Development of AI-Based Optimization Technology for Efficient Quantum Computing; Paper Presented at ICDM 2026 / Research Team Led by Professor Kim Jangho (Department of Artificial Intelligence)

  • 26.11.18 / 홍유민

Lee Hye-eun, a Ph.D. candidate in the Department of AI and Software Engineering at Kookmin University, working under the guidance of Professor Kim Jangho’s research team in the School of Artificial Intelligence, has developed a new AI-based initialization technique that enhances the learning efficiency of quantum algorithms. The study was accepted for presentation at the IEEE International Conference on Data Mining (ICDM 2026), a prestigious international conference in the field of data mining, and was presented on November 13 in Shenyang, China.
​
A variational quantum algorithm (VQA) is a technique that solves complex optimization problems by iteratively adjusting the parameters of a quantum circuit. However, since the optimization speed and final results can vary significantly depending on the initial parameters used to start training, setting appropriate initial values has long been considered a major challenge.
​
To address this issue, the research team proposed a technique called “Task-Conditioned Distributional Initialization,” which simultaneously learns the initial parameters and the search range based on the characteristics of the problem. While existing methods focused on predicting a single initial value, the proposed technique utilizes a hypernetwork—an artificial intelligence model—to determine both the appropriate initial values and the search range during the learning process for each specific problem. It is also designed to enhance the stability of the learning process by leveraging information from similar problems.
​
Experimental results confirmed faster convergence performance compared to existing initialization methods for the XYZ Hamiltonian problem—used in quantum physics simulations—and the MaxCut combinatorial optimization problem. Specifically, for the 10-qubit XYZ problem, the number of optimization iterations required for convergence decreased by approximately 83% compared to existing point-based learning methods, and the MaxCut problem also showed a reduction of up to approximately 73%. The accuracy of the final solution was also comparable to that of existing methods or improved under certain conditions.
​
This research is significant in that it can enhance computational efficiency in environments where similar types of optimization problems must be solved repeatedly by efficiently setting the initial values of quantum algorithms on a problem-by-problem basis. It is expected to be applied to various quantum computing applications, such as quantum physics simulations and combinatorial optimization, in the future.
 
Meanwhile, the paper titled “Learning Task-Conditioned Distributional Initializers for Variational Quantum Algorithms” was authored by Ph.D. candidate Hye-Eun Lee of Kookmin University as the first author and Professor  Kim Jangho as the corresponding author.

This content is translated from Korean to English using the AI translation service DeepL and may contain translation errors such as jargon/pronouns.

If you find any, please send your feedback to kookminpr@kookmin.ac.kr so we can correct them.

 

View original article [click]

Development of AI-Based Optimization Technology for Efficient Quantum Computing; Paper Presented at ICDM 2026 / Research Team Led by Professor Kim Jangho (Department of Artificial Intelligence)

Lee Hye-eun, a Ph.D. candidate in the Department of AI and Software Engineering at Kookmin University, working under the guidance of Professor Kim Jangho’s research team in the School of Artificial Intelligence, has developed a new AI-based initialization technique that enhances the learning efficiency of quantum algorithms. The study was accepted for presentation at the IEEE International Conference on Data Mining (ICDM 2026), a prestigious international conference in the field of data mining, and was presented on November 13 in Shenyang, China.
​
A variational quantum algorithm (VQA) is a technique that solves complex optimization problems by iteratively adjusting the parameters of a quantum circuit. However, since the optimization speed and final results can vary significantly depending on the initial parameters used to start training, setting appropriate initial values has long been considered a major challenge.
​
To address this issue, the research team proposed a technique called “Task-Conditioned Distributional Initialization,” which simultaneously learns the initial parameters and the search range based on the characteristics of the problem. While existing methods focused on predicting a single initial value, the proposed technique utilizes a hypernetwork—an artificial intelligence model—to determine both the appropriate initial values and the search range during the learning process for each specific problem. It is also designed to enhance the stability of the learning process by leveraging information from similar problems.
​
Experimental results confirmed faster convergence performance compared to existing initialization methods for the XYZ Hamiltonian problem—used in quantum physics simulations—and the MaxCut combinatorial optimization problem. Specifically, for the 10-qubit XYZ problem, the number of optimization iterations required for convergence decreased by approximately 83% compared to existing point-based learning methods, and the MaxCut problem also showed a reduction of up to approximately 73%. The accuracy of the final solution was also comparable to that of existing methods or improved under certain conditions.
​
This research is significant in that it can enhance computational efficiency in environments where similar types of optimization problems must be solved repeatedly by efficiently setting the initial values of quantum algorithms on a problem-by-problem basis. It is expected to be applied to various quantum computing applications, such as quantum physics simulations and combinatorial optimization, in the future.
 
Meanwhile, the paper titled “Learning Task-Conditioned Distributional Initializers for Variational Quantum Algorithms” was authored by Ph.D. candidate Hye-Eun Lee of Kookmin University as the first author and Professor  Kim Jangho as the corresponding author.

This content is translated from Korean to English using the AI translation service DeepL and may contain translation errors such as jargon/pronouns.

If you find any, please send your feedback to kookminpr@kookmin.ac.kr so we can correct them.

 

View original article [click]

TOP