arXiv subjects

Keisuke Sugiura

Publications and source records attributed to Keisuke Sugiura.

At least 19 recordsLinked to original sources

arXiv2025

InstantFT: An FPGA-Based Runtime Subsecond Fine-tuning of CNN Models

Training deep neural networks (DNNs) requires significantly more computation and memory than inference, making runtime adaptation of DNNs challenging on resource-limited IoT platforms. We propose InstantFT, an FPGA-based method for ultra-fast CNN fine-tuning on IoT devices, by optimizing the forward and backward computations in parameter-efficient fine-tuning (PEFT). Experiments on datasets with concept drift demonstrate that InstantFT fine-tunes a pre-trained CNN 17.4x faster than existing Low-Rank Adaptation (LoRA)-based approaches, while achieving comparable accuracy. Our FPGA-based InstantFT reduces the fine-tuning time to just 0.36s and improves energy-efficiency by 16.3x, enabling on-the-fly adaptation of CNNs to non-stationary data distributions.

cs.LG↗

arXiv2025

PointODE: Lightweight Point Cloud Learning with Neural Ordinary Differential Equations on Edge

Embedded edge devices are often used as a computing platform to run real-world point cloud applications, but recent deep learning-based methods may not fit on such devices due to limited resources. In this paper, we aim to fill this gap by introducing PointODE, a parameter-efficient ResNet-like architecture for point cloud feature extraction based on a stack of MLP blocks with residual connections. We leverage Neural ODE (Ordinary Differential Equation), a continuous-depth version of ResNet originally developed for modeling the dynamics of continuous-time systems, to compress PointODE by reusing the same parameters across MLP blocks. The point-wise normalization is proposed for PointODE to handle the non-uniform distribution of feature points. We introduce PointODE-Elite as a lightweight version with 0.58M trainable parameters and design its dedicated accelerator for embedded FPGAs. The accelerator consists of a four-stage pipeline to parallelize the feature extraction for multiple points and stores the entire parameters on-chip to eliminate most of the off-chip data transfers. Compared to the ARM Cortex-A53 CPU, the accelerator implemented on a Xilinx ZCU104 board speeds up the feature extraction by 4.9x, leading to 3.7x faster inference and 3.5x better energy-efficiency. Despite the simple architecture, PointODE-Elite shows competitive accuracy to the state-of-the-art models on both synthetic and real-world classification datasets, greatly improving the trade-off between accuracy and inference cost.

cs.LG↗

arXiv2025

ElasticZO: A Memory-Efficient On-Device Learning with Combined Zeroth- and First-Order Optimization

Zeroth-order (ZO) optimization is being recognized as a simple yet powerful alternative to standard backpropagation (BP)-based training. Notably, ZO optimization allows for training with only forward passes and (almost) the same memory as inference, making it well-suited for edge devices with limited computing and memory resources. In this paper, we propose ZO-based on-device learning (ODL) methods for full-precision and 8-bit quantized deep neural networks (DNNs), namely ElasticZO and ElasticZO-INT8. ElasticZO lies in the middle between pure ZO- and pure BP-based approaches, and is based on the idea to employ BP for the last few layers and ZO for the remaining layers. ElasticZO-INT8 achieves integer arithmetic-only ZO-based training for the first time, by incorporating a novel method for computing quantized ZO gradients from integer cross-entropy loss values. Experimental results on the classification datasets show that ElasticZO effectively addresses the slow convergence of vanilla ZO and shrinks the accuracy gap to BP-based training. Compared to vanilla ZO, ElasticZO achieves 5.2-9.5% higher accuracy with only 0.072-1.7% memory overhead, and can handle fine-tuning tasks as well as full training. ElasticZO-INT8 further reduces the memory usage and training time by 1.46-1.60x and 1.38-1.42x without compromising the accuracy. These results demonstrate a better tradeoff between accuracy and training cost compared to pure ZO- and BP-based approaches, and also highlight the potential of ZO optimization in on-device learning.

