AI
-
'Team Atlanta', in which KAIST Professor Insu Yun research team participated, won the DARPA AI Cyber Challenge in the US, with a prize of 5.5 billion KRW
<Photo1. Group Photo of Team Atlanta>
Team Atlanta, led by Professor Insu Yun of the Department of Electrical and Electronic Engineering at KAIST and Tae-soo Kim, an executive from Samsung Research, along with researchers from POSTECH and Georgia Tech, won the final championship at the AI Cyber Challenge (AIxCC) hosted by the Defense Advanced Research Projects Agency (DARPA). The final was held at the world's largest hacking conference, DEF CON 33, in Las Vegas on August 8 (local time).
With this achievement, the team won a prize of $4 million (approximately 5.5 billion KRW), demonstrating the excellence of their AI-based autonomous cyber defense technology on the global stage.
<Photo2.Championship Commemorative:On the left and right are tournament officials. From the second person, Professor Tae-soo Kim(Samsung Research / Georgia Tech), Researcher Hyeong-seok Han (Samsung Research America), and Professor Insu Yun (KAIST)>
The AI Cyber Challenge is a two-year global competition co-hosted by DARPA and the Advanced Research Projects Agency for Health (ARPA-H). It challenges contestants to automatically analyze, detect, and fix software vulnerabilities using AI-based Cyber Reasoning Systems (CRS). The total prize money for the competition is $29.5 million, with the winning team receiving $4 million.
In the final, Team Atlanta scored a total of 392.76 points, a difference of over 170 points from the second-place team, Trail of Bits, securing a dominant victory. The CRS developed by Team Atlanta successfully and automatically detected various types of vulnerabilities and patched a significant number of them in real time.
Among the 7 finalist teams, an average of 77% of the 70 intentionally injected vulnerabilities were found, and 61% of them were patched. The teams also found 18 additional unknown vulnerabilities in real software, proving the potential of AI security technology.
All CRS technologies, including those of the winning team, will be provided as open-source and are expected to be used to strengthen the security of core infrastructure such as hospitals, water, and power systems.
<Photo3. Final Scoreboard: An overwhelming victory with over 170 points>
Professor Insu Yun of KAIST, a member of Team Atlanta, stated, "I am very happy to have achieved such a great result. This is a remarkable achievement that shows Korea's cyber security research has reached the highest level in the world, and it was meaningful to show the capabilities of Korean researchers on the world stage. I will continue to conduct research to protect the digital safety of the nation and global society through the fusion of AI and security technology."
KAIST President Kwang-hyung Lee stated, "This victory is another example that proves KAIST is a world-leading institution in the field of future cyber security and AI convergence. We will continue to provide full support to our researchers so they can compete and produce results on the world stage."
<Photo4. Results Announcement>
2025.08.10 View 29 -
Material Innovation Realized with Robotic Arms and AI, Without Human Researchers
<(From Left) M.S candidate Dongwoo Kim from KAIST, Ph.D candidate Hyun-Gi Lee from KAIST, Intern Yeham Kang from KAIST, M.S candidate Seongjae Bae from KAIST, Professor Dong-Hwa Seo from KAIST, (From top right, from left) Senior Researcher Inchul Park from POSCO Holdings, Senior Researcher Jung Woo Park, senior researcher from POSCO Holdings>
A joint research team from industry and academia in Korea has successfully developed an autonomous lab that uses AI and automation to create new cathode materials for secondary batteries. This system operates without human intervention, drastically reducing researcher labor and cutting the material discovery period by 93%.
* Autonomous Lab: A platform that autonomously designs, conducts, and analyzes experiments to find the optimal material.
KAIST (President Kwang Hyung Lee) announced on the 3rd of August that the research team led by Professor Dong-Hwa Seo of the Department of Materials Science and Engineering, in collaboration with the team of LIB Materials Research Center in Energy Materials R&D Laboratories at POSCO Holdings' POSCO N.EX.T Hub (Director Ki Soo Kim), built the lab to explore cathode materials using AI and automation technology.
Developing secondary battery cathode materials is a labor-intensive and time-consuming process for skilled researchers. It involves extensive exploration of various compositions and experimental variables through weighing, transporting, mixing, sintering*, and analyzing samples.
* Sintering: A process in which powder particles are heated to form a single solid mass through thermal activation.
The research team's autonomous lab combines an automated system with an AI model. The system handles all experimental steps—weighing, mixing, pelletizing, sintering, and analysis—without human interference. The AI model then interprets the data, learns from it, and selects the best candidates for the next experiment.
<Figure 1. Outline of the Anode Material Autonomous Exploration Laboratory>
To increase efficiency, the team designed the automation system with separate modules for each process, which are managed by a central robotic arm. This modular approach reduces the system's reliance on the robotic arm.
The team also significantly improved the synthesis speed by using a new high-speed sintering method, which is 50 times faster than the conventional low-speed method. This allows the autonomous lab to acquire 12 times more material data compared to traditional, researcher-led experiments.
