We have established the Accton–NCKU Joint Research Center (智邦–成大聯合研發中心). In 2025, Accton and Edgecore announced the launch of the new Nexvec™ solution, designed to provide enterprises with an integrated and open platform for AI infrastructure. This solution integrates disaggregated networking with composable compute and storage, combining Edgecore’s open networking product line, Liqid’s dynamic resource allocation, and the self-developed Nous controller to enable fully automated Day 0 to Day 2 operations. Its primary goal is to simplify AI deployment, support scalable and programmable architectures, and accelerate enterprise adoption of AI applications. As chair of the Accton–NCKU Joint Research Center, Prof. Chuang leads several key initiatives, particularly focusing on service deployment workflows in the Liqid mini AI center. Our research explores Kubernetes-based marketplace deployment, data lake frameworks such as Apache Iceberg within composable AI architectures, and AI toolchains including MLFlow, Open Deep Research, OpenWebUI, and AnythingLLM. These efforts aim to enable Model Context Protocol (MCP) integration and AI agent frameworks within Liqid’s composable infrastructure. In 2025, our research emphasis is on scalable distributed LLM architectures. Leveraging Liqid’s hardware composability, which supports small- to medium-scale GPU clusters, we pursue a holistic approach to distributed Large Language Models—focusing on scalability, resource efficiency, and performance breakthroughs: Scalable Distributed Architectures: Using frameworks such as Exo, GPUStack, and Ray, we support flexible, multi-vendor GPU and cross-platform deployments, enabling models at the 405B-parameter scale. Advanced Resource Optimization: Through intelligent model sharding and heterogeneous GPU collaboration, we significantly reduce single-GPU VRAM requirements and extend support across diverse hardware ecosystems, including NVIDIA, AMD, and Intel. High-Performance Network Optimization: With Edgecore’s QoS technologies, we address latency bottlenecks to deliver near single-machine inference speeds in distributed environments—achieving performance at a fraction of the traditional cost. Edgecore Networks Announces its Nexvec™ Solution for Enterprise AI - Edgecore Networks

We are spearheading a monumental initiative for the Ministry of Education, taking charge of the service development for the Digital Diploma and Digital Wallet (TW-DIW) projects for all university in Taiwan based on the Distributed Ledger Technologies (DLT). This service has rapidly expanded its reach, demonstrating versatility across various fields, including insurance and medical services, and other projects of digital certificates. It's about revolutionizing how credentials are managed and verified, setting a new standard for digital trust and efficiency nationwide. 教育部電子證書驗證系統

Wen join in the smart city project led by Itochu in Tainan’s Shalun Green Energy Science City, serving as a key hub connecting academic research, government policy, and smart city applications. In this project, our mission is to integrate cutting-edge technological innovations with urban governance, energy management, and smart living solutions, thereby enabling Shalun to become Taiwan’s flagship demonstration site for low-carbon digital innovation: Smart Energy and Data Governance: Building on the outcomes of previous smart energy projects, we lead the effort in system data decoupling and the establishment of the Shalun Data Lake and City Databank. By providing APIs aligned with data governance standards, we ensure secure data sharing and cross-platform interoperability, laying the foundation for scalable applications. Generative AI and Smart Applications: We develop the Shalun Generative AI Chatbot based on TAIDE LLM. This service, based on the LLM function calling capability, enables residents and visitors to access real-time information on pedestrian and traffic flows, as well as energy usage, positioning Shalun as a pioneer in AI-driven urban innovation. AI-Empowered Low-Carbon Living: Through initiatives such as the GreenCoin low-carbon lifestyle platform and the development of medical and health innovation zones, we integrate machine learning–based prediction capabilities to design affordable incentives that encourage residents to adopt net-zero practices, such as car-sharing and participation in Virtual Power Plant (VPP) demand response programs. These efforts are driving Shalun’s transformation into a model low-carbon smart city. 沙崙智慧綠能科學城智慧接駁服務 免費提供站對站預約接送 於2025年運行 20250923 臺南沙崙預見未來 沙崙智慧綠能科學城 2025版中文 Related Research Outcomes Lo Pang-Yun Ting, Yu-Hao Chiang, Yu-Hsiang Chang, Michihiro Konagai, Kun-Ta Chuang, "Optimized Battery Controlling for Smart Homes with Dynamic Insensitive-Zone Adjustment", GCCE 2024: 1257-1259

We join the project of Infineon Technologies on 'Advanced Automotive and Wireless Communication Semiconductor'. We are leveraging cutting-edge AI technologies to focus on two critical areas: wBMS Battery Management for wBMS: Developing advanced transformer-based and Mamba-based models for precise State-of-Charge (SOC), State-of-Health (SOH), and Remaining Useful Life (RUL) prediction. We implement a python-library, call MambaBSP, to achieve high precision of Batter State Prediction based on Mamba technologies. Non-intrusive Passenger Occupation Detection: Utilizing WIFI Channel State Information (CSI) and Deep Learning methodologies to accurately detect car passengers without physical sensors, enhancing safety and smart cabin features.

