an understanding of meet semilattices, but to illustrate general categorical lessons about limits, adjunctions, and monoidal functors.

** January 31 **: Jules Hedges (Oxford)

Open games are a foundation of non-cooperative game theory that is strongly
compositional: all games are built from small pieces using sequential and
parallel composition operators. They form the morphisms of a symmetric monoidal
category, and string diagrams provide a useful and intuitive way of visualising
them that is an alternative to the traditional extensive form. I will focus on
the foundations, which involves a careful analysis of how observable and
counterfactual information interact in a game. Time permitting I'll discuss a
close connection with open learners, from the paper Backprop as Functor by
Fong, Spivak and Tuyéras.

** February 7 **: Brendan Fong

An overview of Ellerman's *Partition Logic*.