%52%65%6c%69%61%62%69%6c%69%74%79
-
KAIST-KakaoBank Speeds Up 'Explainable AI' by 11 Times: "Boosts Financial AI Reliability
< (From left) Professor Jaesik Choi of the Kim Jaechul Graduate School of AI, Ph.D candidate Chanwoo Lee, Ph.D candidate Youngjin Park >
The research team led by Professor Jaesik Choi of KAIST's Kim Jaechul Graduate School of AI, in collaboration with KakaoBank Corp, announced that they have developed an accelerated explanation technology that can explain the basis of an Artificial Intelligence (AI) model's judgment in real-time. This research achievement significantly increases the practical applicability of Explainable Artificial Intelligence (hereinafter XAI) technology in fields requiring real-time decision-making, such as financial services, by achieving an average processing speed 8.5 times faster, and up to 11 times faster, than existing explanation algorithms for AI model predictions.
In the financial sector, a clear explanation for decisions made by AI systems is essential. Especially in services directly related to customer rights, such as loan screening and anomaly detection, regulatory demands to transparently present the basis for the AI model's judgment are increasingly stringent. However, conventional Explainable Artificial Intelligence (XAI) technologies required the repeated calculation of hundreds to thousands of baselines to generate accurate explanations, resulting in massive computational costs. This was a major factor limiting the application of XAI technology in real-time service environments.
To address this issue, Professor Choi's research team developed the 'ABSQR (Amortized Baseline Selection via Rank-Revealing QR)' framework for accelerating explanation algorithms. ABSQR noticed that the value function matrix generated during the AI model explanation process has a low-rank structure. It introduced a method to select only a critical few baselines from the hundreds available. This drastically reduced the computation complexity, which was previously proportional to the number of baselines, to be proportional only to the number of selected critical baselines, thereby maximizing computational efficiency while maintaining explanatory accuracy.
Specifically, ABSQR operates in two stages. The first stage systematically selects important baselines using Singular Value Decomposition (SVD) and Rank-Revealing QR decomposition techniques. Unlike existing random sampling methods, this is a deterministic selection method aimed at preserving information recovery, which guarantees the accuracy of the explanation while significantly reducing computation. The second stage introduces an amortized inference mechanism, which reuses the pre-calculated weights of the baselines through cluster-based search, allowing the system to provide an explanation for the model's prediction result in real-time service environments without repeatedly evaluating the model. The research team verified the superiority of ABSQR through experiments on various real-world datasets. Tests on standard datasets across five sectors—finance, marketing, and demographics—showed that ABSQR achieved an average processing speed 8.5 times faster than existing explanation algorithms that use all baselines, with a maximum speed improvement of over 11 times. Furthermore, the degradation of explanatory accuracy due to speed acceleration was minimized, maintaining up to 93.5% of the explanation accuracy compared to the baseline algorithm. This level is sufficient to meet the explanation quality required in real-world applications.