The AI for Mental Good project aims to develop an AI-assisted counseling support system for schools, addressing the growing challenges of student mental health. By combining Retrieval-Augmented Generation (RAG) with local Large Language Models (LLMs), the system can rapidly retrieve and analyze relevant case data, provide structured guidance to teachers and counselors, and enhance knowledge sharing across institutions. The platform emphasizes immediacy, professionalism, and data security, supporting crisis response, structured case management, and professional training. In particular, the project will study the utility of vector databases and explore methods to embed counseling knowledge securely within the RAG framework, ensuring both high retrieval accuracy and strict protection of sensitive data. With features like risk alerts, anonymized case databases, and local deployment, it strengthens counseling capacity, reduces delays, and builds a sustainable knowledge hub for mental health education.

Improper handling of industrial chemicals can cause severe environmental damage, especially when processing methods are illegal. To mitigate this risk, a blockchain-based industrial chemical tracking system can create a secure digital footprint for each chemical. By leveraging blockchain, the system builds trust across the supply chain, allowing all stakeholders to transparently track and verify the movement of industrial chemicals. This not only enhances accountability but also helps operators reduce management costs and streamline packaging processes.

We initiate a new 'Security for LLM' research sub-project in NSTC TACC . Detecting poisoned content in Retrieval-Augmented Generation (RAG) for local LLMs is crucial because compromised knowledge bases can silently mislead models into producing false or biased answers. Unlike cloud systems, local LLMs often operate in sensitive domains—such as healthcare, finance, or government—where even a few injected malicious texts can distort decision-making and undermine trust. Robust detection mechanisms not only protect against disinformation and data integrity attacks but also ensure that local deployments maintain reliability, security, and compliance in mission-critical applications.

We partner with TEPCO PowerGrid Company (東京電力公司) on the “Quality of Life Services” project, based on Japan’s Utility 3.0 concept. By analyzing data from water, electricity, and gas meters, the project develops services that enhance daily living: Smart Home: Optimize appliances to save energy and improve comfort. Water Leakage Detection: Real-time leak monitoring via meter data. Healthcare: Vital sign and behavior monitoring for early warnings at home. Utility Data Collection (UDC) At the Shalun Smart Green Energy Science City (Zone D, Taiwan), we built a Utility Data Collection (UDC) system with TEPCO to gather data from meters, sensors, appliances, and weather sources. The UDC provides a high-throughput, reliable data pipeline, bridging households, vendors, and Energy Service Companies (ESCOs). It supports analytics and enables new services such as Quality of Life applications. Key Contributions Water Leakage Detection Service Occupancy-Aware Delivery Optimization using NILM-based electricity data Coupon-based Demand Response to balance Household Electricity Load Encourages off-peak energy use with local business coupons. Lowers community load and stabilizes the power grid. Related Research Outcomes Lo Pang-Yun Ting, You-Cheng Guo, Chi-Chun Lin, Shih-Hsun Lin and Kun-Ta Chuang*, “MAC-DP: Multi-Agent Control for Dynamic Placement of Electric Vehicle Charging Stations”, International Journal of Data Science and Analytics (JDSA), accepted. Lo Pang-Yun Ting and Huan-Yang Wang and Jhe-Yun Jhang and Kun-Ta Chuang, “Online Spatial-Temporal EV Charging Scheduling with Incentive Promotion”, accepted by ACM Transactions on Intelligent Systems and Technology, 2024 (Impact Factor 2023: 7.2) Lo Pang-Yun Ting and Rong Chao and Chai-Shi Chang and Kun-Ta Chuang, 2024, “An Explore-Exploit Workload-Bounded Strategy for Rare Event Detection in Massive Energy Sensor Time Series”, ACM Transactions on Intelligent Systems and Technology, 15(4), 69:1--69:25 (Impact Factor 2023: 7.2) Lo Pang-Yun Ting and Shan-Yun Teng and Szu-Chan Wu and Kun-Ta Chuang, 2022, “Interactive planning of revisiting-free itinerary for signed-for delivery”, International Journal of Data Science and Analytics, 14(4), 439—456 (Impact Factor 2023: 3.4) Lo Pang-Yun Ting and Chi-Chun Lin and Shih-Hsun Lin and Yu-Lin Chu and Kun-Ta Chuang, 2024, Multi-agent Reinforcement Learning for Online Placement of Mobile EV Charging Stations, Advances in Knowledge Discovery and Data Mining - 28th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2024, Taipei, Taiwan, May 7-10, 2024, Proceedings, Part V, 284—296 (oral presentation, acceptance rate 19%) Moritz Rosar and Kun-Ta Chuang, 2024, Using Blockchain for Incentive-Driven FHIR-based Health Record Exchange, Proceedings of the 39th ACM/SIGAPP Symposium on Applied Computing, SAC 2024, Avila, Spain, April 8-12, 2024, 298—299 Lo Pang-Yun Ting and Po-Hui Wu and Hsiu-Ying Chung and Kun-Ta Chuang, 2022, An Incentive Dispatch Algorithm for Utilization-Perfect EV Charging Management, Advances in Knowledge Discovery and Data Mining - 26th Pacific-Asia Conference, PAKDD 2022, Chengdu, China, May 16-19, 2022, Proceedings, Part III, 132—146 (blind-review, acceptance rate=19.3%, out of 627 submissions)

