[Paper Review] Quantum generative adversarial network for generating discrete data

Haozhen Situ, Zhimin He+2 more|arXiv (Cornell University)|Jul 3, 2018

Quantum Computing Algorithms and Architecture17 citations

TL;DR

This paper proposes a hybrid quantum-classical generative adversarial network (GAN) that generates discrete classical data—such as text—using a parameterized quantum circuit as the generator and a classical neural network as the discriminator. The method leverages only native one- and two-qubit gates available on current quantum hardware, avoids data encoding/output bottlenecks, and implicitly encodes probability distributions into quantum states, enabling efficient discrete data generation without the vanishing gradient issues common in classical GANs.

ABSTRACT

Quantum machine learning has recently attracted much attention from the community of quantum computing. In this paper, we explore the ability of generative adversarial networks (GANs) based on quantum computing. More specifically, we propose a quantum GAN for generating classical discrete distribution, which has a classical-quantum hybrid architecture and is composed of a parameterized quantum circuit as the generator and a classical neural network as the discriminator. The parameterized quantum circuit only consists of simple one-qubit rotation gates and two-qubit controlled-phase gates that are available in current quantum devices. Our scheme has the following characteristics and potential advantages: (i) It is intrinsically capable of generating discrete data (e.g., text data), while classical GANs are clumsy for this task due to the vanishing gradient problem. (ii) Our scheme avoids the input/output bottlenecks embarrassing most of the existing quantum learning algorithms that either require to encode the classical input data into quantum states, or output a quantum state corresponding to the solution instead of giving the solution itself, which inevitably compromises the speedup of the quantum algorithm. (iii) The probability distribution implicitly given by data samples can be loaded into a quantum state, which may be useful for some further applications.

Motivation & Objective

• To address the challenge of generating discrete data, such as text, using quantum machine learning models.
• To overcome the vanishing gradient problem that hinders classical GANs when generating discrete distributions.
• To eliminate input and output bottlenecks common in quantum machine learning by avoiding full quantum state encoding and outputting classical data directly.
• To enable the implicit encoding of data probability distributions into quantum states for downstream quantum applications.

Proposed method

• The generator is a parameterized quantum circuit composed of one-qubit rotation gates and two-qubit controlled-phase gates, realizable on current noisy intermediate-scale quantum (NISQ) devices.
• The discriminator is a classical feedforward neural network that evaluates the realism of generated data samples.
• The generator produces a quantum state that encodes the probability distribution of discrete data, which is then measured to yield classical discrete outputs.
• The training process alternates between optimizing the generator to produce more realistic data and the discriminator to better distinguish real from generated data.
• The quantum circuit's output is measured in the computational basis to obtain discrete classical data, bypassing the need for full quantum state tomography.
• The method implicitly stores the data distribution in a quantum state, which may be useful for further quantum processing.

Experimental results

Research questions

• RQ1 Can a quantum generator with only native gates produce realistic discrete data distributions without suffering from vanishing gradients?
• RQ2 How can quantum GANs avoid the input and output bottlenecks that limit the speedup of existing quantum machine learning algorithms?
• RQ3 Can the probability distribution of discrete data be implicitly encoded into a quantum state through measurement of a parameterized quantum circuit?
• RQ4 What is the feasibility of generating classical discrete data (e.g., text) using a hybrid quantum-classical GAN architecture on near-term quantum hardware?

Key findings

• The proposed quantum GAN architecture is intrinsically capable of generating discrete data, overcoming the vanishing gradient problem that plagues classical GANs in this setting.
• The method avoids the input and output bottlenecks common in quantum machine learning by directly outputting classical discrete data instead of requiring quantum state output.
• The probability distribution of the generated data is implicitly encoded in the quantum state of the generator, enabling potential use in subsequent quantum algorithms.
• The use of only one- and two-qubit gates ensures compatibility with current noisy intermediate-scale quantum (NISQ) devices, enhancing practical feasibility.