< ABSQR Framework Overview. (1) The baseline selection stage utilizes the low-rank structure of the value function matrix to select only a small number of key baselines, and (2) the accelerated search stage reuses the pre-calculated baseline weight coefficients based on clusters. This dramatically reduces the computation complexity, which was proportional to the number of baselines, to be proportional only to the number of selected key baselines. >
A KakaoBank official stated, "We will continue relentless research and development to enhance the reliability and convenience of financial services and introduce innovative financial technologies that customers can experience." Chanwoo Lee and Youngjin Park, co-first authors from KAIST, explained the significance of the research: "This methodology solves the crucial acceleration problem for real-time application in the financial sector, proving that it is possible to provide users with the reasons behind a learning model's decision in real-time." They added, "This research provides new insights into what constitutes unnecessary computation and the selection of important baselines in explanation algorithms, practically contributing to the improvement of explanation technology efficiency." This research, co-authored by PhD candidates Chanwoo Lee and Youngjin Park from the KAIST Kim Jaechul Graduate School of AI, and researchers Hyeongeun Lee and Yeeun Yoo from the KakaoBank Financial Technology Research Institute, was presented on November 12 at the 'CIKM 2025 (ACM International Conference on Information and Knowledge Management)', the world's highest-authority academic conference in the field of information and knowledge management. ※ Paper Title: Amortized Baseline Selection via Rank-Revealing QR for Efficient Model Explanation
※ Author Information:
※ Author Information: DOI: https://doi.org/10.1145/3746252.3761036
Co-First Authors: Chanwoo Lee (KAIST Kim Jaechul Graduate School of AI), Youngjin Park (KAIST Kim Jaechul Graduate School of AI), Hyeogeun Lee (KakaoBank), Yeeun Yoo (KakaoBank)
Co-Authors: Daehee Han (KakaoBank), Junho Choi (KAIST Kim Jaechul Graduate School of AI), Kunhyung Kim (KAIST Kim Jaechul Graduate School of AI)
Corresponding Authors: Nari Kim (KAIST Kim Jaechul Graduate School of AI), Jaesik Choi (KAIST Kim Jaechul Graduate School of AI)
Meanwhile, this research achievement was conducted through KakaoBank's industry-academia research project 'Advanced Research on Explainable Artificial Intelligence Algorithms in the Financial Sector' and the Ministry of Science and ICT/Institute for Information & Communications Technology Planning and Evaluation (IITP) supported project 'Development of Explainable Artificial Intelligence Technology Providing Explainability in a Plug-and-Play Manner and Verification of Explanation Provision for AI Systems.'
2025.12.12 View 3355 -
KAIST Develops Neuromorphic Semiconductor Chip that Learns and Corrects Itself
< Photo. The research team of the School of Electrical Engineering posed by the newly deveoped processor. (From center to the right) Professor Young-Gyu Yoon, Integrated Master's and Doctoral Program Students Seungjae Han and Hakcheon Jeong and Professor Shinhyun Choi >
- Professor Shinhyun Choi and Professor Young-Gyu Yoon’s Joint Research Team from the School of Electrical Engineering developed a computing chip that can learn, correct errors, and process AI tasks
- Equipping a computing chip with high-reliability memristor devices with self-error correction functions for real-time learning and image processing
Existing computer systems have separate data processing and storage devices, making them inefficient for processing complex data like AI. A KAIST research team has developed a memristor-based integrated system similar to the way our brain processes information. It is now ready for application in various devices including smart security cameras, allowing them to recognize suspicious activity immediately without having to rely on remote cloud servers, and medical devices with which it can help analyze health data in real time.
KAIST (President Kwang Hyung Lee) announced on the 17th of January that the joint research team of Professor Shinhyun Choi and Professor Young-Gyu Yoon of the School of Electrical Engineering has developed a next-generation neuromorphic semiconductor-based ultra-small computing chip that can learn and correct errors on its own.
< Figure 1. Scanning electron microscope (SEM) image of a computing chip equipped with a highly reliable selector-less 32×32 memristor crossbar array (left). Hardware system developed for real-time artificial intelligence implementation (right). >
What is special about this computing chip is that it can learn and correct errors that occur due to non-ideal characteristics that were difficult to solve in existing neuromorphic devices. For example, when processing a video stream, the chip learns to automatically separate a moving object from the background, and it becomes better at this task over time.
This self-learning ability has been proven by achieving accuracy comparable to ideal computer simulations in real-time image processing. The research team's main achievement is that it has completed a system that is both reliable and practical, beyond the development of brain-like components.
The research team has developed the world's first memristor-based integrated system that can adapt to immediate environmental changes, and has presented an innovative solution that overcomes the limitations of existing technology.
< Figure 2. Background and foreground separation results of an image containing non-ideal characteristics of memristor devices (left). Real-time image separation results through on-device learning using the memristor computing chip developed by our research team (right). >
At the heart of this innovation is a next-generation semiconductor device called a memristor*. The variable resistance characteristics of this device can replace the role of synapses in neural networks, and by utilizing it, data storage and computation can be performed simultaneously, just like our brain cells.
