Can Reverse Nearest Neighbors Perceive Unknowns?

Abstract

A novel open set classifier is presented in this work, where the neighborhood of a test instance is determined using the principles of Reverse k-nearest neighbors (RkNN). The RkNN count of an instance can have any non-negative value less or equal to the size of the training set. While dealing with an open dataset, consisting of known and unknown classes, the zero count can provide a possible solution for detecting the unknown class. Positive RkNN count along with the nearest RkNN distance information are used to determine the known class classifications. Experiments are carried out on ten real world datasets, with various openness values on five state-of-the-art open set learners and the proposed scheme. Their performance is measured on three evaluating metrics namely accuracy, average F 1 over known and unknown classes, and Known class F 1. Empirical results indicate comparable to superior performance delivered by the proposed method over the state-of-the-art approaches on all but one dataset. INDEX TERMS Open set classification, unknown class detection, known class classification, reverse nearest neighborhood.

Multi‐label learning on principles of reverse k‐nearest neighbourhood

“…Sadhukhan and Palit (2019) proposed an oversampling scheme for multi‐label imbalanced datasets, where the space for oversampling is determined via reverse‐nearest neighbourhood. In a recent work by Sadhukhan (2020), a scheme is devised on principles of reverse nearest neighbourhood to tackle the problem of open set classification.…”

Open-set HRRP recognition method based on joint sparse representation

“…Aiming at the problem of target recognition in an open environment, Scheirer and other scholars defined the problem of open set recognition (OSR) and established the theoretical framework of OSR [3] . Under this framework, scholars have proposed a series of algorithms, which can be roughly divided into the following categories:1) Based on support vector machine (SVM), such as W-SVM [4] , PI-SVM , etc.；2) Based on sparse representation (SR), such as SR-OSR algorithm [6] ；3) Based on distance criteria, such as Nearest Non-Outlier [7] , reverse k-nearest neighbor classifier [8] , etc.；4) Based on deep neural network (DNN), such as deep open classifier [9] , category condition encoder [10] , etc.；5) Based on edge distribution, such as Extremum Value Machine algorithm [11] . Under the open set condition, the OSR method has achieved a good recognition effect, but the relevant achievements are mainly concentrated in the field of computer vision.…”