<Figure 2. Synthesis of Cathode Material Using a High-Speed Sintering Device>
The vast amount of data collected is automatically interpreted by the AI model to extract information such as synthesized phases and impurity ratios. This data is systematically stored to create a high-quality database, which then serves as training data for an optimization AI model. This creates a closed-loop experimental system that recommends the next cathode composition and synthesis conditions for the automated system.
* Closed-loop experimental system: A system that independently performs all experimental processes without researcher intervention.
Operating this intelligent automation system 24 hours a day can secure more than 12 times the experimental data and shorten material discovery time by 93%. For a project requiring 500 experiments, the system can complete the work in about 6 days, whereas a traditional researcher-led approach would take 84 days.
During development, POSCO Holdings team managed the overall project planning, reviewed the platform design, and co-developed the partial module design and AI-based experimental model. The KAIST team, led by Professor Dong-hwa Seo, was responsible for the actual system implementation and operation, including platform design, module fabrication, algorithm creation, and system verification and improvement.
Professor Dong-Hwa Seo of KAIST stated that this system is a solution to the decrease in research personnel due to the low birth rate in Korea. He expects it will enhance global competitiveness by accelerating secondary battery material development through the acquisition of high-quality data.
<Figure 3. Exterior View (Side) of the Cathode Material Autonomous Exploration Laboratory>
POSCO N.EX.T Hub plans to apply an upgraded version of this autonomous lab to its own research facilities after 2026 to dramatically speed up next-generation secondary battery material development. They are planning further developments to enhance the system's stability and scalability, and hope this industry-academia collaboration will serve as a model for using innovative technology in real-world R&D.
<Figure 4. Exterior View (Front) of the Cathode Material Autonomous Exploration Laboratory>
The research was spearheaded by Ph.D. student Hyun-Gi Lee, along with master's students Seongjae Bae and Dongwoo Kim from Professor Dong-Hwa Seo’s lab at KAIST. Senior researchers Jung Woo Park and Inchul Park from LIB Materials Research Center of POSCO N.EX.T Hub's Energy Materials R&D Laboratories (Director Jeongjin Hong) also participated.
2025.08.06 View 32 -
KAIST Successfully Presents the Future of AI Transformation and Physical AI Strategy at the 1st National Strategic Technology Forum
<(Front row, fourth from the right) President Kwang Hyung Lee of KAIST, (back row, fifth from the right) Forum co-host Representative Hyung-Doo Choi, (back row, sixth from the left) Forum co-host Representative Han-Kyu Kim, along with ruling and opposition party members of the Science, ICT, Broadcasting, and Communications Committee and the Trade, Industry, Energy, SMEs, and Startups Committee, as well as Professors Hoe-Jun Yoo and Jung Kim from KAIST)>
KAIST (President Kwang Hyung Lee) announced on July that it had successfully held the “1st National Strategic Technology Forum” at the National Assembly Members' Office Building that day under the theme “The Future of Artificial Intelligence Transformation (AX): Physical AI.” This bipartisan policy forum aimed to discuss strategies for technology hegemony by leveraging Korea’s strengths in AI semiconductors and manufacturing.
The forum was hosted by KAIST and co-organized by Representative Hyung-Du Choi (People Power Party), the secretary of the National Assembly's Science, ICT, Broadcasting, and Communications Committee, and Representative Han-Kyu Kim (Democratic Party), a member of the Trade, Industry, Energy, SMEs, and Startups Committee. It marks the beginning of a five-part forum series, scheduled monthly through the rest of the year except for October.
The overarching theme, “Artificial Intelligence Transformation (AX),” was designed to address the structural changes reshaping industry, the economy, and society due to the spread of generative AI.
< KAIST President Kwang Hyung Lee delivering his remarks >
The first session focused on “Physical AI,” reflecting how AI innovation—sparked by the proliferation of large language models (LLMs)—is rapidly expanding into the physical realm through ultra-low-power, ultra-lightweight semiconductors. This includes applications in robotics, sensors, and edge devices. Physical AI refers to technologies that interact directly with the real world through AI integration with robotics, autonomous driving, and smart factories. It is drawing attention as a promising next-generation field where Korea can secure a strategic edge, given its strengths in semiconductors and manufacturing.
<Hoi-Jun Yoo, Dean of the KAIST Graduate School of AI Semiconductor>
Hoi-Jun Yoo, Dean of the KAIST Graduate School of AI Semiconductor, gave a presentation titled “The Second AI Innovation Enabled by Ultra-Low-Power AI Semiconductors and Lightweight AI Models,” covering semiconductor trends for implementing Physical AI, academic and industrial strategies for robotics and semiconductors, and Korea’s development direction for “K-Physical AI.”
<Professor Jung Kim, the head of KAIST’s Department of Mechanical and Aerospace Engineering>
Following that, Professor Jung Kim, the head of KAIST’s Department of Mechanical and Aerospace Engineering gave a talk on “Trends in Physical AI and Humanoid Robots,” predicting a new industrial paradigm shaped by AI-robot convergence. He presented global trends, Korea’s development trajectory, and survival strategies for humanoid robots that can supplement or replace human intellectual and physical functions.
During the open discussion that followed, participating lawmakers and experts engaged in in-depth conversations about the need for bipartisan strategies and collaboration.
