Learners are Almost Free Compact Closed

The category of learners has a pleasant symmetric formulation when the morphisms are considered up to a coarser equivalence than the one originally described in the paper "Backprop as Functor".A quotient of this modified category gives a new construction of the free compact closed category on a symmetric monoidal category.In this note I expand on an observation made at the end of [15] and further discussed online [16].If the definition of a learner [7] is tweaked slightly so that learners are considered identical under a coarser equivalence relation than before, a striking symmetry in the definition is revealed.This new category comes quite close to being compact closed.There is a strict symmetric monoidal involution and each object has a cup and cap morphism connecting it to its "dual" under this involution.However, the snake equations fail and so these are not true duals.Quotienting the category to force the equations to hold results in a compact closed category, and we end the note by showing that it is in fact the free compact closed category on the starting symmetric monoidal category.Compact closed categories support a more powerful diagrammatic calculus than bare symmetric monoidal categories, so this work may make it easier to design and reason about learners compositionally.The operation that dualises a learner does not appear to have been previously noticed and there is much to investigate.As asked by David Spivak, what is the dual of a neuron?

Authors

Institutions

Publication Details

Journal
Electronic Proceedings in Theoretical Computer Science
Published
2025-09-22
DOI
https://doi.org/10.4204/eptcs.429.3
Primary Topic
Mathematical and Theoretical Analysis
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
article

Learners are Almost Free Compact Closed

Mary S. Riley
Electronic Proceedings in Theoretical Computer Science
Mathematical and Theoretical Analysis
article

Learners are Almost Free Compact Closed

Mary S. Riley
article en

Abstract

The category of learners has a pleasant symmetric formulation when the morphisms are considered up to a coarser equivalence than the one originally described in the paper "Backprop as Functor".A quotient of this modified category gives a new construction of the free compact closed category on a symmetric monoidal category.In this note I expand on an observation made at the end of [15] and further discussed online [16].If the definition of a learner [7] is tweaked slightly so that learners are considered identical under a coarser equivalence relation than before, a striking symmetry in the definition is revealed.This new category comes quite close to being compact closed.There is a strict symmetric monoidal involution and each object has a cup and cap morphism connecting it to its "dual" under this involution.However, the snake equations fail and so these are not true duals.Quotienting the category to force the equations to hold results in a compact closed category, and we end the note by showing that it is in fact the free compact closed category on the starting symmetric monoidal category.Compact closed categories support a more powerful diagrammatic calculus than bare symmetric monoidal categories, so this work may make it easier to design and reason about learners compositionally.The operation that dualises a learner does not appear to have been previously noticed and there is much to investigate.As asked by David Spivak, what is the dual of a neuron?

Electronic Proceedings in Theoretical Computer ScienceVol. 429
New York University Abu Dhabi (AE)
Openalex Percentile: Top 68%
Mathematical and Theoretical Analysis
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.