*Memristor: A compound word of memory and resistor, next-generation electrical device whose resistance value is determined by the amount and direction of charge that has flowed between the two terminals in the past.
The research team designed a highly reliable memristor that can precisely control resistance changes and developed an efficient system that excludes complex compensation processes through self-learning. This study is significant in that it experimentally verified the commercialization possibility of a next-generation neuromorphic semiconductor-based integrated system that supports real-time learning and inference.
This technology will revolutionize the way artificial intelligence is used in everyday devices, allowing AI tasks to be processed locally without relying on remote cloud servers, making them faster, more privacy-protected, and more energy-efficient.
“This system is like a smart workspace where everything is within arm’s reach instead of having to go back and forth between desks and file cabinets,” explained KAIST researchers Hakcheon Jeong and Seungjae Han, who led the development of this technology. “This is similar to the way our brain processes information, where everything is processed efficiently at once at one spot.”
The research was conducted with Hakcheon Jeong and Seungjae Han, the students of Integrated Master's and Doctoral Program at KAIST School of Electrical Engineering being the co-first authors, the results of which was published online in the international academic journal, Nature Electronics, on January 8, 2025.
*Paper title: Self-supervised video processing with self-calibration on an analogue computing platform based on a selector-less memristor array ( https://doi.org/10.1038/s41928-024-01318-6 )
This research was supported by the Next-Generation Intelligent Semiconductor Technology Development Project, Excellent New Researcher Project and PIM AI Semiconductor Core Technology Development Project of the National Research Foundation of Korea, and the Electronics and Telecommunications Research Institute Research and Development Support Project of the Institute of Information & communications Technology Planning & Evaluation.
2025.01.17 View 18294 -
KAIST and Hyundai Motors Collaborate to Develop Ultra-Fast Hydrogen Leak Detection within 0.6 Seconds
Recently, as the spread of eco-friendly hydrogen cars increases, the importance of hydrogen sensors is also on the rise. In particular, achieving technology to detect hydrogen leaks within one second remains a challenging task. Accordingly, the development of the world's first hydrogen sensor that meets the performance standards of the U.S. Department of Energy has become a hot topic.
A team at KAIST led by Dr. Min-Seung Jo from Professor Jun-Bo Yoon's team in the Department of Electrical and Electronic Engineering has successfully achieved all of its desired performance indicators, meeting globally recognized standards through collaboration with the Electromagnetic Energy Materials Research Team at Hyundai Motor Company's Basic Materials Research Center and Professor Min-Ho Seo of Pusan National University. On January 10th, the research group announced that the world's first hydrogen sensor with a speed of less than 0.6 seconds had been developed.
In order to secure faster and more stable hydrogen detection technology than existing commercialized hydrogen sensors, the KAIST team began developing a next-generation hydrogen sensor in 2021 together with Hyundai Motor Company, and succeeded after two years of development.
< Figure 1. (Left) The conceptual drawing of the structure of the coplanar heater-integrated hydrogen sensor. Pd nanowire is stably suspended in the air even with its thickness of 20 nm. (Right) A graph of hydrogen sensor performance operating within 0.6 seconds for hydrogen at a concentration of 0.1 to 4% >
Existing hydrogen sensor research has mainly focused on sensing materials, such as catalytic treatments or the alloying of palladium (Pd) materials, which are widely used in hydrogen sensors. Although these studies showed excellent performance with certain performance indicators, they did not meet all of the desired performance indicators and commercialization was limited due to the difficulty of batch processing.
To overcome this, the research team developed a sensor that satisfied all of the performance indicators by combining independent micro/nano structure design and process technology based on pure palladium materials. In addition, considering future mass production, pure metal materials with fewer material restrictions were used rather than synthetic materials, and a next-generation hydrogen sensor was developed that can be mass-produced based on a semiconductor batch process.
