Abstract:Kernel methods are of current interest in quantum machine learning due to similarities with quantum computing in how they process information in high-dimensional feature (Hilbert) spaces. Kernels are believed to offer particular advantages when they cannot be computed classically, so a kernel matrix with indisputably nonclassical elements is desirable provided it can be generated efficiently in a particular physical machine. Kerr nonlinearities, known to be a route to universal continuous variable (CV) quantum computation, may be able to play this role for quantum machine learning. We propose a quantum hardware kernel implementation scheme based on superconducting quantum circuits. The scheme does not use qubits or quantum circuits but rather exploits the analogue features of Kerr coupled modes. Our approach is more akin to the growing number of analog machine learning schemes based on sampling quantum probabilities directly in an engineered device by stochastic quantum control.

What problem does this paper attempt to address?

The paper proposes a novel machine learning model based on quantum systems, called Kerr Kernel Quantum Learning Machine (KKQLM). It utilizes the nonlinear Kerr effect in quantum mechanics to create and process high-dimensional representations of data without computing the kernel matrix using classical methods. The core of the research is the design of a quantum hardware implementation based on superconducting quantum circuits, which does not rely on the traditional qubit-based architecture of quantum computers. In the context of machine learning tasks, especially binary classification problems, the protocol proposed in the paper directly generates the kernel from physical device measurements instead of computing the kernel matrix beforehand. This allows the system to leverage the randomness and non-classical characteristics inherent in quantum systems, such as the negative Wigner function of quantum states, to perform learning tasks. The paper demonstrates how to generate quantum encoding states using Kerr nonlinearity and construct decision functions by measuring specific operators on these states, such as the parity shift. Specifically, they propose two protocols: a sequential protocol where the quantum system continuously updates its state based on the input data stream, and a parallel protocol that constructs the input required for the Support Vector Machine algorithm by measuring the similarity (i.e., kernel) between quantum states. The paper illustrates this concept with examples of single-mode and two-mode scenarios, highlighting the capability of quantum kernels in handling non-classical quantum states and how this kernel is realized in continuous-variable quantum systems through Kerr nonlinearity. This approach is of great significance for developing new quantum-enhanced machine learning algorithms as it provides a different approach from traditional qubit-based algorithms.

A Kerr kernel quantum learning machine

Quantum Kerr Learning

Enhancing Quantum Machine Learning: The Power of Non-Linear Optical Reproducing Kernels

Quantum Machine Learning: Quantum Kernel Methods

Quantum Kernel Machine Learning With Continuous Variables

Quantum kernels with Gaussian state encoding for machine learning

Quantum kernels with squeezed-state encoding for machine learning

Distributed and Secure Kernel-Based Quantum Machine Learning

Quantum-Classical Multiple Kernel Learning

Quantum machine learning beyond kernel methods

Continuous-variable quantum kernel method on a programmable photonic quantum processor

CDQKL: Consensus-Based Distributed Quantum Kernel Learning

Hybrid model of the kernel method for quantum computers

Experimental kernel-based quantum machine learning in finite feature space

Quantum Phase Recognition Via Quantum Kernel Methods.

Parameterized Quantum Circuits with Quantum Kernels for Machine Learning: A Hybrid Quantum-Classical Approach

Covariant quantum kernels for data with group structure

Application of quantum machine learning using the quantum kernel algorithm on high energy physics analysis at the LHC

Automatic and effective discovery of quantum kernels

Quantum Machine Learning in Feature Hilbert Spaces