Shufan Yang

Low-density lipoprotein (LDL) deposition at haemodynamically disturbed arte- rial sites drives cerebral atherosclerosis, a leading contributor to stroke. Com- putational fluid dynamics (CFD) studies of cerebral arteries commonly rely on surface-based haemodynamic descriptors such as time-averaged wall shear stress (TAWSS), oscillatory shear index (OSI) and relative residence time (RRT) to identify atheroprone regions, but such descriptors cannot resolve individual particle–wall mechanics. This study presents a one-way coupled CFD–discrete element method (DEM) framework and applies it to LDL-surrogate deposi- tion in a patient-specific middle cerebral artery bifurcation. A non-Newtonian transient CFD model, validated through mesh (1.15% TAWSS difference) and timestep (0.44% TAWSS difference) independence studies, was coupled to a DEM model incorporating Johnson–Kendall–Roberts (JKR) adhesion. Of 10,671 LDL- surrogate particles injected, 2,878 particle–wall contact events were recorded. The bifurcation returned the highest contact density (6.00 × 10^8 no./m^2), consistent with its largest low-TAWSS area fraction (30.56%). However, the right branch accounted for approximately 50% of total contact events despite low-TAWSS, OSI and RRT area fractions comparable to those of the left branch, indicat- ing that surface-based haemodynamic descriptors alone did not account for the observed deposition asymmetry. A sensitivity case using 20 μm particles con- firmed this trend, with the right branch retaining its contact dominance across a twofold change in particle diameter. The coupled CFD–DEM analysis demon- strates that LDL-surrogate deposition behaviour is governed by the combined influence of local geometry and haemodynamic conditions, providing information that is inaccessible to CFD-only analysis

Functional near-infrared spectroscopy (fNIRS) is a non-invasive neuroimaging technique that provides valuable insights into brain activity by measuring haemodynamic changes in blood oxygenation. Despite its potential, the accuracy of fNIRS-based brain image reconstruction is often compromised by motion artefacts. While conventional adaptive signal processing methods can be employed to remove these artefacts, neural networks have demonstrated a more effective alternative. However, traditional neural networks typically have substantial computational and memory requirements, making them unsuitable for resource-constrained wearable platforms. In this study, we propose an optimisation framework for neural network processing specifically designed for wearable devices to enhance the clarity and reliability of fNIRS brain images. Through systematic evaluation and integrate of various datasets on resource limited computing platform, we establish a standardised a standardised proceeding pipeline that can be applied across various fNIRS datasets. The proposed framework is validated on three datasets, demonstrating significant improvements in signal quality and image reconstruction accuracy, while achieving a 24% reduction in memory footprint optimisation. Our findings suggest that adopting a universal preprocessing optimisation strategy could standardise fNIRS data analysis for wearable devices, enabling more consistent and interpretable results across studies. This advancement contributes to the broader application of fNIRS in clinical and neurorehabilitation research, make real-time neuroimaging more feasible and effective.

Due to iterative matrix multiplications or gradient computations, machine learning modules often require a large amount of processing power and memory. As a result, they are often not feasible for use in wearable devices, which have limited processing power and memory. In this study, we propose an ultralow-power and real-time machine learning-based motion artifact detection module for functional near-infrared spectroscopy (fNIRS) systems. We achieved a high classification accuracy of 97.42%, low field-programmable gate array (FPGA) resource utilization of 38354 lookup tables and 6024 flip-flops, as well as low power consumption of 0.021 W in dynamic power. These results outperform conventional CPU support vector machine (SVM) methods and other state-of-the-art SVM implementations. This study has demonstrated that an FPGA-based fNIRS motion artifact classifier can be exploited while meeting low power and resource constraints, which are crucial in embedded hardware systems while keeping high classification accuracy.

Multi-Level Intermediate Representation (MLIR) is gaining increasing attention in reconfigurable hardware communities due to its capability to represent various abstract levels for software compilers. This project aims to be the first to provide an end-to-end framework that leverages open-source, cross-platform compilation technology to generate MLIR from SYCL. Additionally, it aims to explore a lowering pipeline that converts MLIR to RTL using open-source hardware intermediate representation (IR) and compilers. Furthermore, it aims to couple the generated hardware module with the host CPU using vendor-specific crossbars. Our preliminary results demonstrated the feasibility of lowering customized MLIR to RTL, thus paving the way for an end-to-end compilation.

Recent research has shown that large natural language processing (NLP) models have been vulnerable to a security threat called backdoor attacks. The current privately-triggered backdoor attack achieves the attack target by modifying the label of the poisoned data to a specified label. Still, the result often ignores the consistency of the sample semantics and the label, thus failing to achieve stealthy attacks on system users and system deployers. To address this issue, we use human-written text perturbations, which are less likely to be detected as machine-manipulation and removed by defense systems, as a backdoor trigger and apply a mild adversarial perturbation to the poisoned samples before implanting the backdoor, thereby addressing the semantic inconsistency caused by modifying the labels of the poisoned data. Furthermore, we use learnable backdoor triggers to improve the stealth of backdoor attacks while avoiding the conflict between textual adversarial perturbations and backdoor implantation. Experiments show that our attack achieves close to 100% attack success rates on SST-2, OLID and AG’s News datasets without affecting the utility of existing NLP models.