An online nearest neighbor query method for high-dimensional data based on hash learning

Abstract

The invention discloses an online high-dimensional data nearest neighbor query method based on hash learning, Firstly, the loss function is designed according to the similarity or dissimilarity of samples and the range of the loss function is enlarged, Then, a new objective function is proposed based on the principle that the hash model needs to keep the history information and minimize the loss of the current data. By analyzing the convergence of the online hash algorithm, the optimal value of the objective function is found. On this basis, for the data points to be queried, the closest datapoints can be quickly queried, the average accuracy results converge steadily, and the update of hash function in the iterative learning process is greatly reduced.

Description

technical field [0001] The invention relates to an online nearest neighbor query method, in particular to an online high-dimensional data nearest neighbor query method based on hash learning. Background technique [0002] Nearest Neighbor Search is an important research direction in the field of information retrieval, and it is widely used in image retrieval and data mining. The commonly used technologies for nearest neighbor query mainly include tree-based and hash-based methods. But when the data dimension becomes larger, the efficiency of tree-based neighbor retrieval will be greatly limited. The hash-based method compresses the original data into a low-dimensional binary code through a hash function, and then sorts and retrieves it under the Hamming distance, so this method has the advantages of being fast, efficient and dimensionally insensitive. Currently, the most researched hashing method is a batch processing method that uniformly trains all data. This method cann...

AI Technical Summary

Problems solved by technology

However, the update iteration process is frequent when recalculating the hash code, so that the computational overhead demand is too large as the data increases

Moreover, the hash model of the above method still has the problems of fast update frequency of the hash function and weak stability of the hash model in the online iterative learning process.

The reasons are: (1) design the loss function, and set the similar and dissimilar samples as a unified threshold on the entire data set; (2) only update the hash function according to the difference between two adjacent projection vectors as small as possible, which cannot guarantee the model stability

Images