Representative Hyung-Du Choi (People Power Party) stated, “Through this forum as a platform for public discourse, I will work to ensure that legislation and policy align with the direction of the science and technology field, and that necessary measures are taken promptly to strengthen national competitiveness.”
Representative Han-Kyu Kim (Democratic Party) emphasized, “As strategic planning in science and technology accelerates, it becomes more difficult to coordinate policies involving multiple ministries. Forums like this, which enable ongoing communication among stakeholders, are instrumental in finding effective solutions.”
KAIST President Kwang Hyung Lee remarked, “Although Korea is a latecomer in the generative AI field, we have a unique opportunity to gain strategic superiority in Physical AI, thanks to our technological capabilities in manufacturing, semiconductors, and robotics.” He added, “I hope lawmakers from both the ruling and opposition parties, along with experts, will come together regularly to devise practical policies and contribute to the advancement of Korea’s science and technology.”
<Poster of National Strategic Technology Forum>
This forum series aims to explore policy and institutional solutions to help Korea gain technological leadership in a global context where strategic technologies—such as AI, semiconductors, biotechnology, and energy—directly influence national security and economic sovereignty. Lawmakers from both the Science, ICT, Broadcasting, and Communications Committee and the Trade, Industry, Energy, SMEs, and Startups Committee will continue to participate, fostering bipartisan dialogue. The forums are coordinated by the KAIST Policy Research Institute for National Strategic Technologies.
2025.07.31 View 150 -
KAIST GESS Team Awarded Honorable Mention at 2025 Entrepreneurship Olympiad
<Photo: eaureco team at the final pitch>
The KAIST Global Entrepreneurship Summer School (GESS) winning team, eaureco, earned an Honorable Mention at the 2025 Entrepreneurship Olympiad, held July 21–23 at Stanford Faculty Club and hosted by Techdev Academy. Competing in the college track, the team showcased their innovative solution among participants from top institutions including Stanford University, UC Berkeley, UCLA, and UC San Diego.
Team eaureco—comprising KAIST undergraduate and graduate students Jiwon Park(Semiconductor Systems Engineering), Si Li Sara (Julia) Aow, Lunar Sebastian Widjaja (both Civil & Environmental Engineering), Seoyeon Jang (Impact MBA), and Isabel Alexandra Cornejo Lima (BTM/Global Digital Innovation)—presented a B2B solution that upcycles discarded seaweed into biodegradable ice packs for cold-chain companies. Their business model was recognized for its alignment with sustainability, resource circulation, and social impact goals.
<Photo: eaureco team preparing for the final pitch>
The team’s ability to rapidly adapt their pitch based on mentor feedback and clearly communicate the value of their idea to judges contributed to their recognition. This accomplishment further highlights the impact of KAIST's GESS program, which supports students in building real-world entrepreneurial skills through immersive learning experiences in Silicon Valley.
“The GESS program helped us refine every aspect of our business idea—from identifying the problem to developing a go-to-market strategy,” said Si Li Sara (Julia) Aow, a member of the eaureco team. “We’re grateful for the opportunity to showcase our work on a global stage and hope to continue developing innovations that drive meaningful change.”
“This award reaffirms the creative potential and practical capabilities of KAIST students in global innovation ecosystems,” said Dr. Soyoung Kim, Vice President of International Office. “We will continue to invest in programs like GESS to empower our students as future leaders in entrepreneurship.”
The Entrepreneurship Olympiad is a global event designed to foster innovation, entrepreneurship, and collaboration among young change-makers. This year’s program featured keynote talks, panels, and workshops led by industry pioneers including Marc Tarpenning (Co-founder, Tesla Motors), Pat Brown (Founder, Impossible Foods), and other influential entrepreneurs from the biotech, fintech, and deeptech sectors.
The Honorable Mention recognition underscores KAIST’s commitment to global entrepreneurship education and the growing international visibility of the GESS program.
2025.07.29 View 395 -
Immune Signals Directly Modulate Brain's Emotional Circuits: Unraveling the Mechanism Behind Anxiety-Inducing Behaviors
KAIST's Department of Brain and Cognitive Sciences, led by Professor Jeong-Tae Kwon, has collaborated with MIT and Harvard Medical School to make a groundbreaking discovery. For the first time globally, their joint research has revealed that cytokines, released during immune responses, directly influence the brain's emotional circuits to regulate anxiety behavior.
The study provided experimental evidence for a bidirectional regulatory mechanism: inflammatory cytokines IL-17A and IL-17C act on specific neurons in the amygdala, a region known for emotional regulation, increasing their excitability and consequently inducing anxiety. Conversely, the anti-inflammatory cytokine IL-10 was found to suppress excitability in these very same neurons, thereby contributing to anxiety alleviation.
In a mouse model, the research team observed that while skin inflammation was mitigated by immunotherapy (IL-17RA antibody), anxiety levels paradoxically rose. This was attributed to elevated circulating IL-17 family cytokines leading to the overactivation of amygdala neurons.
Key finding: Inflammatory cytokines IL-17A/17C promote anxiety by acting on excitable amygdala neurons (via IL-17RA/RE receptors), whereas anti-inflammatory cytokine IL-10 alleviates anxiety by suppressing excitability through IL-10RA receptors on the same neurons.
