%4a%61%65%73%69%6b%20%43%68%6f%69
-
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 3984 -
How Does AI Think? KAIST Achieves First Visualization of the Internal Structure Behind AI Decision-Making
<(From Left) Ph.D candidate Daehee Kwon, Ph.D candidate Sehyun lee, Professor Jaesik Choi>
Although deep learning–based image recognition technology is rapidly advancing, it still remains difficult to clearly explain the criteria AI uses internally to observe and judge images. In particular, technologies that analyze how large-scale models combine various concepts (e.g., cat ears, car wheels) to reach a conclusion have long been recognized as a major unsolved challenge.
KAIST (President Kwang Hyung Lee) announced on the 26th of November that Professor Jaesik Choi’s research team at the Kim Jaechul Graduate School of AI has developed a new explainable AI (XAI) technology that visualizes the concept-formation process inside a model at the level of circuits, enabling humans to understand the basis on which AI makes decisions.
The study is evaluated as a significant step forward that allows researchers to structurally examine “how AI thinks.”
Inside deep learning models, there exist basic computational units called neurons, which function similarly to those in the human brain. Neurons detect small features within an image—such as the shape of an ear, a specific color, or an outline—and compute a value (signal) that is transmitted to the next layer.
In contrast, a circuit refers to a structure in which multiple neurons are connected to jointly recognize a single meaning (concept). For example, to recognize the concept of cat ear, neurons detecting outline shapes, neurons detecting triangular forms, and neurons detecting fur-color patterns must activate in sequence, forming a functional unit (circuit).
Up until now, most explanation techniques have taken a neuron-centric approach based on the idea that “a specific neuron detects a specific concept.” However, in reality, deep learning models form concepts through cooperative circuit structures involving many neurons. Based on this observation, the KAIST research team proposed a technique that expands the unit of concept representation from “neuron → circuit.”
The research team’s newly developed technology, Granular Concept Circuits (GCC), is a novel method that analyzes and visualizes how an image-classification model internally forms concepts at the circuit level.
GCC automatically traces circuits by computing Neuron Sensitivity and Semantic Flow. Neuron Sensitivity indicates how strongly a neuron responds to a particular feature, while Semantic Flow measures how strongly that feature is passed on to the next concept. Using these metrics, the system can visualize, step-by-step, how basic features such as color and texture are assembled into higher-level concepts.
The team conducted experiments in which specific circuits were temporarily disabled (ablation). As a result, when the circuit responsible for a concept was deactivated, the AI’s predictions actually changed.
In other words, the experiment directly demonstrated that the corresponding circuit indeed performs the function of recognizing that concept.
This study is regarded as the first to reveal, at a fine-grained circuit level, the actual structural process by which concepts are formed inside complex deep learning models. Through this, the research suggests practical applicability across the entire explainable AI (XAI) domain—including strengthening transparency in AI decision-making, analyzing the causes of misclassification, detecting bias, improving model debugging and architecture, and enhancing safety and accountability.
The research team stated, “This technology shows the concept structures that AI forms internally in a way that humans can understand,” adding that “this study provides a scientific starting point for researching how AI thinks.”
Professor Jaesik Choi emphasized, “Unlike previous approaches that simplified complex models for explanation, this is the first approach to precisely interpret the model’s interior at the level of fine-grained circuits,” and added, “We demonstrated that the concepts learned by AI can be automatically traced and visualized.”
< Overview of the Conceptual Circuit Proposed by the Research Team >
This study, with Ph.D. candidates Dahee Kwon and Sehyun Lee from KAIST Kim Jaechul Graduate School of AI as co–first authors, was presented on October 21 at the International Conference on Computer Vision (ICCV).
Paper title: Granular Concept Circuits: Toward a Fine-Grained Circuit Discovery for Concept Representations
Paper link: https://openaccess.thecvf.com/content/ICCV2025/papers/Kwon_Granular_Concept_Circuits_Toward_a_Fine-Grained_Circuit_Discovery_for_Concept_ICCV_2025_paper.pdf
This research was supported by the Ministry of Science and ICT and the Institute for Information & Communications Technology Planning & Evaluation (IITP) under the “Development of Artificial Intelligence Technology for Personalized Plug-and-Play Explanation and Verification of Explanation” project, the AI Research Hub Project, and the KAIST AI Graduate School Program, and was carried out with support from the Defense Acquisition Program Administration (DAPA) and the Agency for Defense Development (ADD) at the KAIST Center for Applied Research in Artificial Intelligence.
