Federated learning (FL) performs collaborative training of deep learning models among multiple clients, safeguarding data privacy, security, and legal adherence by preserving training data locally. Despite the benefits of FL, its wider implementation is hindered by communication overheads and potential privacy risks. Transiting locally updated model parameters between edge clients and servers demands high communication bandwidth, leading to high latency and Internet infrastructure constraints. Furthermore, recent works have shown that the malicious server can reconstruct clients’ training data from gradients, significantly escalating privacy threats and violating regularizations. Different defense techniques have been proposed to address this information leakage from the gradient or updates, including introducing noise to gradients, performing model compression (such as sparsification), and feature perturbation. However, these methods either impede model convergence or entail substantial communication costs, further exacerbating the communication demands in FL. To develop an efficient and privacy-preserving FL, we introduce an autoencoder-based method for compressing and, thus, perturbing the model parameters. The client utilizes an autoencoder to acquire the representation of the local model parameters and then shares it as the compressed model parameters with the server, rather than the true model parameters. The use of the autoencoder for lossy compression serves as an effective protection against information leakage from the updates. Additionally, the perturbation is intrinsically linked to the autoencoder’s input, thereby achieving a perturbation with respect to the parameters of different layers. Moreover, our approach can reduce 4.1 × the communication rate compared to federated averaging. We empirically validate our method using two widely-used models within the context of federated learning, considering three datasets, and assess its performance against several well-established defense frameworks. The results indicate that our approach attains a model performance nearly identical to that of unmodified local updates, while effectively preventing information leakage and reducing communication costs in comparison to other methods, including noisy gradients, gradient sparsification, and PRECODE.

A document by Xupeng Ai . Click on the document to view its contents.

A document by Md Shamim Hasan. Click on the document to view its contents.

In this manuscript we investigate the implications of adopting a double cyclic prefix in the orthogonal timefrequency space modulation. Our study first focuses on the analysis of the modulated signal and on the development of a useful model for the received signal in the presence of a doubly selective fading channel. On the one hand, our mathematical results allow us to accurately assess the impact of pulse shaping on the structure of the transmitted waveform and on its power spectral density, and to develop some simple rules for allocating multiple pilot symbols within each orthogonal time-frequency space symbol. On the other hand, they are exploited to develop a novel algorithm for pilot-aided channel estimation, whose output provides a detailed representation of the communication channel; for this reason, it can be used for sensing at the transmit side in integrated sensing and communication applications, or for channel equalization at the receive side in digital communications. Our numerical results evidence that our channel estimation & equalization algorithm outperforms the other related techniques available in the technical literature at the price of a limited increase in computational complexity.

Connected and autonomous vehicles (CAVs) rely heavily on a suite of perception sensors such as cameras and LIDAR for accurate and robust on-road decision making. However, these sensors have a limitation of line-of-sight (LOS), which can significantly restrict their effectiveness. To overcome this limitation, vehicle-to-everything (V2X) networks are being developed to wirelessly receive non-line-of-sight (nLOS) data from surrounding sources such as roadside units (RSUs). By leveraging nLOS data, V2X networks can significantly enhance the situational awareness of CAVs, improving traffic efficiency and road safety. In this paper, we present selected use-cases of V2X networks such as vehicle-to-vehicle (V2V) communication, vehicle platooning and cooperative perception using a beamsteering antenna that offers a wide bandwidth and fast beamsteering.

Touching a warm stimulator with the base of the finger and a cold stimulator with the middle part of the finger causes a sense of warmth at the fingertip outside the cold stimulator, revealing Extrapolation of Thermal Sensation (ETS). Although ETS shares similarities with the Thermal Grill Illusion (TGI) regarding spatial thermal integration, the spatial distributions of sensations are different. The TGI is limited inside boundaries that envelop physical stimuli, whereas ETS crosses the boundaries. This difference indicates ETS, although TGI is reproduced accompanied by overestimation of the cold stimulus, which is influenced by the spinal segmental distance between warm and cold stimuli, it remains to be seen whether ETS carries out the same. The present study investigated ETS and TGI by simultaneous warm, cold, and neutral stimulation on the fingers or the lower leg. To show the difference between ETS and TGI, we manipulated the segmental distance between warm and cold stimuli and observed the resulting perceived temperature of neutral and cold stimulators. The perceived temperatures of ETS as well as TGI varied in units of segmental distance, however, ETS was not reproduced where TGI was. Thus, we concluded ETS has a different mechanism from the TGI. The experimental results suggest a non-uniform intersegmental connection contributes to the lower reproducibility of ETS on the lower leg. INDEX TERMS Extrapolation of thermal sensation, thermal grill illusion, thermal referral

Despite the impressive performance of current deep learning models in the field of medical imaging, the transfer of the lung segmentation task in X-ray images to clinical practice is still a pending task. In this study, we explored the performance of a fully automatic framework for lung fields segmentation in chest X-ray images, based on the combination of the Segment Anything Model (SAM) with prompt capabilities, and the You Only Look Once (YOLO) model to provide effective prompts. Transfer learning, loss functions and several validation strategies were intensively assessed. This provides a complete benchmark that enables future research studies to fairly compare new segmentation strategies. The results achieved demonstrate significant robustness and generalization capability against the variability in sensors, populations, disease manifestations, device processing and imaging conditions. The proposed framework is computationally efficient, can address bias in training over multiple datasets, and has the potential to be applied across other domains and modalities.

