AI training costs cut to one-sixtieth ... Yonsei University develops ‘customized quantum circuit’ technology

[Financial News] A quantum computing technology that can dramatically reduce the time and computational costs involved in AI training has been developed by South Korean researchers.
Yonsei University said on the 26th that Professor Park Yu-rang of its College of Medicine and a research team led by Sim Woo-seop of Yonsei University Health System (YUHS) had developed a ‘customized quantum circuit preselection algorithm’ to improve AI training efficiency. The findings were accepted by NeurIPS 2026, a leading global AI conference, in recognition of their academic merit.
Quantum AI learns complex data by connecting ‘qubits,’ the basic computational units of quantum computers. Until now, however, researchers could not determine which connection structure was best suited to learning from the data, so they had to run and evaluate numerous circuits directly. This led to massive computational costs and computational errors.
The research team devised a method to identify the most suitable circuit structure by analyzing only the characteristics of the data before training the AI itself. This sharply reduces complex calculations while securing both the ability to distinguish data and training stability.
In performance evaluations using images and disease data, among other things, the team’s algorithm selected the optimal circuit at about one-sixtieth the cost of existing high-performing comparison methods.
The team also completed validation using an actual quantum computer, an ‘IBM Quantum System One,’ installed at Yonsei University’s Songdo International Campus. Experiments on the actual device also demonstrated its practical potential: the system achieved superior performance with fewer operations than when circuits were arranged randomly or an existing simple linear structure was used.
Professor Park Yu-rang of Yonsei University said, “With quantum computers, the key is how to choose the operations the data requires, rather than blindly making circuits larger.” He added, “This research will provide an important foundation for understanding the learning potential of quantum AI and improving its efficiency.”  

[email protected] Kim Man-gi Reporter