The researchers further elucidated that the anti-inflammatory cytokine IL-10 works to reduce the excitability of these amygdala neurons, thereby mitigating anxiety responses.
This research marks the first instance of demonstrating that immune responses, such as infections or inflammation, directly impact emotional regulation at the level of brain circuits, extending beyond simple physical reactions. This is a profoundly significant achievement, as it proposes a crucial biological mechanism that interlinks immunity, emotion, and behavior through identical neurons within the brain.
The findings of this research were published in the esteemed international journal Cell on April 17th of this year.
Paper Information:
Title: Inflammatory and anti-inflammatory cytokines bidirectionally modulate amygdala circuits regulating anxiety
Journal: Cell (Vol. 188, 2190–2220), April 17, 2025
DOI: https://doi.org/10.1016/j.cell.2025.03.005
Corresponding Authors: Professor Gloria Choi (MIT), Professor Jun R. Huh (Harvard Medical School)
2025.07.24 View 338 -
KAIST Designs a New Atomic Catalyst for Air Pollution Reduction
<(From Left)Professor Jong Hun Kim from Inha University, Dr. Gyuho Han and Professor Jeong Young Park from KAIST>
Platinum diselenide (PtSe2) is a two-dimensional multilayer material in which each layer is composed of platinum (Pt) and selenium (Se). It is known that its excellent crystallinity and precise control of interlayer interactions allow modulation of various physical and chemical properties. Due to these characteristics, it has been actively researched in multiple fields, including semiconductors, photodetectors, and electrochemical devices. Now, a research team has proposed a new design concept in which atomically dispersed platinum on the surface of platinum diselenide can function as a catalyst for gas reactions. Through this, they have proven its potential as a next-generation gas-phase catalyst technology for high-efficiency carbon dioxide conversion and carbon monoxide reduction.
KAIST (President Kwang Hyung Lee) announced on July 22 that a joint research team led by Endowed Chair Professor Jeong Young Park from the Department of Chemistry, along with Professor Hyun You Kim's team from Chungnam National University and Professor Yeonwoong (Eric) Jung's team from the University of Central Florida (UCF), has achieved excellent carbon monoxide oxidation performance by utilizing platinum atoms exposed on the surface of platinum diselenide, a type of two-dimensional transition metal dichalcogenide (TMD).
To maximize catalytic performance, the research team designed the catalyst by dispersing platinum atoms uniformly across the surface, departing from the conventional use of bulk platinum. This strategy allows more efficient catalytic reactions using a smaller amount of platinum. It also enhances electronic interactions between platinum and selenium by tuning the surface electronic structure. As a result, the platinum diselenide film with a thickness of a few nanometers showed superior carbon monoxide oxidation performance across the entire temperature range compared to a conventional platinum thin film under identical conditions.
In particular, carbon monoxide and oxygen were evenly adsorbed on the surface in similar proportions, increasing the likelihood that they would encounter each other and react, which significantly enhanced the catalytic activity. This improvement is primarily attributed to the increased exposure of surface platinum atoms resulting from selenium vacancies (Se-vacancies), which provide adsorption sites for gas molecules.
The research team confirmed in real-time that these platinum atoms served as active adsorption sites during the actual reaction process, using ambient-pressure X-ray photoelectron spectroscopy (AP-XPS) conducted at the Pohang Accelerator Laboratory. This high-precision analysis was enabled by advanced instrumentation capable of observing surfaces at the nanometer scale under ambient pressure conditions. At the same time, computer simulations based on density functional theory (DFT) demonstrated that platinum diselenide exhibits distinct electronic behavior compared to conventional platinum.
*Density Functional Theory (DFT): A quantum mechanical method for calculating the total energy of a system based on electron density.
Professor Jeong Young Park stated, “This research presents a new design strategy that utilizes platinum diselenide, a two-dimensional layered material distinct from conventional platinum catalysts, to enable catalytic functions optimized for gas-phase reactions.” He added, “The electronic interaction between platinum and selenium created favorable conditions for the balanced adsorption of carbon monoxide and oxygen. By designing the catalyst to exhibit higher reactivity across the entire temperature range than conventional platinum, we improved its practical applicability. This enabled a high-efficiency catalytic reaction mechanism through atomic-level design, a two-dimensional material platform, and precise adsorption control.”
This research was co-authored by Dr. Gyuho Han from the Department of Chemistry at KAIST, Dr. Hyuk Choi from the Department of Materials Science and Engineering at Chungnam National University, and Professor Jong Hun Kim from Inha University. The study was published on July 3 in the world-renowned journal Nature Communications.
Paper Title: Enhanced catalytic activity on atomically dispersed PtSe2 two-dimensional layers
DOI: 10.1038/s41467-025-61320-0
This research was supported by the Mid-Career Researcher Program of the Ministry of Science and ICT, the Core Research Institute Program of the Ministry of Education, the National Strategic Technology Materials Development Project, the U.S. National Science Foundation (NSF) CAREER Program, research funding from Inha University, and the Postdoctoral Researcher Program (P3) at UCF. Accelerator-based analysis was conducted in cooperation with the Pohang Accelerator Laboratory and the Korea Basic Science Institute (KBSI).
