Using relative von Neumann and Shannon entropies for feature fusion

Abstract

This paper develops a new quantum inspired feature fusion method based on relative von Neumann entropy. The motivation is to more effectively reduce data redundancy and further improve the completeness and conciseness of the existing feature data. Regarding this, we quantise the source dataset into the collection of basic quantum states and constructed a weighted linking network to calculate the relative von Neumann entropies between feature samples. Thus, the detection and fusion of the duplicate feature samples in a subset is turned to the computation of the average relative von Neumann entropy and the measurement probabilities of the quantised feature samples. In parallel, the classical feature fusion method based on relative Shannon entropy is also proposed following similar idea. The experimental results show that the proposed feature fusion methods perform better in their completeness, conciseness, and stability.

KEYWORDS:

• Duplicate detection
• feature fusion
• relative von Neumann entropy
• relative Shannon entropy
• quantum inspired method

Additional information

Funding

This work was supported by the Zhejiang Provincial Natural Science Foundation of China [grant number LY18F020013] and the National Natural Science Foundation of China [grant number 61703127].