Introduction

We are researching computing platforms of various scales, ranging from edge devices around us to large-scale cloud computing infrastructures. Recently, we have been working on AI (Artificial Intelligence) algoritms and their hardware implementation for edge devices with limited computing resources, in-network computing using network-attached FPGAs (Field-Programmable Gate Arrays) and GPUs (Graphics Processing Units), and high-performance distributed machine learning and data processing.

■On-Device Learning Algorithms for Edge Devices
・Research Objective: Accuracy of tiny machine learning applications is often affected by various environmental factors. We introduce a neural network based on-device learning (ODL) approach to address this issue by retraining in deployed environments.
- Title: Kazuki Sunaga, Masaaki Kondo, Hiroki Matsutani, "Addressing Gap between Training Data and Deployed Environment by On-Device Learning", IEEE Micro, Special Issue on tinyML, Vol.43, No.6, pp.66-73, Nov/Dec 2023.

・Research Objective: We propose a neural network based on-device learning (ODL) approach and its hardware implementation for low-cost embedded FPGAs (Field-Programmable Gate Arrays).
- Title: Mineto Tsukada, Masaaki Kondo, Hiroki Matsutani, "A Neural Network-Based On-device Learning Anomaly Detector for Edge Devices", IEEE Transactions on Computers (TC), Vol.69, No.7, pp.1027-1044, Jul 2020.

・Research Objective: We extend a neural network based on-device learning (ODL) approach for federated learning so that edge devices can exchange their trained results and update their model by using those collected from the other edge devices.
- Title: Rei Ito, Mineto Tsukada, Hiroki Matsutani, "An On-Device Federated Learning Approach for Cooperative Model Update between Edge Devices", IEEE Access, Vol.9, pp.92986-92998, Jun 2021.

■On-Device Finetuning Algorithms for DNN/CNN Models
・Research Objective: We propose a lightweight finetuning algorithm for low-cost edge devices so that deep neural network models can be finetuned in deployed environments.
- Title: Hiroki Matsutani, Masaaki Kondo, Kazuki Sunaga, Radu Marculescu, "Skip2-LoRA: A Lightweight On-device DNN Fine-tuning Method for Low-cost Edge Devices", Proc. of the 30th Asia and South Pacific Design Automation Conference (ASP-DAC'25), pp.51-57, Jan 2025.

■FPGA-Based Accelerators for Robotics Applications
・Research Objective: Point cloud registration serves as a basis for vision and robotic applications. We propose an efficient deep-learning-based registration approach for low-cost embedded FPGAs (Field-Programmable Gate Arrays).
- Title: Keisuke Sugiura, Hiroki Matsutani, "FPGA-Accelerated Correspondence-free Point Cloud Registration with PointNet Features", ACM Transactions on Reconfigurable Technology and Systems (TRETS), Just Accepted.

・Research Objective: We focus on vision transformer models that combine CNN backbone and MHSA (Multi-Head Self-Attention). Specifically, we propose a lightweight hybrid model that uses Neural ODE (Ordinary Differential Equation) to reduce parameter size for low-cost embedded FPGAs (Field-Programmable Gate Arrays).
- Title: Ikumi Okubo, Keisuke Sugiura, Hiroki Matsutani, "A Cost-Efficient FPGA-Based CNN-Transformer using Neural ODE", IEEE Access, Vol.12, pp.155773-155788, Oct 2024.

・Research Objective: Path planning is an important component for realizing autonomous mobile robots. We propose a lightweight deep-learning-based 2D/3D path planning approach for low-cost embedded FPGAs (Field-Programmable Gate Arrays).
- Title: Keisuke Sugiura, Hiroki Matsutani, "An Integrated FPGA Accelerator for Deep Learning-based 2D/3D Path Planning", IEEE Transactions on Computers (TC), Vol.73, No.6, pp.1442-1456, Jun 2024.

・Research Objective: LiDAR (Light Detection and Ranging) SLAM (Simultaneous Localization and Mapping) serves as a basis for navigation in robotics applications. We propose a universal resource-efficient accelerator design of 2D LiDAR SLAM for low-cost embedded FPGAs (Field-Programmable Gate Arrays).
- Title: Keisuke Sugiura, Hiroki Matsutani, "A Universal LiDAR SLAM Accelerator System on Low-cost FPGA", IEEE Access, Vol.10, pp.26931-26947, Mar 2022.

■Communication Efficient Distributed Machine Learning Systems
・Research Objective: In distributed reinforcement learning, we can assume that actors are located in edge side while a learner is located in cloud side, and their long-haul communication imposes significant overheads. We propose new architecture that performs experience sampling at the edge side to improve communication efficiency.
- Title: Shin Morishima, Hiroki Matsutani, "An Efficient Distributed Reinforcement Learning Architecture for Long-haul Communication between Actors and Learner", IEEE Access, Vol.12, pp.71479-71491, May 2024.

Areas of Research

・Machine Learning
・Computer System
・On-device Learning

Social Contributions

・Advancement in Computing Efficiency
・Sustainable Technology Development
・Enhancement of Privacy and Security