This study explores the potential enhancement of radio signal propagation in intricate environments through the deployment of basic resonant dipole scatterers. We employ the MATLAB ray tracer algorithm to examine the impact of dipolar scatterers on the power received at the user's location. The ray tracing model is modified for a dipolar scatterer model based on its bi-static scattering cross-section. Our analytical and simulation results indicate that the deployment of extra scattering components is nearly ineffective when a direct link is present. Conversely, in scenarios where there is no line-ofsight, the proposed approach can considerably boost the power delivered. This cost-effective and straightforward method could complement other channel optimization strategies, such as the application of reconfigurable intelligent surfaces.

As biological wide-field visual neurons in locusts, lobula giant motion detectors (LGMDs) can effectively predict collisions and trigger avoidance before the collision occurs. This capability has extensive potential applications in the field of autonomous driving, unmanned aerial vehicles, and more. Currently, describing the LGMD characteristics is divided into two viewpoints, one emphasizing the presynaptic visual pathway and the other emphasizing the postsynaptic LGMDs neuron. Indeed, both have their research support leading to the emergence of two computational models, but both lack a biophysical description of the behavior in the individual LGMD neuron. This paper aims to mimic and explain LGMD's individual behavior based on fractional spiking neurons and construct a biomimetic visual model for the LGMD compatible with these two characteristics. Methods: We implement the visual model in the form of spikes by choosing an event camera rather than a conventional CMOS camera to simulate the photoreceptors and follow the topology of the ON/OFF visual pathway, enabling it to incorporate the lateral inhibition to mimic the LGMD's system from the bottom up. Second, most computational models of motion perception use only the dendrites within the LGMD neurons as the ideal pathway for linear summation, ignoring dendritic effects inducing neuronal properties. Thus, we introduced fractional spiking neuron (FSN) circuits into the model by altering dendritic morphological parameters to simulate multiscale spike frequency adaptation (SFA) observed in LGMDs. In addition, we have attempted to add one more circuit of dendritic trees into fractional spiking neurons to be compatible with the postsynaptic FFI in LGMDs and provide a novel explanatory approach and a predictive model for studying LGMD neurons. Results: Finally, we test that the event-driven biomimetic visual model can achieve collision detection and looming selection in different complex scenes, especially fast-moving objects.

Following the recent rise in generative artificial intelligence (GenAI) tools, fundamental questions about their wider impacts have started reverberating around various disciplines. To this end, this work was undertaken: (i) firstly, to track the unfolding landscape of general issues surrounding GenAI tools; (ii) secondly, as an exploratory inquiry to elucidate the specific opportunities and limitations of GenAI tools as part of the technology-assisted enhancement of mechanical engineering education and professional practices. As part of the investigation, we conduct and present a brief scientometric analysis of recently published studies to unravel the emerging trend on the subject matter. Further, experimentation was done with selected GenAI tools (Bard, ChatGPT, DALL.E, and 3DGPT) for mechanical engineering-related tasks. The study identified several pedagogical and professional opportunities and guidelines for deploying GenAI tools in mechanical engineering. Besides, the study highlights some pitfalls of GenAI tools for analytical reasoning tasks (e.g., subtle errors in computation involving unit conversions) and sketching/image generation tasks (e.g., poor demonstration of symmetry).

The Choroid Plexus (ChP) plays a vital role in brain homeostasis, serving as part of the BloodCerebrospinal Fluid Barrier, contributing to brain clearance pathways and being the main source of cerebrospinal fluid. Since the involvement of ChP in neurological and psychiatric disorders is not entirely established and currently under investigation, accurate and reproducible segmentation of this brain structure on large cohorts remains challenging. This paper presents ASCHOPLEX, a deep-learning tool for the automated segmentation of human ChP from structural MRI data that integrates existing software architectures like 3D UNet, UNETR, and DynUNet to deliver accurate ChP volume estimates. Here we trained ASCHOPLEX on 128 T1-w MRI images comprising both controls and patients with Multiple Sclerosis. ASCHOPLEX’s performances were evaluated using traditional segmentation met rics; manual segmentation by experts served as ground truth. To overcome the generalizability problem that affects data-driven approaches, an additional fine-tuning procedure (ASCHOPLEXtune) was implemented on 77 T1-w PET/MRI images of both controls and depressed patients. ASCHOPLEX showed superior performance compared to commonly used methods like FreeSurfer and Gaussian Mixture Model both in terms of Dice Coefficient (ASCHOPLEX 0.80, ASCHOPLEXtune 0.78) and estimated ChP volume error (ASCHOPLEX 9.22%, ASCHOPLEXtune 9.23%). These results highlight the high accuracy, reliability, and reproducibility of ASCHOPLEX ChP segmentations.ASCHOPLEX Toolboox is available on GitLab:  https://gitlab.dei.unipd.it/fair/aschoplex