2025.07.22 View 358 -
KAIST Successfully Implements 3D Brain-Mimicking Platform with 6x Higher Precision
<(From left) Dr. Dongjo Yoon, Professor Je-Kyun Park from the Department of Bio and Brain Engineering, (upper right) Professor Yoonkey Nam, Dr. Soo Jee Kim>
Existing three-dimensional (3D) neuronal culture technology has limitations in brain research due to the difficulty of precisely replicating the brain's complex multilayered structure and the lack of a platform that can simultaneously analyze both structure and function. A KAIST research team has successfully developed an integrated platform that can implement brain-like layered neuronal structures using 3D printing technology and precisely measure neuronal activity within them.
KAIST (President Kwang Hyung Lee) announced on the 16th of July that a joint research team led by Professors Je-Kyun Park and Yoonkey Nam from the Department of Bio and Brain Engineering has developed an integrated platform capable of fabricating high-resolution 3D multilayer neuronal networks using low-viscosity natural hydrogels with mechanical properties similar to brain tissue, and simultaneously analyzing their structural and functional connectivity.
Conventional bioprinting technology uses high-viscosity bioinks for structural stability, but this limits neuronal proliferation and neurite growth. Conversely, neural cell-friendly low-viscosity hydrogels are difficult to precisely pattern, leading to a fundamental trade-off between structural stability and biological function. The research team completed a sophisticated and stable brain-mimicking platform by combining three key technologies that enable the precise creation of brain structure with dilute gels, accurate alignment between layers, and simultaneous observation of neuronal activity.
The three core technologies are: ▲ 'Capillary Pinning Effect' technology, which enables the dilute gel (hydrogel) to adhere firmly to a stainless steel mesh (micromesh) to prevent it from flowing, thereby reproducing brain structures with six times greater precision (resolution of 500 μm or less) than conventional methods; ▲ the '3D Printing Aligner,' a cylindrical design that ensures the printed layers are precisely stacked without misalignment, guaranteeing the accurate assembly of multilayer structures and stable integration with microelectrode chips; and ▲ 'Dual-mode Analysis System' technology, which simultaneously measures electrical signals from below and observes cell activity with light (calcium imaging) from above, allowing for the simultaneous verification of the functional operation of interlayer connections through multiple methods.
< Figure 1. Platform integrating brain-structure-mimicking neural network model construction and functional measurement technology>
The research team successfully implemented a three-layered mini-brain structure using 3D printing with a fibrin hydrogel, which has elastic properties similar to those of the brain, and experimentally verified the process of actual neural cells transmitting and receiving signals within it.
Cortical neurons were placed in the upper and lower layers, while the middle layer was left empty but designed to allow neurons to penetrate and connect through it. Electrical signals were measured from the lower layer using a microsensor (electrode chip), and cell activity was observed from the upper layer using light (calcium imaging). The results showed that when electrical stimulation was applied, neural cells in both upper and lower layers responded simultaneously. When a synapse-blocking agent (synaptic blocker) was introduced, the response decreased, proving that the neural cells were genuinely connected and transmitting signals.
Professor Je-Kyun Park of KAIST explained, "This research is a joint development achievement of an integrated platform that can simultaneously reproduce the complex multilayered structure and function of brain tissue. Compared to existing technologies where signal measurement was impossible for more than 14 days, this platform maintains a stable microelectrode chip interface for over 27 days, allowing the real-time analysis of structure-function relationships. It can be utilized in various brain research fields such as neurological disease modeling, brain function research, neurotoxicity assessment, and neuroprotective drug screening in the future."
The research, in which Dr. Soo Jee Kim and Dr. Dongjo Yoon from KAIST's Department of Bio and Brain Engineering participated as co-first authors, was published online in the international journal 'Biosensors and Bioelectronics' on June 11, 2025.
※Paper: Hybrid biofabrication of multilayered 3D neuronal networks with structural and functional interlayer connectivity
※DOI: https://doi.org/10.1016/j.bios.2025.117688
2025.07.16 View 476 -
Professor Jung-woo' Choi ‘s Team Comes in First at the World's Top Acoustic AI Challenge
<Photo1. (From left) Ph.D candidate Yong-hoo Kwon, M.S candidate Do-hwan Kim, Professor Jung-woo Choi, Dr. Dong-heon Lee>
'Acoustic separation and classification technology' is a next-generation artificial intelligence (AI) core technology that enables the early detection of abnormal sounds in areas such as drones, fault detection of factory pipelines, and border surveillance systems, or allows for the separation and editing of spatial audio by sound source when producing AR/VR content.
On the 11th of July, a research team led by Professor Jung-woo Choi of KAIST's Department of Electrical and Electronic Engineering won first place in the 'Spatial Semantic Segmentation of Sound Scenes' task of the 'DCASE2025 Challenge,' the world's most prestigious acoustic detection and analysis competition.
This year’s challenge featured 86 teams competing across six tasks. In this competition, the KAIST research team achieved the best performance in their first-ever participation to Task 4. Professor Jung-woo Choi’s research team consisted of Dr. Dong-heon, Lee, Ph.D. candidate Young-hoo Kwon, and M.S. candidate Do-hwan Kim.