cs.LG↗

arXiv2024

A Cost-Efficient FPGA Implementation of Tiny Transformer Model using Neural ODE

Transformer has been adopted to image recognition tasks and shown to outperform CNNs and RNNs while it suffers from high training cost and computational complexity. To address these issues, a hybrid approach has become a recent research trend, which replaces a part of ResNet with an MHSA (Multi-Head Self-Attention). In this paper, we propose a lightweight hybrid model which uses Neural ODE (Ordinary Differential Equation) as a backbone instead of ResNet so that we can increase the number of iterations of building blocks while reusing the same parameters, mitigating the increase in parameter size per iteration. The proposed model is deployed on a modest-sized FPGA device for edge computing. The model is further quantized by QAT (Quantization Aware Training) scheme to reduce FPGA resource utilization while suppressing the accuracy loss. The quantized model achieves 79.68% top-1 accuracy for STL10 dataset that contains 96$\times$96 pixel images. The weights of the feature extraction network are stored on-chip to minimize the memory transfer overhead, allowing faster inference. By eliminating the overhead of memory transfers, inference can be executed seamlessly, leading to accelerated inference. The proposed FPGA implementation accelerates the backbone and MHSA parts by 34.01$\times$, and achieves an overall 9.85$\times$ speedup when taking into account the software pre- and post-processing. The FPGA acceleration leads to 7.10$\times$ better energy efficiency compared to the ARM Cortex-A53 CPU. The proposed lightweight Transformer model is demonstrated on Xilinx ZCU104 board for the image recognition of 96$\times$96 pixel images in this paper and can be applied to different image sizes by modifying the pre-processing layer.

cs.LG↗

arXiv2024

An FPGA-Based Accelerator for Graph Embedding using Sequential Training Algorithm

A graph embedding is an emerging approach that can represent a graph structure with a fixed-length low-dimensional vector. node2vec is a well-known algorithm to obtain such a graph embedding by sampling neighboring nodes on a given graph with a random walk technique. However, the original node2vec algorithm typically relies on a batch training of graph structures; thus, it is not suited for applications in which the graph structure changes after the deployment. In this paper, we focus on node2vec applications for IoT (Internet of Things) environments. To handle the changes of graph structures after the IoT devices have been deployed in edge environments, in this paper we propose to combine an online sequential training algorithm with node2vec. The proposed sequentially-trainable model is implemented on an FPGA (Field-Programmable Gate Array) device to demonstrate the benefits of our approach. The proposed FPGA implementation achieves up to 205.25 times speedup compared to the original model on ARM Cortex-A53 CPU. Evaluation results using dynamic graphs show that although the accuracy is decreased in the original model, the proposed sequential model can obtain better graph embedding that achieves a higher accuracy even when the graph structure is changed.

cs.LG↗

arXiv2024

FPGA-Accelerated Correspondence-free Point Cloud Registration with PointNet Features

Point cloud registration serves as a basis for vision and robotic applications including 3D reconstruction and mapping. Despite significant improvements on the quality of results, recent deep learning approaches are computationally expensive and power-hungry, making them difficult to deploy on resource-constrained edge devices. To tackle this problem, in this paper, we propose a fast, accurate, and robust registration for low-cost embedded FPGAs. Based on a parallel and pipelined PointNet feature extractor, we develop custom accelerator cores namely PointLKCore and ReAgentCore, for two different learning-based methods. They are both correspondence-free and computationally efficient as they avoid the costly feature matching step involving nearest-neighbor search. The proposed cores are implemented on the Xilinx ZCU104 board and evaluated using both synthetic and real-world datasets, showing the substantial improvements in the trade-offs between runtime and registration quality. They run 44.08-45.75x faster than ARM Cortex-A53 CPU and offer 1.98-11.13x speedups over Intel Xeon CPU and Nvidia Jetson boards, while consuming less than 1W and achieving 163.11-213.58x energy-efficiency compared to Nvidia GeForce GPU. The proposed cores are more robust to noise and large initial misalignments than the classical methods and quickly find reasonable solutions in less than 15ms, demonstrating the real-time performance.