2025.11.26 View 4827 -
KAIST Researchers Unveil an AI that Generates "Unexpectedly Original" Designs
< Photo 1. Professor Jaesik Choi, KAIST Kim Jaechul Graduate School of AI >
Recently, text-based image generation models can automatically create high-resolution, high-quality images solely from natural language descriptions. However, when a typical example like the Stable Diffusion model is given the text "creative," its ability to generate truly creative images remains limited. KAIST researchers have developed a technology that can enhance the creativity of text-based image generation models such as Stable Diffusion without additional training, allowing AI to draw creative chair designs that are far from ordinary.
Professor Jaesik Choi's research team at KAIST Kim Jaechul Graduate School of AI, in collaboration with NAVER AI Lab, developed this technology to enhance the creative generation of AI generative models without the need for additional training.
< Photo 2. Gayoung Lee, Researcher at NAVER AI Lab; Dahee Kwon, Ph.D. Candidate at KAIST Kim Jaechul Graduate School of AI; Jiyeon Han, Ph.D. Candidate at KAIST Kim Jaechul Graduate School of AI; Junho Kim, Researcher at NAVER AI Lab >
Professor Choi's research team developed a technology to enhance creative generation by amplifying the internal feature maps of text-based image generation models. They also discovered that shallow blocks within the model play a crucial role in creative generation. They confirmed that amplifying values in the high-frequency region after converting feature maps to the frequency domain can lead to noise or fragmented color patterns. Accordingly, the research team demonstrated that amplifying the low-frequency region of shallow blocks can effectively enhance creative generation.
Considering originality and usefulness as two key elements defining creativity, the research team proposed an algorithm that automatically selects the optimal amplification value for each block within the generative model.
Through the developed algorithm, appropriate amplification of the internal feature maps of a pre-trained Stable Diffusion model was able to enhance creative generation without additional classification data or training.
< Figure 1. Overview of the methodology researched by the development team. After converting the internal feature map of a pre-trained generative model into the frequency domain through Fast Fourier Transform, the low-frequency region of the feature map is amplified, then re-transformed into the feature space via Inverse Fast Fourier Transform to generate an image. >
The research team quantitatively proved, using various metrics, that their developed algorithm can generate images that are more novel than those from existing models, without significantly compromising utility.
In particular, they confirmed an increase in image diversity by mitigating the mode collapse problem that occurs in the SDXL-Turbo model, which was developed to significantly improve the image generation speed of the Stable Diffusion XL (SDXL) model. Furthermore, user studies showed that human evaluation also confirmed a significant improvement in novelty relative to utility compared to existing methods.
Jiyeon Han and Dahee Kwon, Ph.D. candidates at KAIST and co-first authors of the paper, stated, "This is the first methodology to enhance the creative generation of generative models without new training or fine-tuning. We have shown that the latent creativity within trained AI generative models can be enhanced through feature map manipulation."
They added, "This research makes it easy to generate creative images using only text from existing trained models. It is expected to provide new inspiration in various fields, such as creative product design, and contribute to the practical and useful application of AI models in the creative ecosystem."
< Figure 2. Application examples of the methodology researched by the development team. Various Stable Diffusion models generate novel images compared to existing generations while maintaining the meaning of the generated object. >
This research, co-authored by Jiyeon Han and Dahee Kwon, Ph.D. candidates at KAIST Kim Jaechul Graduate School of AI, was presented on June 16 at the International Conference on Computer Vision and Pattern Recognition (CVPR), an international academic conference.* Paper Title: Enhancing Creative Generation on Stable Diffusion-based Models* DOI: https://doi.org/10.48550/arXiv.2503.23538
This research was supported by the KAIST-NAVER Ultra-creative AI Research Center, the Innovation Growth Engine Project Explainable AI, the AI Research Hub Project, and research on flexible evolving AI technology development in line with increasingly strengthened ethical policies, all funded by the Ministry of Science and ICT through the Institute for Information & Communications Technology Promotion. It also received support from the KAIST AI Graduate School Program and was carried out at the KAIST Future Defense AI Specialized Research Center with support from the Defense Acquisition Program Administration and the Agency for Defense Development.
2025.06.20 View 8827