The developed device is a differential coplanar device in which the heater and sensing materials are integrated side by side on the same plane to overcome the uneven temperature distribution of existing gas sensors, which have a structure where the heater, insulating layer, and sensing materials are stacked vertically. The palladium nanomaterial, which is a sensing material, has a completely floating structure and is exposed to air from beneath, maximizing the reaction area with a gas to ensure a fast reaction speed. In addition, the palladium sensing material operates at a uniform temperature throughout the entire area, and the research team was able to secure a fast operation speed, wide sensing concentration, and temperature/humidity insensitivity by accurately controlling temperature-sensitive sensing performance.
< Figure 2. Electron microscopy of the coplanar heater-integrated hydrogen sensor (left) Photo of the entire device (top right) Pd nanowire suspended in the air (bottom right) Cross section of Pd nanowire >
The research team packaged the fabricated device with a Bluetooth module to create an integrated module that wirelessly detects hydrogen leaks within one second and then verified its performance. Unlike existing high-performance optical hydrogen sensors, this one is highly portable and can be used in a variety of applications where hydrogen energy is used.
Dr. Min-Seung Jo, who led the research, said, “The results of this research are of significant value as they not only operate at high speeds by exceeding the performance limits of existing hydrogen sensors, but also secure the reliability and stability necessary for actual use, and can be used in various places such as automobiles, hydrogen charging stations, and homes.” He also revealed his future plans, saying, “Through the commercialization of this hydrogen sensor technology, I would like to contribute to advancing the safe and eco-friendly use of hydrogen energy.”
< Figure 3. (Left) Real-time hydrogen detection results from the coplanar heater-integrated hydrogen sensor integrated and packaged in wireless communication and an app for mobile phone. (Middle) LED blinking cycle control in accordance with the hydrogen concentration level. (Right) Results of performance confirmation of the detection within 1 second in a real-time hydrogen leak demo >
The research team is currently working with Hyundai Motor Company to manufacture the device on a wafer scale and then mount it on a vehicle module to further verify detection and durability performance.
This research, conducted by Dr. Min-Seung Jo as the first author, has three patent applications filed in the U.S. and Korea, and was published in the renowned international academic journal 'ACS Nano'. (Paper title: Ultrafast (∼0.6 s), Robust, and Highly Linear Hydrogen Detection up to 10% Using Fully Suspended Pure Pd Nanowire). (Impact Factor: 18.087). ( https://pubs.acs.org/doi/10.1021/acsnano.3c06806?fig=fig1&ref=pdf )
The research was conducted through support from the National Research Foundation of Korea's Nano and Materials Technology Development Project and support and joint development efforts from Hyundai Motor Company's Basic Materials Research Center.
2024.01.25 View 12117 -
Prof. Seong Publishes English Book on Reliability in Digital Control Systems
Prof. Poong-Hyun Seong of Department of Nuclear and Quantum Engineering has recently published an English-language book on reliability and risk issues in large scale safety-critical digital control systems used in complex facilities such as nuclear power plants.
The book entitled “Reliability and Risk Issues in Large Scale Safety-critical Digital Control Systems” is a result of Prof. Seong’s collaboration with some KAIST graduates who used to be under his guidance. The 303-page publication has been published by Springer, one of the world’s leading publishers of academic journals, as part of the Springer Series in Reliability Engineering.
The book consists of four parts; part I deals with issues related to hardware, part II software, part III human factors and finally the last part integrated systems. It can be purchased through some on-line book stores such as Amazon.com.
Prof. Seong served as an editor-in-chief for Nuclear Engineering and Technology (NET), an international journal of Korean Nuclear Society (KNS), from 2003 to 2008. He also worked as a chair of the Human Factors Division (HFD) of American Nuclear Society (ANS) from 2006 to 2007. Prof. Seong is now a commissioner of Korea Nuclear Safety Commission which is the nation’s highest committee on Nuclear Safety.
2008.12.26 View 25411