cs.RO↗

arXiv2023

An Integrated FPGA Accelerator for Deep Learning-based 2D/3D Path Planning

Path planning is a crucial component for realizing the autonomy of mobile robots. However, due to limited computational resources on mobile robots, it remains challenging to deploy state-of-the-art methods and achieve real-time performance. To address this, we propose P3Net (PointNet-based Path Planning Networks), a lightweight deep-learning-based method for 2D/3D path planning, and design an IP core (P3NetCore) targeting FPGA SoCs (Xilinx ZCU104). P3Net improves the algorithm and model architecture of the recently-proposed MPNet. P3Net employs an encoder with a PointNet backbone and a lightweight planning network in order to extract robust point cloud features and sample path points from a promising region. P3NetCore is comprised of the fully-pipelined point cloud encoder, batched bidirectional path planner, and parallel collision checker, to cover most part of the algorithm. On the 2D (3D) datasets, P3Net with the IP core runs 24.54-149.57x and 6.19-115.25x (10.03-59.47x and 3.38-28.76x) faster than ARM Cortex CPU and Nvidia Jetson while only consuming 0.255W (0.809W), and is up to 1049.42x (133.84x) power-efficient than the workstation. P3Net improves the success rate by up to 28.2% and plans a near-optimal path, leading to a significantly better tradeoff between computation and solution quality than MPNet and the state-of-the-art sampling-based methods.

cs.RO↗

arXiv2023

A Low-Cost Neural ODE with Depthwise Separable Convolution for Edge Domain Adaptation on FPGAs

High-performance deep neural network (DNN)-based systems are in high demand in edge environments. Due to its high computational complexity, it is challenging to deploy DNNs on edge devices with strict limitations on computational resources. In this paper, we derive a compact while highly-accurate DNN model, termed dsODENet, by combining recently-proposed parameter reduction techniques: Neural ODE (Ordinary Differential Equation) and DSC (Depthwise Separable Convolution). Neural ODE exploits a similarity between ResNet and ODE, and shares most of weight parameters among multiple layers, which greatly reduces the memory consumption. We apply dsODENet to a domain adaptation as a practical use case with image classification datasets. We also propose a resource-efficient FPGA-based design for dsODENet, where all the parameters and feature maps except for pre- and post-processing layers can be mapped onto on-chip memories. It is implemented on Xilinx ZCU104 board and evaluated in terms of domain adaptation accuracy, inference speed, FPGA resource utilization, and speedup rate compared to a software counterpart. The results demonstrate that dsODENet achieves comparable or slightly better domain adaptation accuracy compared to our baseline Neural ODE implementation, while the total parameter size without pre- and post-processing layers is reduced by 54.2% to 79.8%. Our FPGA implementation accelerates the inference speed by 23.8 times.

cs.LG↗

arXiv2022

Formation of moons and equatorial ridge around top-shaped asteroids after surface landslide

Top-shaped asteroids have been observed among near-Earth asteroids. About half of them are reported to have moons (on the order of $\sim 1$wt.\% of the top-shaped primary) and many of them have an equatorial ridge. A recent study has shown that the enigmatic top-shaped figure of asteroids (e.g., Ryugu, Bennu, and Didymos) could result from an axisymmetric landslide of the primary during a fast spin-up near the breakup rotation period. Such a landslide would inevitably form a particulate disk around an asteroid with a short timescale ($\sim 3$ hours). However, the long-term full dynamical evolution is not investigated. Here, we perform a continuous simulation ($\sim 700$ hours) that investigates the sequence of events from the surface landslide that forms a top-shaped asteroid and a particulate disk to disk evolution. We show that the disk quickly spreads and produces moons (within $\sim 300$ hours). The mass of the formed moon is consistent with what is observed around the top-shaped asteroids. We also demonstrate that an equatorial ridge is naturally formed because a fraction of the disk particles re-accretes selectively onto the equatorial region of the primary. We envision that Ryugu and Bennu could once have an ancient moon that was later lost due to a successive moon's orbital evolution. Alternatively, at the top-shaped asteroid that has a moon, such as Didymos, no significant orbital evolution of the moon has occurred that would result in its loss. Our study would also be qualitatively applicable to any rubble-pile asteroids near the breakup rotation period.

