Quantum circuit-based graph attention mechanism implementation method

Abstract

The invention provides a graph attention mechanism implementation method based on a quantum circuit, and belongs to the technical field of quantum computing. According to the method, entanglement evolution is carried out on an initial node feature vector corresponding to a node i of a target image and an initial node feature vector corresponding to an adjacent node j in a quantum GAT, score calculation is carried out, and a processing feature vector corresponding to the node i is finally output, so that a quantum calculation mode of a graph attention mechanism is realized, and the method has the advantages of being high in practicability and the like. And m auxiliary quantum bits are added in the construction mode of the quantum circuit to increase the flexibility of the input feature vector dimension corresponding to the quantum state, and meanwhile, the nonlinear performance of the quantum circuit is enhanced, and compared with a traditional method, the method provided by the invention has the advantages that the quantity of parameters needing to be trained is greatly reduced, the storage space is also greatly reduced, and the training efficiency is improved. And the calculation speed can be increased by utilizing parallelism, so that the method has a wide application prospect.

Description

technical field [0001] The invention relates to the technical field of quantum computing, in particular to a method for realizing a graph attention mechanism based on quantum circuits. Background technique [0002] The graph neural network GNN applies deep learning to the graph structure (Graph), and the graph convolutional network GCN can perform convolution operations on the Graph. The success of GCN has made deep learning in the graph field flourish. With the deepening of research, the shortcomings of GCN have become more and more obvious: it relies on Laplacian matrix and cannot be directly used for directed graphs; model training depends on the entire graph structure , cannot be used in dynamic graphs; there is no way to assign different weights to neighbor nodes during convolution. Therefore, in 2018, the graph attention network GAT (Graph Attention Network) was proposed to solve the problems existing in GCN. [0003] Graph Convolutional Network (GCN) tells us that c...

AI Technical Summary

Problems solved by technology

[0004] 1. Unable to complete the induction task, that is, to deal with the dynamic graph problem;

[0005] 2. Dealing with the bottleneck of directed graphs, it is not easy to assign different learning weights to different adjacent nodes;

[0006] 3. GCN cannot assign different weights to each neighbor. GCN treats all neighbor nodes equally during convolution, and cannot assign different weights according to the importance of nodes.

However, in the case of a large amount of data, the amount of calculation required by the attention mechanism is still very large, and the number of parameters that need to be trained is astonishing, and the calculation speed is relatively slow.

Images