We participate in NXP Semiconductor's Advanced Semiconductor R&D Center and Automotive Industry Linkage project, focusing on optimizing IC testing processes through ATE and enhancing test pattern correlation with graph-based deep learning models. Related Research Outcomes Lo Pang-Yun Ting, Yu-Hao Chiang, Yi-Tung Tsai, Hsu-Chao Lai, Kun-Ta Chuang, “DeCo: Defect-Aware Modeling with Contrasting Matching for Optimizing Task Assignment in Online IC Testing”, International Joint Conference on Artificial Intelligence, IJCAI 2025.

We pioneered the creation of "用數據看台灣", an innovative open-data visualization platform designed to democratize access to critical public information. This platform features a dynamic web visualization framework that presents both real-time and statistical data, allowing users to explore complex datasets with ease. It visualizes a wide range of data, including demographic trends, economic indicators, public health statistics, environmental data, and various government statistics. By offering interactive dashboards and customizable views, the platform empowers researchers, policymakers, and the general public to make data-driven decisions, foster a deeper understanding of societal issues, and enhance government transparency. This initiative represents a significant step towards building a more data-literate society and promoting informed civic engagement nationwide. Check TaiwanStat for more details. 用數據看台灣

We continuously monitor body temperature changes to identify infectious diseases early and assess the risk of poor prognosis. This approach can also detect potential disease clusters, helping to reduce complications and comorbidities. Using smart IoT devices, we collected large-scale health data and established personalized risk indicators based on temperature fluctuations and discomfort levels. These indicators serve as a valuable reference for developing infectious disease prevention strategies. Related Research Outcomes Chun-Yin Yeh and Yi-Ting Chung and Kun-Ta Chuang and Yu-Chen Shu and Hung-Yu Kao and Po-Lin Chen and Wen-Chien Ko and Nai-Ying Ko, 2021, “An innovative wearable device for monitoring continuous body surface temperature (HEARThermo): Instrument validation study”, JMIR mHealth and uHealth, 9(2), e19210 (Impact Factor 2023: 5.4) Yi-Ting Chung and Chun-Yin Yeh and Yu-Chen Shu and Kun-Ta Chuang and Chang-Chun Chen and Hung-Yu Kao and Wen-Chien Ko and Po-Lin Chen and Nai-Ying Ko, 2020, “Continuous temperature monitoring by a wearable device for early detection of febrile events in the SARS-CoV-2 outbreak in Taiwan, 2020”, Journal of Microbiology, Immunology, and Infection, 53(3), 503–504 (Impact Factor 2023: 4.5) Chun-Yin Yeh and Bing-Ze Lu and Wei-Jie Liang and Yu-Chen Shu and Kun-Ta Chuang and Po-Lin Chen and Wen-Chien Ko and Nai-Ying Ko, 2019, “Trajectories of hepatic and coagulation dysfunctions related to a rapidly fatal outcome among hospitalized patients with dengue fever in Tainan, 2015.”, PLoS Neglected Tropical Diseases, 13(12), e0007817 (Impact Factor 2023: 3.4) Hsu-Chao Lai, Po-Hsiang Fang, Yi-Ting Wu, Lo Pang-Yun Ting, Kun-Ta Chuang, "Confidence-Based Power-Efficient Framework for Sleep Stage Classification on Consumer Wearables", accepted by IEEE BigData 2024 (regular paper, acceptance rate 18.8%) Moritz Rosar and Kun-Ta Chuang, 2024, Using Blockchain for Incentive-Driven FHIR-based Health Record Exchange, Proceedings of the 39th ACM/SIGAPP Symposium on Applied Computing, SAC 2024, Avila, Spain, April 8-12, 2024, 298—299

The primary goal of the Energy Trading Platform is to facilitate Energy Trading and Demand Response (DR) by leveraging distributed ledger technology (DLT) and the immutability of blockchain data. Energy production and consumption data collected by the Building Energy Management System (BEMS) through Advanced Metering Infrastructure (AMI), along with all transaction records, are securely stored on both the Ethereum blockchain and the IOTA Tangle. Compared with traditional platforms, this solution significantly reduces hardware costs while ensuring data integrity and security. A live demonstration will be carried out in the Shalun Smart Green Energy Science City, accompanied by a proof-of-concept (POC) implementation. The platform offers two core services: Energy Trading – Enables bidding of renewable energy within the BEMS, allowing surplus electricity to be efficiently redistributed and utilized across buildings. Demand Bidding – Allows participants to engage in demand response bidding with the Taiwan Power Company (TPC) via aggregators, while securely recording electricity savings and corresponding rewards.