AI for Science
Our laboratory is built on the core technologies of computer architecture and high-performance computing, fusing them with advanced technologies such as artificial intelligence (AI), the Internet of Things (IoT), and big-data analytics, with the goal of contributing to the development of industry and society and to the preservation of cultural heritage.
In recent years, we have been actively promoting interdisciplinary research that combines AI with a wide range of application fields — for example, ancient-document restoration, anomaly detection, 4D printing, time-series forecasting, smart agriculture via edge computing, and fusion with biomedical engineering. We approach these challenges from both hardware and software, working consistently from system design through implementation and application.
AI × Edge Computing & Model Optimization
Deep learning models are highly accurate, but their enormous computation and memory requirements make them hard to deploy on edge devices. Through compression techniques such as pruning, quantization, and knowledge distillation, token reduction for Vision Transformers, and hardware/software co-design for FPGA / RISC-V, we build AI that runs fast and low-power even under tight computational constraints.
AI × Object Detection & Industrial Vision
Fast, high-precision visual recognition is essential for quality control and automation on manufacturing lines. We adapt lightweight YOLO-family object detection for industrial imagery, applying it to PCB defect detection, fruit quality grading, and dish-collecting robots. Unsupervised anomaly detection (PatchCore, memory banks, segment fusion) further builds practical systems capable of catching previously unseen defects.
AI × Cultural Heritage & Ancient Documents
A vast quantity of Japanese classical books and ancient documents remain unorganized, alongside oracle bone inscriptions roughly 3,000 years old. Using deep learning, we achieve automatic recognition of cursive Japanese and oracle bone script, restoration of degraded documents via diffusion models (DDRM), and multimodal contextual understanding — aiming to accelerate the digital transmission of cultural heritage with AI technology.
AI × Bioinformatics Analysis
Combining artificial intelligence with established bioinformatics methods, we systematically analyze single-cell multi-omics data. Through an automatic cell-type recognition model based on scRNA-seq, a multi-omics integration framework for scRNA-seq / scATAC-seq, and deep learning models that integrate single-cell data with spatial transcriptomics, we build a computational foundation contributing to multi-layered analysis of complex biological processes and the advancement of precision medicine.
AI × IoT & Robotics
We achieve real-time AI inference on edge devices with extremely limited computing resources. From fully on-board object tracking on a 29g nano UAV to automated cherry tomato harvesting with RGB-D vision and a robotic arm, and digital-twin biogas plant monitoring, we develop real-world applications that fuse AI with IoT and robotics — pursuing low-power lightweight models and system integration that works in the field.
AI × Quantum
Exploiting qubit superposition and entanglement delivers high parallel processing power for complex, high-dimensional data. Using a quantum evolutionary algorithm, we jointly optimize feature selection and network architecture parameters for deep learning models applied to medical datasets on breast cancer, Parkinson's disease, and heart disease — validated on real quantum hardware, substantially reducing redundant features while improving diagnostic accuracy.