Task 4 titled 'Spatial Semantic Segmentation of Sound Scenes' is a highly demanding task requiring the analysis of spatial information in multi-channel audio signals with overlapping sound sources. The goal was to separate individual sounds and classify them into 18 predefined categories. The research team plans to present their technology at the DCASE workshop in Barcelona this October.
<Figure 1. Example of an acoustic scene with multiple mixed sounds>
Early this year, Dr. Dong-heon Lee developed a state-of-the-art sound source separation AI that combines Transformer and Mamba architectures. During the competition, centered around researcher Young-hoo Kwon, they completed a ‘chain-of-inference architecture' AI model that performs sound source separation and classification again, using the waveforms and types of the initially separated sound sources as clues. This AI model is inspired by human’s auditory scene analysis mechanism that isolates individual sounds by focusing on incomplete clues such as sound type, rhythm, or direction, when listening to complex sounds.
Through this, the team was the only participant to achieve double-digit performance (11 dB) in 'Class-Aware Signal-to-Distortion Ratio Improvement (CA-SDRi)*,' which is the measure for ranking how well the AI separated and classified sounds, proving their technical excellence.
Class-Aware Signal-to-Distortion Ratio Improvement (CA-SDRi): Measures how much clearer (less distorted) the desired sound is separated and classified compared to the original audio, in dB (decibels). A higher number indicates more accurate and cleaner sound separation.
Prof. Jung-woo Choi remarked, "The research team has showcased world-leading acoustic separation AI models for the past three years, and I am delighted that these results have been officially recognized." He added, "I am proud of every member of the research team for winning first place through focused research, despite the significant increase in difficulty and having only a few weeks for development."
<Figure 2. Time-frequency patterns of sound sources separated from a mixed source>
The IEEE DCASE Challenge 2025 was held online, with submissions accepted from April 1 to June 15 and results announced on June 30. Since its launch in 2013, the DCASE Challenge has served as a premier global platform of IEEE Signal Processing Society for showcasing cutting-edge AI models in acoustic signal processing.
This research was supported by the Mid-Career Researcher Support Project and STEAM Research Project of the National Research Foundation of Korea, funded by the Ministry of Education, Science and Technology, as well as support from the Future Defense Research Center, funded by the Defense Acquisition Program Administration and the Agency for Defense Development.
2025.07.13 View 403 -
KAIST Kicks Off the Expansion of its Creative Learning Building, a 50th Anniversary Donation Landmark
KAIST announced on July 10th that it held a groundbreaking ceremony on July 9th for the expansion of its Creative Learning Building. This project, which celebrates the university's 50th anniversary, will become a significant donation-funded landmark and marks the official start of its construction.
<(From left) President Kwang Hyung Lee, Former President Sung-Chul Shin>
The groundbreaking ceremony was attended by key donors who graced the occasion, including KAIST President Kwang Hyung Lee, former President Sung-Chul Shin, Alumni Association President Yoon-Tae Lee, as well as parents and faculty member.
The Creative Learning Building serves as a primary space where KAIST undergraduate and graduate students attend lectures, functioning as a central hub for a variety of classes and talks. It also houses student support departments, including the Student Affairs Office, establishing itself as a student-centric complex that integrates educational, counseling, and welfare functions.
This expansion is more than just an increase in educational facilities; it's being developed as a "donation landmark" embodying KAIST's identity and future vision. Designed with a focus on creative convergence education, this project aims to create a new educational hub that organically combines education, exchange, and welfare functions
The campaign included over 230 participants, including KAIST alumni Byung-gyu Chang, Chairman of Krafton, former Alumni Association President Ki-chul Cha, Dr. Kun-mo Chung (former Minister of Science and Technology), as well as faculty members, parents, and current students. They collectively raised 6.5 billion KRW in donations. The total cost for this expansion project is 9 billion KRW, encompassing a gross floor area of 3,222.92㎡ across five above-ground floors, with completion targeted for September 2026.
2025.07.10 View 462 -
Professor Moon-Jeong Choi Appointed as an Advisor for the ITU's 'AI for Good Global Summit'
Professor Moon-Jeong Choi from KAIST’s Graduate School of Science and Technology Policy has been appointed as an advisor for "Innovate for Impact" at the AI for Good Global Summit, organized by the International Telecommunication Union (ITU), a specialized agency of the United Nations (UN).
The ITU is the UN's oldest specialized agency in the field of information and communication technology (ICT) and serves as a crucial body for coordinating global ICT policies and standards.
This advisory committee was formed to explore global cooperation strategies for realizing the social value of Artificial Intelligence (AI) and promoting sustainable development. Experts from around the world are participating as committee members, with Professor Choi being the sole Korean representative.
<Moon-Jeong Choi from KAIST’s Graduate School of Science and Technology Policy>
The AI for Good Global Summit is taking place in Geneva, Switzerland from July 8 to 11. It is organized by the ITU in collaboration with approximately 40 other UN-affiliated organizations. The summit aims to address global challenges facing humanity through the use of AI technology, focusing on key agenda items such as identifying AI application cases, discussing international policies and technical standards, and strengthening global partnerships.
