Electronic Systems Design (Malta)

Vision Transformers (ViTs) have outperformed traditional Convolutional Neural Network architectures and achieved state-of-the-art results in various computer vision tasks. Since ViTs are computationally expensive, the models either have to be pruned to run on resource-limited edge devices only or have to be executed on remote cloud servers after receiving the raw data transmitted over fluctuating networks. The resulting degraded performance or high latency all hinder their widespread applications. In this paper, we present Janus, the first framework for low-latency cloud-device collaborative Vision Transformer inference over dynamic networks. Janus overcomes the intrinsic model limitations of ViTs and realizes collaboratively executing ViT models on both cloud and edge devices, achieving low latency, high accuracy, and low communication overhead. Specifically, Janus judiciously combines token pruning techniques with a carefully designed fine-to-coarse model splitting policy and non-static mixed pruning policy. It attains a balance between accuracy and latency by dynamically selecting the optimal pruning level and split point. Experimental results across various tasks demonstrate that Janus enhances throughput by up to 5.15x and reduces latency violation ratios by up to 98.7% when compared with baseline approaches under various network environments.

Synthetic data, generated using generative AI techniques, closely mimics the characteristics of real data while enhancing privacy for sensitive health data. This study evaluates synthetic tabular data based on fidelity and utility for predictive models. Fidelity is measured through univariate distribution and bivariate differential pairwise correlations, while utility is measured by comparing machine learning model performance trained on synthetic and real data. Results show highly similar model performance on synthetic and real data. We also explore the potential of using synthetic data for hyperparameter tuning. Our findings reveal a strong correlation between prediction accuracy on synthetic and real data, suggesting that hyperparameters optimized using synthetic data can be effectively applied to models trained on real datasets for optimal results.

Oil spills severely damage marine life and coastal environments. To reduce their polluting effect on the ecosystem, it is important to promptly react to potential spills for early detection and monitoring. In this paper, we propose a drone-based solution with a deep-learning U-net model. It processes the radar backscattering dominated by the specular component in calm ocean conditions to detect contaminated sea surfaces with oil spills. Results show that our approach achieves a high detection rate exceeding 90% for thick oil slicks in the range of 1-10 mm.

Focusing on design education program related to the Sustainable Development Goals (SDGs) for Japanese high school students, this research aims to develop a set of opensource tools that can be used to visualize the design thinking educational processes used by students. To implement a new type of SDGs related design educational program that promotes solving the world's issues in environmental and social changes, educational tools to support students in design thinking processes are not available. In collaboration with teachers, we conducted a workshop with four sessions at Fukusho High School in Fukuoka City for Year 1 students under the Global Management Program. We worked with teachers to understand the design thinking processes used in the workshop and identify issues that students faced during the workshop. Through the findings, a set of visualization tools can be developed and proposed to support the students in the design thinking processes in future workshops. The visualization tool will enable students to learn freely and develop a high awareness of social issues. Also, teachers can use and adapt the tools to meet the needs of the educational objectives. Besides, as an open-source, the outputs of the research can easily be contributed back to society.

A more accurate and more efficient grinding method for optical glasses is demanded as high-resolution imaging devices become more popular. We propose a reaction-induced-slurry-assisted (RISA) grinding method as a chemical mechanical grinding method using cerium oxide slurry instead of ordinary coolant. In this study, the grinding performance of RISA grinding for BK7 optical glass is experimentally evaluated from the viewpoints of sub-surface damage and repeatability of ductile regime grinding at various wheel feed rates. From the grinding tests, RISA grinding causes less sub-surface damages and successfully generates ductile regime surface repeatedly under 3.0 mm/min wheel feed rate.

A numerical implementation of the conductor sub-division method to estimate buried cable earth-impedances is developed in this paper. The skin-effect of the resistive and inductive frequency varying components is taken into account. The Finite Element Method and the exact Pollaczek analytical formulation are also implemented to compare and validate the here obtained results for an underground cable test system, respectively. As a result of this numerical exploration, we have evidenced the relation between the ground-return paths of the current with respect to the frequency dependent space-grid numerical solution for an underground cable power system.

A high-performance perovskite/polymer piezoelectric nanogenerator for next generation self-powered wireless micro/nanodevices.

We report the fabrication of a solution-processed n-type Thin Film Transistor (TFT) with current on/off ratios of 104, a turn-on voltage (VON) of 1.2 V and a threshold voltage (VT) of 6.2 V. The TFT incorporates an insoluble and intractable dielectric layer (k = 7–9) prepared in situ from solution-processed and then photopolymerised ligand-stabilised, inorganic/organic TiO2 nanorods. A solution processed zinc oxide (ZnO) layer acts as the semiconductor. The new surface-modified TiO2 nanorods were synthesised using a ligand replacement process with a monolayer coating of photopolymerisable 10-undecynylphosphonic acid (10UCYPA) to render them both soluble in common organic solvents and be photopolymerisable using UV-illumination after having been deposited on substrate surfaces from solution and drying.

Parallel turning technology has been attracting attention as an important technology to enhance the productivity of multitasking machine tools. To maximize the productivity advantage of parallel turning, chatter avoidance or suppression is one of the most noteworthy concerns. In this study, a novel chatter suppression technique using tool swing motion is provided by a feed drive system. The optimal design methodology of the tool swing motion for effective chatter suppression is also introduced based on its analogy with the spindle speed variation technique under the shared-surface parallel turning and rigid-tool and flexible-workpiece assumptions. The proposed method was evaluated with regard to the chatter stabilizing performance and workpiece runout as compared to conventional equal pitch turning and unequal pitch turning for chatter suppression. As a result, the proposed tool swing parallel turning exhibited a high chatter stabilizing performance without eccentricity of the workpiece and enhanced surface quality, although particular swing marks were left on the machined surface.