Be first to read the latest tech news, Industry Leader's Insights, and CIO interviews of medium and large enterprises exclusively from CIO Advisor APAC
Advances in quantum computing have raised expectations for the technology's future impact on real-world applications.
FREMONT, CA: Recent studies have shown that quantum computers can handle correlations and input complexities far beyond classical computers. The University of Innsbruck and The Institute for Quantum Computing (IQC) recently collaborated on a project. A classical computer with a quantum system combining reliability has been proposed. This confirms that quantum computer learning models can be more powerful for applications where specific information is essential but less data is required. Understanding which applications will benefit from quantum technologies advances at a breakneck pace. Data access fundamentally alters the question when evaluating quantum computers' capabilities to aid in machine learning.
Power of data
Occasionally, the concept of quantum advantage over a conventional computer is expressed in terms of computational complexity classes. Factoring large numbers and modelling quantum systems are examples of Bounded Quantum Polynomial (BQP) time problems that quantum computers are more adept at solving than classical systems. On the other hand, problems involving bounded probabilistic polynomials (BPPs) can be easily solved on conventional computers. The researchers demonstrated that learning algorithms equipped with data from quantum processes such as fusion or chemical reactions establish a new class of problems called BPP/Samp that can perform tasks more efficiently than traditional algorithms without data. It is a subclass of issues that can be solved efficiently with polynomial-size (P/poly) advice. This demonstrates that grasping the quantum advantage for various machine learning tasks also requires an examination of publicly available data.
Projected Quantum Kernel
Quantum machine learning is classified into two stages: quantum embedding the data and evaluating an applied function. Within a kernel learning framework, a procedure was developed for determining the possibility of advantage. The most significant and instructive of these was the development of a new geometric test. A geometric proof established that existing quantum kernels possessed a more accessible geometry that favoured memorization over comprehension.
This resulted in developing a projected quantum kernel, which converts the quantum embedding to a classical representation. While computing directly with a conventional computer is tricky, this representation has several practical advantages over remaining entirely in quantum space.
In short, the discovery of quantum kernels and the creation of a machine learning problem-solving guide have combined to introduce new work into machine learning. Quantifying the difference in prediction errors between quantum and classical machine learning models requires prediction error bounds. In practice, the comparison should be made between a suite of optimised classical machine learning models rather than an efficiency-defined difference to the nearest efficient classical machine learning model. A traditional machine learning method can ensure equivalent or higher performance prediction across function values and labels regardless of the geometric difference. When the geometry varies significantly, the quantum machine learning model yields better results.