As an "Innovate for Impact" advisor, Professor Choi will evaluate AI application cases from various countries, participating in case analyses primarily focused on public interest and social impact. The summit will move beyond discussions of technical performance to focus on how AI can contribute to the public good, with diverse case studies from around the world being debated. Notably, during a policy panel discussion at the summit, Professor Choi will discuss policy frameworks for AI transparency, inclusivity, and fairness under the theme of "Responsible AI Development."
Professor Choi commented, "I believe the social impact of technology mirrors the values and systems of each nation. As a society's core values permeate technology, the way AI is developed and used varies significantly from country to country. These differences lead to diverse manifestations of AI's impact on society." She further emphasized, "Korea's vision of becoming an AI powerhouse should not merely be about technological superiority, but rather about enhancing social capital through human-centered AI and realizing communal values that enable us to live together."
Professor Moon-Jeong Choi currently serves as the Dean of the Graduate School of Science and Technology Policy. She is also an external director for the National Information Society Agency (2023-present) and chair of the Korea-OECD Digital Society Initiative (2024-present).
For more information about the AI for Good Global Summit, please visit the official website: https://aiforgood.itu.int.
2025.07.08 View 462 -
Development of Core NPU Technology to Improve ChatGPT Inference Performance by Over 60%
Latest generative AI models such as OpenAI's ChatGPT-4 and Google's Gemini 2.5 require not only high memory bandwidth but also large memory capacity. This is why generative AI cloud operating companies like Microsoft and Google purchase hundreds of thousands of NVIDIA GPUs. As a solution to address the core challenges of building such high-performance AI infrastructure, Korean researchers have succeeded in developing an NPU (Neural Processing Unit)* core technology that improves the inference performance of generative AI models by an average of over 60% while consuming approximately 44% less power compared to the latest GPUs.
*NPU (Neural Processing Unit): An AI-specific semiconductor chip designed to rapidly process artificial neural networks.
On the 4th, Professor Jongse Park's research team from KAIST School of Computing, in collaboration with HyperAccel Inc. (a startup founded by Professor Joo-Young Kim from the School of Electrical Engineering), announced that they have developed a high-performance, low-power NPU (Neural Processing Unit) core technology specialized for generative AI clouds like ChatGPT.
The technology proposed by the research team has been accepted by the '2025 International Symposium on Computer Architecture (ISCA 2025)', a top-tier international conference in the field of computer architecture.
The key objective of this research is to improve the performance of large-scale generative AI services by lightweighting the inference process, while minimizing accuracy loss and solving memory bottleneck issues. This research is highly recognized for its integrated design of AI semiconductors and AI system software, which are key components of AI infrastructure.
While existing GPU-based AI infrastructure requires multiple GPU devices to meet high bandwidth and capacity demands, this technology enables the configuration of the same level of AI infrastructure using fewer NPU devices through KV cache quantization*. KV cache accounts for most of the memory usage, thereby its quantization significantly reduces the cost of building generative AI clouds.
*KV Cache (Key-Value Cache) Quantization: Refers to reducing the data size in a type of temporary storage space used to improve performance when operating generative AI models (e.g., converting a 16-bit number to a 4-bit number reduces data size by 1/4).
The research team designed it to be integrated with memory interfaces without changing the operational logic of existing NPU architectures. This hardware architecture not only implements the proposed quantization algorithm but also adopts page-level memory management techniques* for efficient utilization of limited memory bandwidth and capacity, and introduces new encoding technique optimized for quantized KV cache.
*Page-level memory management technique: Virtualizes memory addresses, as the CPU does, to allow consistent access within the NPU.
Furthermore, when building an NPU-based AI cloud with superior cost and power efficiency compared to the latest GPUs, the high-performance, low-power nature of NPUs is expected to significantly reduce operating costs.
Professor Jongse Park stated, "This research, through joint work with HyperAccel Inc., found a solution in generative AI inference lightweighting algorithms and succeeded in developing a core NPU technology that can solve the 'memory problem.' Through this technology, we implemented an NPU with over 60% improved performance compared to the latest GPUs by combining quantization techniques that reduce memory requirements while maintaining inference accuracy, and hardware designs optimized for this".
He further emphasized, "This technology has demonstrated the possibility of implementing high-performance, low-power infrastructure specialized for generative AI, and is expected to play a key role not only in AI cloud data centers but also in the AI transformation (AX) environment represented by dynamic, executable AI such as 'Agentic AI'."
This research was presented by Ph.D. student Minsu Kim and Dr. Seongmin Hong from HyperAccel Inc. as co-first authors at the '2025 International Symposium on Computer Architecture (ISCA)' held in Tokyo, Japan, from June 21 to June 25. ISCA, a globally renowned academic conference, received 570 paper submissions this year, with only 127 papers accepted (an acceptance rate of 22.7%).
※Paper Title: Oaken: Fast and Efficient LLM Serving with Online-Offline Hybrid KV Cache Quantization
※DOI: https://doi.org/10.1145/3695053.3731019
Meanwhile, this research was supported by the National Research Foundation of Korea's Excellent Young Researcher Program, the Institute for Information & Communications Technology Planning & Evaluation (IITP), and the AI Semiconductor Graduate School Support Project.
