A Categorical Foundation of Rough Sets
Yoshihiko Kakutani · 2025
Read A Categorical Foundation of Rough Sets by Yoshihiko Kakutani (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-04218.txt.
Introduction
Yoshihiko Kakutani gives rough set theory a category theory foundation in this 2025 paper. Rough sets approximate ordinary sets from above and below and are used widely in data mining. Adjunctions were known to sit underneath the approximations, but they alone do not characterize the theory.
Kakutani reformulates rough sets as adjunctions between preordered sets in full generality. The version stays closed under composition and can express attribute reduction and data insertion, and it even suggests decision rules for unseen data. Changing the answer set refines the theory with ease, leading toward fuzzy rough sets, and an enriched version covers those as well.
Catalog record
| CREATOR | Yoshihiko Kakutani |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 56 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-04218
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A Categorical Foundation of Rough Sets Yoshihiko Kakutani Abstract Rough sets are approximations of concrete sets. The theory of rough sets has been used widely for data-mining. While it is well-known that adjunctions are underlying in rough approximations, such adjunctions are not enough for characterization of rough sets. This paper provides a way to characterize rough sets in terms of category theory. We reformulate rough sets as adjunctions between preordered sets in a general way. Our formulation of rough sets can enjoy benefits of adjunctions and category theory. Especially, our characterization is closed under composition. We can also explain the notions of attribute reduction and data insertion in our theory. It is novel that our theory enables us to guess decision rules for unknown data. If we change the answer set, we can get a refinement of rough sets without any difficulty. Our refined rough sets lead rough fuzzy sets or more general approximations of functions. Moreover, our theory of rough sets can be generalized in the manner of enriched category theory. The derived enriched theory covers the usual theory of fuzzy rough sets. 1 Introduction The concept of rough sets was proposed by Pawlak [ 15 ] in order to deal with vagueness. While a crisp set has a clear boundary, a rough set has a vague boundary. Mathematically, a rough set consists of two kinds of approximations, an upper approximation and a lower one. The region between upper and lower approximations represents the boundary of a property. Rough sets have been studied constantly in the field of data-mining, and recently, much attention is paid to their practical applications. Both in a theoretical aspect and in a practical aspect, we can refer to kinds of extensions of rough sets [ 3 , 12 , 34 , 9 , 17 , 19 , 7 , 38 ] . Although rough approximations can be calculated directly, we formulate them with adjunctions in this work. An adjunction between preordered sets is often referred to as a Galois connection. Some Galois connections in rough set theory [ 6 , 4 ] were studied but are not enough for categorical reasoning of rough set theory. Our adjunctions are division of a known adjunction and explain how approximations are categorically constructed from original data. In rough set theory, data are given as a collection of objects. Each object has attributes and is classified into ‘yes’ or ‘no’. Our analysis is aimed at characterizing conditions on attributes for the classification. When objects x and y have the same attributes, we write x\equiv y . Clearly, {\equiv} is an equivalence relation. Let \mu be the set of all objects classified into yes. The lower approximation of \mu means the subset of {X}/{\equiv} consisting of classes surely in yes. On the other hand, the concept of the upper approximation of \mu is the subset possibly in yes. These approximations are the essential concept of a rough set. In our formulation, the lower approximation is characterized as the right adjoint of the trivial inclusion, and dually, the upper approximation is characterized as the left adjoint. In some axiomatizations [ 12 , 34 ] , an upper approximation is the dual of a lower approximation via a kind of negation operator. In our framework, the duality is represented as the symmetry between left and right adjoints. Given an equivalence relation on objects, there are two adjunctions, right and left adjoints, in our formalization. From the composite of these two adjunctions, a known Galois connection [ 6 ] can be derived. Our theory has some advantages over other category theoretic approaches. First, approximations are closed under composition in our theory. In other words, we have a lower/upper approximation correctly even if we approximate more than once. Second, the notions of attribute reduction and data insertion can be explained in a sophisticated way. Third, we can guess decision rules for unknown data automatically. While the usual rough set theory can be derived with a surjective functor in our theory, a non-surjective functor enables us to guess decision rules. We show some examples for the above points in the paper. Since our framework follows quite standard approach in category theory, we can generalize the theory easily. While a set of objects is usually discrete, we can consider a preorder on it. Though such a generalization may not be required by practical applications, it makes the theory clear and coherent. For another extension, we can replace the answer set of ‘yes’ and ‘no’ with a more complex structure. Moreover, we can apply enriched category theory to our framework. This enriched version covers the usual theory of fuzzy rough sets [ 3 ] . Fuzzy sets [ 37 ] have been studied widely as another approach to vagueness. A fuzzy set has an extension of a membership function with intermediate values. From the early days of the history of rough sets, similarities and differences between fuzzy sets and rough sets were discussed [ 16 , 33 ] . Fusion of rough sets and fuzzy sets is known as fuzzy rough sets [ 3 ] . Theories of fuzzy rough sets are usually provided algebraically. This paper gives another reasoning to fuzzy rough sets via enriched category theory. From the viewpoint of enriched category theory, a preordered set is a 2 -enriched category and a monotone function between preordered sets is a 2 -enriched functor, where 2 is the trivial lattice \{\bot,\top\} . Therefore, we replace 2 with a complete commutative residuated lattice L . Significance of residuated lattices in generalizations of fuzzy rough sets was already known [ 21 , 23 ] . If we take the interval between 0 and 1 as a residuated lattice, we can get the usual theory of fuzzy rough sets from our theory. More general structures like quantaloids [ 26 , 27 ] can be considered for enrichment. We do not focus on such generalization, but we expect that most part of this work can be generalized. Related Work While we are formalizing the theory in terms of category theory, most studies on rough sets are directed to characterize the theory as a set of algebraic axioms. Although the studies on Galois connections [ 6 , 4 ] are strongly related to our work, their aim is still an algebraic characterization. Relation of Galois connections and rough sets is studied also in formal concept analysis [ 29 , 4 , 30 , 35 ] . Such studies provides another direction of generalization of rough set theory. The intersection of formal concept analysis and our work is just the traditional rough set theory. The approach to generalize rough sets to fuzzy rough sets [ 20 , 31 , 21 , 23 , 13 ] is similar to our enrichment method. In those studies, a relation is replaced with a fuzzy relation over a residuated lattice. Some aspects of such generalization can be explained in terms of enriched category theory. Also fuzzy extensions [ 2 , 22 , 11 ] of formal concept analysis are studied algebraically. Studies on fuzzy Galois connections [ 1 , 32 ] have more direct relation to our enrichment of adjunctions. A complete residuated lattice is