Singapore University of Technology and Design

In comparison to traditional vehicular edge computing (VEC) architectures, the air-ground integrated vehicular edge computing (AGI-VEC) network, which combines unmanned aerial vehicles (UAV) and high-altitude platforms (HAP), offers significant advantages. These include seamless coverage, long-distance transmission, reduced latency, enhanced throughput, and alleviation of network congestion, all of which contribute to a more efficient service experience for the Internet of vehicles. To minimize the task offloading delay, this paper investigates a joint multi-computation equipment selection, resource allocation, and UAV trajectory design problem based on energy harvesting in the AGI-VEC network. In order to solve this problem, we first reformulate the problem into a Markov decision process (MDP). And then, we propose a hybrid dual-clip multi-agent proximal policy optimization (DC-MAPPO) algorithm based on the multi-agent proximal policy optimization (MAPPO) algorithm, which introduces the idea of dual-clip (DC), and the adaptive discount factor to enhance the stability and convergence speed of the algorithm. Simulation results demonstrate that the proposed algorithm outperforms other baseline algorithms in reducing task offloading delay.

Large language models (LLMs) have rapidly advanced, but their growing compute demands limit accessibility in under-resourced regions like Southeast Asia (SEA). While hybrid architectures combining Attention and State-Space Models (SSMs) offer efficiency gains, most rely on sequential interleaving, leaving the potential of parallel-head mixing largely under-explored. However, the recent Falcon-H1 family of models has demonstrated that parallel-head hybrid architectures are not only viable, but scalable to state-of-the-art levels. I propose investigating this parallel-head architecture as a foundation for efficient, multilingual SEA LLMs. My short-term goal is to adapt Falcon-H1-1.5B via vocabulary expansion and continuous pretraining, mitigating token fragmentation and enabling low-resource adaptation to 9 SEA languages. In the longer term, I will develop a dynamic token routing mechanism to optimize token-level compute allocation within hybrid layers, aiming to maximize efficiency without sacrificing the expressive power needed for complex multilingual contexts. Evaluation will utilize the SEA-HELM framework to assess whether these parallel-hybrid innovations can democratize access to high-performance AI for SEA communities.

Abstract This work focuses on the validation of attack pattern mining in the context of Industrial Control System (ICS) security. A comprehensive security assessment of an ICS requires the generation of a variety of attack patterns. For this purpose, we have proposed a data-driven technique to generate attack patterns for an ICS. The proposed technique has been used to generate 117,960 attack patterns from data from a water treatment plant. These attack patterns were used to launch attacks on the operational testbed that typically lasted 2 to 4 minutes. Interestingly, some 2-minute attacks impacted the plant, while some attacks of 3 or 4 minutes duration had no observable effect. This suggests that even short-lived attacks can significantly impact operational plants. The proposed technique and the effectiveness of the patterns generated in moving the plant to an anomalous state are valuable when assessing the quality of Intrusion Detection Systems for physical plants. In this work, we present a detailed case study to validate the attack patterns.

The rapid advancement of photonics is driven by growing need for high-capacity transmission in AI and data-center interconnects. Silicon photonics enables low-cost CMOS fabrication, but conventional materials lack second-order nonlinearity and rely on laser-diode arrays for multi-channel links, increasing cost and complexity. A frequency comb provides a single multi-wavelength source that simplifies the system. Meanwhile, aluminum nitride (AlN) is a CMOS-compatible material with wide transparency and significant second-order nonlinearity. Here, we demonstrate high-speed transmission using a frequency comb generated from a high-Q (>1×10⁶) AlN microring on an 8-inch CMOS-compatible platform. A pump laser is amplified, coupled into the resonator to generate comb, and a selected comb line is filtered, reamplified, and intensity modulated. Using a Mach–Zehnder modulator, we transmit 29-GBaud PAM4 and 30-Gbps NRZ signals over 10 km of fiber. The obtained BERs of 10⁻⁸ (PAM4) and 10⁻⁶ (NRZ) surpasses typical FEC limits. Open eye diagrams have high similarity to back-to-back results, demonstrating strong potential of AlN-based combs for scalable cost-effective high-speed communication.