Previous methods have demonstrated remarkable performance in single image super-resolution (SISR) tasks with known and fixed degradation (e.g., bicubic downsampling). However, when the actual degradation deviates from these assumptions, these methods may experience significant declines in performance. In this paper, we propose a Dual Branch Degradation Extractor Network to address the blind SR problem. While some BlindSR methods assume noisefree degradation and others do not explicitly consider the presence of noise in the degradation model, our approach predicts two unsupervised degradation embeddings that represent blurry and noisy information, respectively. The SR network can then be adapted to blur embedding and noise embedding in distinct ways. Furthermore, we treat the degradation extractor as a regularizer to capitalize on the differences between SR and HR images. Extensive experiments on several benchmarks demonstrate that our method achieves SOTA performance in the blind SR problem.

A document by Timothy Wong. Click on the document to view its contents.

With the ever-increasing requirements placed on current and future pulsed power systems in terms of voltage, power, and compactness; solid insulation is a strong candidate for the development of novel insulation systems capable of meeting these specifications. However, the issue of solid-solid interfaces under non-standard and fast-rising impulses must firstly be addressed, as the failure to do so may pose significant risk of electrical breakdown due to reduced dielectric strength across interfacial contacts. In this work, the impulsive breakdown characteristics across dry-mate solid interfaces formed between PVC (polyvinylchloride), Torlon (polyamide-imide), Delrin (polyoxymethylene), Perspex (polymethylmethacrylate), and Ultem (polyetherimide) has been investigated in atmospheric air and under two different impulsive waveforms rising at ∼2400 kV/µs and ∼0.35 kV/µs. The statistical treatment of the obtained impulsive breakdown voltages and time to breakdowns are presented, alongside an analysis of the post-breakdown surfaces and discharge channel morphologies. The results indicate that under low mating pressure conditions (10’s of kPa), the interfacial breakdown strength may be below that of only an air gap with no dielectrics. A correlation between the estimated asperity aspect ratio and the interfacial breakdown strength has been observed. This suggests that under the present experimental conditions, field enhancement around surface asperities may be a dominating factor which defines the breakdown strength of the interface, since the surface asperities do not deform sufficiently to form strong interfacial contact spots, and thus reducing the interfacial tracking resistance. This therefore provides little to impede the development of interfacial discharges. The widths of post-breakdown traces left by plasma channels on the contacting surfaces have also been shown to be dependent on the rate of voltage rise, dV/dt, and on the material forming the interface. The results arising from this work may aid in the future development of high voltage solid insulating systems for power and pulsed power systems.

A document by Timothy Wong. Click on the document to view its contents.

This letter, based on measuring the admittance of a black-boxed grid-following inverter (GFLI) at seven distinct operating points to synthesize an operating-point-parameterized state-space model (OPP SSM), presents an efficient method for analytical prediction of the maximum transferable active power (MTAP) injected by the black-boxed GFLI into the grid under any arbitrary grid condition and reactive power injection level. The proposed method obviates the need for repetitive electromagnetic transient time-domain simulations or frequency-domain impedance criteria by gradually increasing active power output until the MTAP is reached. The feasibility of the synthesized OPP SSM for MTAP prediction has been experimentally validated on a power-controlled GFLI, demonstrating its potential for aiding the integration of inverter-based resources in modern power systems.

This paper outlines a systematic methodology based on the superposition principle to develop a unified model for a Triple Active Bridge (TAB) converter. The proposed model is valid for all operating modes and modulation strategies of the TAB converter. The proposed model accurately predicts the AC port and DC port currents in a TAB converter, and it is used to compute performance metrics of the TAB converter, such as the peak and RMS currents at the AC ports and the average currents at the DC ports. The proposed model is further used to analyze the Zero Voltage Switching (ZVS) boundaries of the TAB converter. One of the features of the proposed model is that it incorporates the impact of transformer magnetizing inductance on the AC and DC port currents as well as the ZVS boundaries of the TAB converter. The model’s accuracy is verified against extensive switching circuit simulations for various operating conditions. The model is used to design the TAB converter parameters. Experimental results from a TAB converter laboratory prototype are presented to showcase the impact of magnetizing inductance on TAB converter performance and ZVS boundaries. The soft-switching boundaries predicted by the model are experimentally verified and efficiency results are presented to highlight the benefits of achieving ZVS.

A document by Tuure Nurminen. Click on the document to view its contents.