Linear isomorphism testing of Boolean functions with small approximate spectral norm

Two Boolean functions f, g : F_2^{n} \to {-1, 1} are called linearly isomorphic if there exists an invertible matrix M \in F_2^{n\times n} such that f\circ M = g. Testing linear isomorphism is a generalization of, now classical in the context of property testing, isomorphism testing between Boolean...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Ghosh, Arijit, Kayal, Chandrima, Paraashar, Manaswi, Roy, Manmatha
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Two Boolean functions f, g : F_2^{n} \to {-1, 1} are called linearly isomorphic if there exists an invertible matrix M \in F_2^{n\times n} such that f\circ M = g. Testing linear isomorphism is a generalization of, now classical in the context of property testing, isomorphism testing between Boolean functions. Linear-invariance of Boolean functions has also been extensively studied in other areas like coding theory and cryptography, and mathematics in general. In this paper, we will study the following two variants of this problem: [1] [Communication complexity: ] Assume that Boolean functions f and g on F_2^{n} are given to Alice and Bob respectively, and the goal is to test linear isomorphism between f and g by exchanging a minimum amount of communication, measured in bits, between Alice and Bob. Our main result is an efficient two-party communication protocol for testing linear isomorphism in terms of the approximate spectral norm of the functions. We will crucially exploit the connection between approximate spectral norm and sign-approximating polynomials. [2] [Query complexity: ] If f: F_2^{n} \to { -1, 1 } is a known function and g : F_2^{n} \to { -1, 1 } be an unknown function with a query access. We study the query complexity of testing linear isomorphism between f and g in terms of the approximate spectral norm of f. As in the case of communication complexity, we will use properties of the approximate spectral norm.
DOI:10.48550/arxiv.2308.02662