• LiDAR-based human tracking for an autonomous robot. • Transforming 3D LiDAR pointcloud to 2D depth image. • Posture recognition using 2D depth image. • Robot operation based on different human postures. Autonomous robots are becoming more and more important in human’ daily life. Human-Robot Interaction (HRI) addresses the challenges of limited robot autonomy in dynamic settings. In human-robot interaction, human-following behavior is a key capability for robots supporting applications such as service robotics, healthcare assistance, and cleaning tasks. Most of the human-following and interaction frameworks are based on pure vision methods, which have limitations such as the range of sensing and different environmental conditions. This paper proposes a novel human-following framework for autonomous cleaning robots using a 3D Light Detection and Ranging (LiDAR) sensor. The proposed framework enables real-time human tracking and following based on point cloud data, complemented by human posture recognition using depth images. Unlike traditional vision-based approaches, this system overcomes challenges in limited perception range and robustness. By recognizing human postures, the robot can adapt its actions to improve task efficiency and responsiveness. Experimental results demonstrate the framework’s enhanced tracking range, robustness in diverse scenarios, and faster posture recognition processing. This study contributes to the advancement of intelligent cleaning robots with improved coordination, flexibility, and interaction capabilities.

Generative AI (GenAI), exemplified by Large Language Models (LLMs), such as OpenAI’s ChatGPT, is revolutionizing various fields. Central to this transformation is Data Center Networking (DCN), which not only provides the infrastructure support for GenAI operation, but also provisions GenAI services to users. Hence, this article explores the interplay between GenAI and DCNs, analyzing their symbiotic relationship and mutual advances. We begin by reviewing the current challenges of DCNs and GenAI-based solutions, such as data augmentation, process automation, and domain transfer. We then discuss the distinctive characteristics of GenAI workloads on DCNs, gaining insights that catalyze the evolution of DCNs to more effectively support GenAI. Moreover, to illustrate the seamless integration of GenAI with DCNs, we present a case study on GenAI-empowered DCN digital twins. Specifically, we employ an LLM equipped with retrieval augmented generation to formulate optimization problems for DCNs (e.g., resource allocation and routing) and adopt diffusion-deep reinforcement learning to solve optimization. The experimental results on a representative DCN optimization problem, i.e., knowledge placement, demonstrate the validity and efficiency of our proposals. We anticipate that this article can promote further research to enhance the virtuous interaction between GenAI and DCNs.

Flexible and wearable surface acoustic wave (SAW) strain sensors, leveraging their inherent wireless and passive operation, have emerged as a transformative platform for on-body biomechanical monitoring. This capability is pivotal because key physiological stimuli such as pressure, force, and tissue deformation often manifest as measurable surface strain, making strain sensing a fundamental modality for non-invasive health assessment. This article systematically reviews recent advances in this field, with a specific focus on design principles, material systems, fabrication techniques, and system integration strategies that address the dual demands of high sensing performance and conformal wearability. It critically examines innovative strategies to address key challenges such as mechanical mismatch with soft tissues, temperature cross-sensitivity, motion artifacts, and signal attenuation in dynamic environments, including dual-channel differential designs, the adoption of flexible and stretchable materials, and intelligent signal processing algorithms. Furthermore, this review provides a critical assessment of barriers (e.g., long-term reliability, under deformation, biocompatibility, and system-level power efficiency) and outlines future research directions, including the development of intelligent systems, multi-modal sensing, and digital twins. By combining recent global research findings, this work serves as a comprehensive reference for developing next-generation, self-sustained wearable sensing systems for personalized biomechanical monitoring.

Organ-on-a-chip (OoC) technology is a powerful tool for creating physiologically relevant microscale models applicable to biomedical studies. Despite the advances in OoC technology, its fabrication method still primarily relies on soft lithography, which has a long design-to-prototype cycle that creates a bottleneck in the acceleration of OoC innovation. To increase the agility of the OoC fabrication process, a system is developed to fabricate OoC using a digital photomask aligned with a microchamber. The approach uses a pre-defined microfluidic chamber customized by xurography and microfluidic channels photopatterned by a digital photomask. The versatility of the approach offered previously unattainable features in the fabrication of OoC, including non-reticular height profiles of the OoC architecture, and real-time modification of channel designs to trap suspension culture (e.g., spheroids). In summary, this work highlights a versatile system to fabricate OoC by direct photopatterning that can accommodate various OoC design requirements of microenvironments of specific organ tissues. It is anticipated that the system can facilitate the rapid fabrication of OoC, potentially supporting advancements in OoC design innovation, which can potentially increase the adoption of the OoC technology for therapeutic screening and elucidation of disease mechanisms in the scientific community.