Professors Jung Kee-chul (left) and Oh Kyoung-su of Soongsil University hold up copies of their 2004 study "Implementation of Neural Networks using," which is considered to have contributed to advancement of modern artificial intelligence by finding application of graphics processing unit in faster artificial neural network. (Yonhap)
Professors Jung Kee-chul (left) and Oh Kyoung-su of Soongsil University hold up copies of their 2004 study "Implementation of Neural Networks using," which is considered to have contributed to advancement of modern artificial intelligence by finding application of graphics processing unit in faster artificial neural network. (Yonhap)

Graphics processing unit acceleration, deemed essential for modern artificial intelligence training, can find its roots in a 2004 study led by two Soongsil University professors.

Professors Oh Kyoung-su and Jung Kee-chul of the university's Global School of Media were recently interviewed by local Yonhap News Agency about how their study provided the basis for the deep learning models applied widely to AI today. Their paper precedes Nvidia's 2007 release of the computer unified device architecture, which allows the use of GPUs for accelerated general-purpose processing.

"The specific method (of our study) is different because it was before CUDA, but the basic principle of operation is the same as the deep learning of today," the two scientists were quoted as saying.

"Implementation of Neural Networks using GPU" was published in 2004 in the international journal Pattern Recognition, presenting a new use of common graphics hardware for a faster artificial neural network. The preliminary results showed about thirty times faster performance using the ATI RADEON 9800 XT board.

The neural network using a GPU is cheaper and has many advantages such as reducing overheads. The findings can be applied when processing massive data, such as in the fields of video processing and pattern recognition, the researchers noted.

The accomplishment was mostly unknown before Google Chief Scientist Jeff Dean mentioned it in an essay published by the American Academy of Arts and Sciences.

"In 2004, computer scientists Kyoung-Su Oh and Keechul Jung showed a nearly twenty-fold improvement for a neural network algorithm using a GPU," Dean wrote in 2022, mentioning it as a milestone in the use of GPUs for implementing deep learning algorithms.

The two scholars were unable to pursue a follow-up study due to a lack of attention toward AI and neural networks at the time, and they also had difficulty securing funds for research due to their being in relatively early stages of their careers. Jung said he had no idea that the GPU-based neural network would be as big as it is today.

"I should've consistently pursued one topic. It's not the issue of the environment, I think I may have made a wrong choice," Jung said.

The pair did say that they are proud to have contributed to the advancement of AI.


minsikyoon@heraldcorp.com