2025.07.07 View 1124 -
KAIST researcher Se Jin Park develops 'SpeechSSM,' opening up possibilities for a 24-hour AI voice assistant.
<(From Left)Prof. Yong Man Ro and Ph.D. candidate Sejin Park>
Se Jin Park, a researcher from Professor Yong Man Ro’s team at KAIST, has announced 'SpeechSSM', a spoken language model capable of generating long-duration speech that sounds natural and remains consistent.
An efficient processing technique based on linear sequence modeling overcomes the limitations of existing spoken language models, enabling high-quality speech generation without time constraints.
It is expected to be widely used in podcasts, audiobooks, and voice assistants due to its ability to generate natural, long-duration speech like humans.
Recently, Spoken Language Models (SLMs) have been spotlighted as next-generation technology that surpasses the limitations of text-based language models by learning human speech without text to understand and generate linguistic and non-linguistic information. However, existing models showed significant limitations in generating long-duration content required for podcasts, audiobooks, and voice assistants. Now, KAIST researcher has succeeded in overcoming these limitations by developing 'SpeechSSM,' which enables consistent and natural speech generation without time constraints.
KAIST(President Kwang Hyung Lee) announced on the 3rd of July that Ph.D. candidate Sejin Park from Professor Yong Man Ro's research team in the School of Electrical Engineering has developed 'SpeechSSM,' a spoken. a spoken language model capable of generating long-duration speech.
This research is set to be presented as an oral paper at ICML (International Conference on Machine Learning) 2025, one of the top machine learning conferences, selected among approximately 1% of all submitted papers. This not only proves outstanding research ability but also serves as an opportunity to once again demonstrate KAIST's world-leading AI research capabilities.
A major advantage of Spoken Language Models (SLMs) is their ability to directly process speech without intermediate text conversion, leveraging the unique acoustic characteristics of human speakers, allowing for the rapid generation of high-quality speech even in large-scale models.
However, existing models faced difficulties in maintaining semantic and speaker consistency for long-duration speech due to increased 'speech token resolution' and memory consumption when capturing very detailed information by breaking down speech into fine fragments.
To solve this problem, Se Jin Park developed 'SpeechSSM,' a spoken language model using a Hybrid State-Space Model, designed to efficiently process and generate long speech sequences.
This model employs a 'hybrid structure' that alternately places 'attention layers' focusing on recent information and 'recurrent layers' that remember the overall narrative flow (long-term context). This allows the story to flow smoothly without losing coherence even when generating speech for a long time. Furthermore, memory usage and computational load do not increase sharply with input length, enabling stable and efficient learning and the generation of long-duration speech.
SpeechSSM effectively processes unbounded speech sequences by dividing speech data into short, fixed units (windows), processing each unit independently, and then combining them to create long speech.
Additionally, in the speech generation phase, it uses a 'Non-Autoregressive' audio synthesis model (SoundStorm), which rapidly generates multiple parts at once instead of slowly creating one character or one word at a time, enabling the fast generation of high-quality speech.
While existing models typically evaluated short speech models of about 10 seconds, Se Jin Park created new evaluation tasks for speech generation based on their self-built benchmark dataset, 'LibriSpeech-Long,' capable of generating up to 16 minutes of speech.
Compared to PPL (Perplexity), an existing speech model evaluation metric that only indicates grammatical correctness, she proposed new evaluation metrics such as 'SC-L (semantic coherence over time)' to assess content coherence over time, and 'N-MOS-T (naturalness mean opinion score over time)' to evaluate naturalness over time, enabling more effective and precise evaluation.
Through these new evaluations, it was confirmed that speech generated by the SpeechSSM spoken language model consistently featured specific individuals mentioned in the initial prompt, and new characters and events unfolded naturally and contextually consistently, despite long-duration generation. This contrasts sharply with existing models, which tended to easily lose their topic and exhibit repetition during long-duration generation.
PhD candidate Sejin Park explained, "Existing spoken language models had limitations in long-duration generation, so our goal was to develop a spoken language model capable of generating long-duration speech for actual human use." She added, "This research achievement is expected to greatly contribute to various types of voice content creation and voice AI fields like voice assistants, by maintaining consistent content in long contexts and responding more efficiently and quickly in real time than existing methods."
This research, with Se Jin Park as the first author, was conducted in collaboration with Google DeepMind and is scheduled to be presented as an oral presentation at ICML (International Conference on Machine Learning) 2025 on July 16th.
Paper Title: Long-Form Speech Generation with Spoken Language Models
DOI: 10.48550/arXiv.2412.18603
Ph.D. candidate Se Jin Park has demonstrated outstanding research capabilities as a member of Professor Yong Man Ro's MLLM (multimodal large language model) research team, through her work integrating vision, speech, and language. Her achievements include a spotlight paper presentation at 2024 CVPR (Computer Vision and Pattern Recognition) and an Outstanding Paper Award at 2024 ACL (Association for Computational Linguistics).
For more information, you can refer to the publication and accompanying demo: SpeechSSM Publications.
2025.07.04 View 988