astro-ph.EP↗

arXiv2022

An Efficient Accelerator for Deep Learning-based Point Cloud Registration on FPGAs

Point cloud registration is the basis for many robotic applications such as odometry and Simultaneous Localization And Mapping (SLAM), which are increasingly important for autonomous mobile robots. Computational resources and power budgets are limited on these robots, thereby motivating the development of resource-efficient registration method on low-cost FPGAs. In this paper, we propose a novel approach for FPGA-based 3D point cloud registration built upon a recent deep learning-based method, PointNetLK. A highly-efficient FPGA accelerator for PointNet-based feature extraction is designed and implemented on both low-cost and mid-range FPGAs (Avnet Ultra96v2 and Xilinx ZCU104). Our accelerator design is evaluated in terms of registration speed, accuracy, resource usage, and power consumption. Experimental results show that PointNetLK with our accelerator achieves up to 21.34x and 69.60x faster registration speed than the CPU counterpart and ICP, respectively, while only consuming 722mW and maintaining the same level of accuracy.

cs.RO↗

arXiv2022

Giant Impact onto a Vesta-Like Asteroid and Formation of Mesosiderites through Mixing of Metallic Core and Surface Crust

Mesosiderites are a type of stony-iron meteorites composed of a mixture of silicates and Fe-Ni metals. The mesosiderite silicates and metals are considered to have originated from the crust and metal core, respectively, of a differentiated asteroid. In contrast, mesosiderites rarely contain the olivine that is mainly included in a mantle. Although a giant impact onto a differentiated asteroid is considered to be a probable mechanism to mix crust and metal materials to form mesosiderites, it is not obvious how such a giant impact can form mesosiderite-like materials without including mantle materials. We conducted numerical simulations of giant impacts onto differentiated asteroids, using the smoothed particle hydrodynamics method, to investigate the detailed distribution of mixed materials on the resultant bodies. For the internal structure of a target body, we used a thin-crust model derived from the magma ocean crystallization model of the asteroid Vesta and a thick-crust and a large-core model suggested from the proximity observation of Vesta by the Dawn probe. In the simulations with the former model, excavation of the metal core requires nearly catastrophic impacts and mantle is exposed over large surface areas. Thus, stony-iron materials produced on its surface are likely to include mantle materials and it is difficult to produce mesosiderite-like materials. Conversely, in the simulations with the latter model, mantle materials are exposed only at impact sites, even when the impacts excavate the metal core, and the formation of a surface with little mantle material and the formation of mesosiderite-like materials are possible. Therefore, our simulations suggest that an internal structure with a thick crust and a large core is more likely as a mesosiderite parent body rather than the thin-crust internal structure inferred from the conventional magma ocean model.

astro-ph.EP↗

arXiv2021

A Universal LiDAR SLAM Accelerator System on Low-cost FPGA

LiDAR (Light Detection and Ranging) SLAM (Simultaneous Localization and Mapping) serves as a basis for indoor cleaning, navigation, and many other useful applications in both industry and household. From a series of LiDAR scans, it constructs an accurate, globally consistent model of the environment and estimates a robot position inside it. SLAM is inherently computationally intensive; it is a challenging problem to realize a fast and reliable SLAM system on mobile robots with a limited processing capability. To overcome such hurdles, in this paper, we propose a universal, low-power, and resource-efficient accelerator design for 2D LiDAR SLAM targeting resource-limited FPGAs. As scan matching is at the heart of SLAM, the proposed accelerator consists of dedicated scan matching cores on the programmable logic part, and provides software interfaces to facilitate the use. Our accelerator can be integrated to various SLAM methods including the ROS (Robot Operating System)-based ones, and users can switch to a different method without modifying and re-synthesizing the logic part. We integrate the accelerator into three widely-used methods, i.e., scan matching, particle filter, and graph-based SLAM. We evaluate the design in terms of resource utilization, speed, and quality of output results using real-world datasets. Experiment results on a Pynq-Z2 board demonstrate that our design accelerates scan matching and loop-closure detection tasks by up to 14.84x and 18.92x, yielding 4.67x, 4.00x, and 4.06x overall performance improvement in the above methods, respectively. Our design enables the real-time performance while consuming only 2.4W and maintaining accuracy, which is comparable to the software counterparts and even the state-of-the-art methods.

