Quantum generative adversarial network algorithm based on conditional constraints

Abstract

The invention discloses a quantum generative adversarial network algorithm based on conditional constraints. The algorithm comprises the following steps of: firstly, preparing a real sample, introducing proper conditional constraints according to the target of a generation task and the features of data, and enabling the real sample, the proper conditional constraints and the features of the data to jointly form a training data set of a network; secondly, designing an appropriate quantum circuit to perform quantum state coding on a classic training sample according to the numerical characteristics of the training data set; thirdly, designing parameterized quantum circuits for constructing a quantum generator G and a quantum discriminator D of a conditional generative adversarial network; and finally, cascading the quantum generator G and the quantum discriminator D, making a training strategy to carry out adversarial training, and carrying out measurement sampling on the trained quantumgenerator G to generate a data result which can fit the real sample and accords with the conditional constraints. According to the algorithm, the network can be effectively guided to generate data meeting specific requirements according to the setting of the conditional constraints, the controllability of a training process is improved, and the quality of the generated data is also improved.

Description

technical field [0001] The invention belongs to quantum machine learning algorithms, and in particular relates to a quantum generation confrontation network algorithm based on conditional constraints. Background technique [0002] In 2014, Goodfellow first proposed the Generative Adversarial Network (GAN), which combines the most important algorithm model in the field of artificial intelligence-the discriminative model and the generative model to obtain a new deep neural network algorithm framework. The main idea is that the generator learns the characteristics of the training sample data and generates false samples, while the discriminator judges the authenticity of the input samples, and through the alternate training of the generator and the discriminator, a better generator is obtained to generate simulated samples. fit the real sample data. Subsequently, this confrontational training idea was widely used to solve various generation and classification problems, especial...

AI Technical Summary

Problems solved by technology

However, similar to the classic GAN scheme, the quantum scheme does not need to assume data distribution in advance, and there is also th

Images