cs.RO↗

arXiv2021

SPH Simulations for Shape Deformation of Rubble-Pile Asteroids Through Spinup: The Challenge for Making Top-Shaped Asteroids Ryugu and Bennu

Asteroid Ryugu and asteroid Bennu, which were recently visited by spacecraft Hayabusa2 and OSIRIS-REx, respectively, are spinning top-shaped rubble piles. Other axisymmetric top-shaped near-Earth asteroids have been observed with ground-based radar, most of which rotate near breakup rotation periods of ~ 3 hours. This suggests that rotation-induced deformation of asteroids through rotational spinup produces top shapes. Although some previous simulations using the Discrete Element Method showed that spinup of rubble piles may produce oblate top shapes, it is still unclear what kinds of conditions such as friction angles of constituent materials and spinup timescales are required for top-shape formation. Here we show, through Smoothed Particle Hydrodynamics simulations of granular bodies spinning-up at different rates, that the rotation-induced deformation of spherical rubble piles before breakup can be classified into three modes according to the friction angle ϕ_{d}: quasi-static and internal deformation for ϕ_{d} < 40 degrees, dynamical and internal deformation for 50 degrees < ϕ_{d} < 60 degrees, and surface landslides for ϕ_{d} > 70 degrees. Note that these apparent large values of friction angle can be acceptable if we consider the effect of cohesion among blocks of a rubble pile under weak gravity. Bodies with ϕ_{d} < 60 degrees evolve into oblate spheroids through internal deformation, but never form pronounced equators defining a top shape. In contrast, bodies with ϕ_{d} > 70 degrees deform into axisymmetric top shapes through an axisymmetric surface landslides if spinup timescales are < a few days. In addition, through slow spinups with timescales > 1 month, bodies with ϕ_{d} > 70 degrees deform into non-axisymmetric shapes via localized landslides. We suggest that rapid spinup mechanisms are preferable for the formation of axisymmetric top shapes.

astro-ph.EP↗

arXiv2020

An FPGA Acceleration and Optimization Techniques for 2D LiDAR SLAM Algorithm

An efficient hardware implementation for Simultaneous Localization and Mapping (SLAM) methods is of necessity for mobile autonomous robots with limited computational resources. In this paper, we propose a resource-efficient FPGA implementation for accelerating scan matching computations, which typically cause a major bottleneck in 2D LiDAR SLAM methods. Scan matching is a process of correcting a robot pose by aligning the latest LiDAR measurements with an occupancy grid map, which encodes the information about the surrounding environment. We exploit an inherent parallelism in the Rao-Blackwellized Particle Filter (RBPF) based algorithms to perform scan matching computations for multiple particles in parallel. In the proposed design, several techniques are employed to reduce the resource utilization and to achieve the maximum throughput. Experimental results using the benchmark datasets show that the scan matching is accelerated by 5.31-8.75x and the overall throughput is improved by 3.72-5.10x without seriously degrading the quality of the final outputs. Furthermore, our proposed IP core requires only 44% of the total resources available in the TUL Pynq-Z2 FPGA board, thus facilitating the realization of SLAM applications on indoor mobile robots.

eess.SP↗

arXiv2019

High-Resolution Simulations of Catastrophic Disruptions: Resultant Shape Distributions

The members of asteroid families have various shapes. We investigate the origin of their shapes by high-resolution impact simulations for catastrophic disruptions using a Smoothed Particle Hydrodynamics code. Collisional remnants produced through our simulations of the catastrophic disruptions mainly have spherical or bilobed shapes. However, no flat remnants with the ratio of minor to major axis lengths c/a < 0.5 are formed. The results of the simulations provide various shapes of asteroids and explain most of the shapes in asteroid families that are supposed to be produced through catastrophic disruptions. However, the present simulations do not explain significantly flat asteroids. We suggest that these flat asteroids may be interlopers or formed through low-velocity collisions between member asteroids.

astro-ph.EP↗

arXiv2019

Collisional Elongation: Possible Origin of Extremely Elongated Shape of 1I/`Oumuamua

Light curve observations of a recently discovered interstellar object 1I/`Oumuamua suggest that this object has an extremely elongated shape with the axis ratio 0.3 or smaller. Planetesimal collisions can produce irregular shapes including elongated shapes. In this paper, we suggest that the extremely elongated shape of 1I/`Oumuamua may be the result of such an impact. To find detailed impact conditions to form the extremely elongated objects, we conduct numerical simulations of planetesimal collisions using Smoothed Particle Hydrodynamics method for elastic dynamics with self-gravity and interparticle friction. Impacts into strengthless target planetesimals with radius 50 m are conducted with various ratios of impactor mass to target mass q, friction angles phi_d, impact velocities v_imp, and impact angles theta_imp. We find that impacts with q \geq 0.5, phi_d \geq 40 degrees, v_imp \leq 40 degrees, and theta_imp \leq 30 degrees produce remnants with the ratio of intermediate to major axis length less than 0.3. This impact condition suggests that the parent protoplanetary disk in the planetesimal collision stage was weakly turbulent (alpha < 10^{-4} for the inner disk) and composed of planetesimals smaller than ~ 7 km to ensure small impact velocity.

astro-ph.EP↗

arXiv2018

Toward Understanding the Origin of Asteroid Geometries: Variety in Shapes Produced by Equal-Mass Impacts

More than a half of asteroids in the main belt have irregular shapes with the ratios of the minor to major axis lengths less than 0.6. One of the mechanisms to create such shapes is collisions between asteroids. The relationship between shapes of collisional outcomes and impact conditions such as impact velocities may provide information on the collisional environments and its evolutionary stages when those asteroids are created. In this study, we perform numerical simulations of collisional destruction of asteroids with radii 50 km and subsequent gravitational reaccumulation using Smoothed Particle Hydrodynamics for elastic dynamics with self-gravity, a model of fracture of rock, and a model of friction of completely damaged rock. We systematically vary the impact velocity from 50 m/s to 400 m/s and the impact angle from 5 degrees to 45 degrees. We investigate shapes of the largest remnants resulting from collisional simulations. As a result, various shapes (bilobed, spherical, flat, elongated, and hemispherical shapes) are formed through equal-mass and low-velocity (50 - 400 m/s) impacts. We clarify a range of the impact angle and velocity to form each shape. Our results indicate that irregular shapes, especially flat shapes, of asteroids with diameters larger than 80 km are likely to be formed through similar-mass and low-velocity impacts, which are likely to occur in the primordial environment prior to the formation of Jupiter.

astro-ph.EP↗

arXiv2016

An Extension of Godunov SPH II: Application to Elastic Dynamics

Godunov Smoothed Particle Hydrodynamics (Godunov SPH) method is a computational fluid dynamics method that utilizes a Riemann solver and achieves the second-order accuracy in space. In this paper, we extend the Godunov SPH method to elastic dynamics by incorporating deviatoric stress tensor that represents the stress for shear deformation or anisotropic compression. Analogously to the formulation of the original Godunov SPH method, we formulate the equation of motion, the equation of energy, and the time evolution equation of deviatoric stress tensor so that the resulting discretized system achieves the second-order accuracy in space. The standard SPH method tends to suffer from the tensile instability that results in unphysical clustering of particles especially in tension-dominated region. We find that the tensile instability can be suppressed by selecting appropriate interpolation for density distribution in the equation of motion for the Godunov SPH method even in the case of elastic dynamics. Several test calculations for elastic dynamics are performed, and the accuracy and versatility of the present method are shown.

physics.comp-ph↗