Though extensive research has been conducted on human locomotion for many decades, problems persist in simulating human movement, hindering the examination of musculoskeletal drivers and clinical conditions. Reinforcement learning (RL) approaches currently applied to human locomotion simulations are proving promising, showcasing musculoskeletal dynamics. These simulations often prove inadequate in recreating natural human locomotion; this inadequacy stems from the lack of incorporation of any reference data on human movement in most reinforcement strategies. To address the presented difficulties, this research has formulated a reward function using trajectory optimization rewards (TOR) and bio-inspired rewards, drawing on rewards from reference movement data collected via a single Inertial Measurement Unit (IMU) sensor. Reference motion data was acquired by positioning sensors on the participants' pelvises. In addition to this, we refined the reward function, leveraging existing work in TOR walking simulations. The modified reward function in the simulated agents, as confirmed by the experimental data, led to improved performance in replicating participant IMU data, resulting in a more realistic simulation of human locomotion. As a bio-inspired defined cost metric, IMU data contributed to a stronger convergence capability within the agent's training process. The models with reference motion data converged faster, showing a marked improvement in convergence rate over those without. Henceforth, human movement simulation can be executed more promptly and across a wider variety of settings, leading to superior simulation results.
Although deep learning has achieved substantial success in various applications, its resilience to adversarial samples is still a critical weakness. A robust classifier was trained using a generative adversarial network (GAN) to mitigate this vulnerability. This paper introduces a novel generative adversarial network (GAN) model and describes its implementation, focusing on its effectiveness in defending against gradient-based adversarial attacks using L1 and L2 constraints. The proposed model, while informed by related work, includes several innovative designs: a dual generator architecture, four unique generator input formulations, and two distinct implementations that yield vector outputs constrained by L and L2 norms. New methods for GAN formulation and parameter tuning are proposed and tested against the limitations of existing adversarial training and defensive GAN strategies, including gradient masking and training complexity. Moreover, an evaluation of the training epoch parameter was conducted to ascertain its influence on the final training outcomes. The optimal GAN adversarial training formulation, indicated by the experimental results, demands a more comprehensive gradient signal from the target classifier. These results additionally illustrate GANs' success in circumventing gradient masking and creating useful perturbations to augment the dataset. The model's performance against PGD L2 128/255 norm perturbation showcases an accuracy over 60%, contrasting with its performance against PGD L8 255 norm perturbation, which maintains an accuracy roughly at 45%. The results demonstrate a transferability of robustness among the constraints of the proposed model. Moreover, a robustness-accuracy trade-off was observed, accompanied by overfitting and the generative and classifying models' capacity for generalization. Selleck Harmine These constraints and concepts for future improvements shall be examined.
Within the realm of car keyless entry systems (KES), ultra-wideband (UWB) technology stands as a progressive solution for keyfob localization, bolstering both precise positioning and secure data transfer. However, the determination of distance for vehicles encounters significant inaccuracies due to non-line-of-sight (NLOS) situations, exacerbated by the vehicle's position. The NLOS problem has prompted the development of methods to reduce point-to-point ranging errors or to calculate the coordinates of the tag by means of neural networks. While promising, certain concerns remain, specifically concerning low accuracy, potential overfitting, or a significant number of parameters. A method of merging a neural network and a linear coordinate solver (NN-LCS) is proposed as a solution to these problems. Two fully connected layers independently extract distance and received signal strength (RSS) features, which are subsequently combined within a multi-layer perceptron (MLP) for distance estimation. The application of the least squares method to error loss backpropagation within neural networks is shown to be viable for distance correcting learning tasks. In conclusion, our model carries out localization as a continuous process, yielding the localization outcomes directly. The outcomes suggest the proposed method possesses both high accuracy and a small model size, which translates to easy deployment on embedded devices with limited processing power.
Industrial and medical applications both rely heavily on gamma imagers. To achieve high-quality images, modern gamma imagers often leverage iterative reconstruction methods that rely heavily on the system matrix (SM). While an accurate SM can be derived from an experimental calibration process employing a point source spanning the FOV, this approach suffers from a protracted calibration time needed to eliminate noise, thereby challenging its application in realistic settings. A time-efficient SM calibration technique for a 4-view gamma imager is described, encompassing short-term SM measurements and deep learning for noise reduction. The key procedure entails fragmenting the SM into numerous detector response function (DRF) image components, classifying these DRFs into varied groups through a dynamically adjusted K-means clustering approach to manage variations in sensitivity, and ultimately individually training distinct denoising deep networks for each DRF category. We analyze the performance of two denoising networks, juxtaposing their results with those obtained using a Gaussian filtering method. The results on denoised SM using deep networks indicate equivalent imaging performance compared to the long-term SM measurements. By optimizing the SM calibration process, the time required for calibration has been reduced drastically from 14 hours to 8 minutes. The SM denoising method under consideration demonstrates promising capabilities in augmenting the output of the 4-view gamma imager, and is widely adaptable to other imaging setups requiring an experimental calibration process.
Though recent Siamese network-based visual tracking methods have excelled in large-scale benchmark testing, challenges remain in effectively separating target objects from distractors with similar visual attributes. In order to resolve the issues highlighted earlier, we present a novel global context attention module for visual tracking. This proposed module gathers and summarizes the overall global scene information to adjust the target embedding, thereby increasing its discriminative power and robustness. Our global context attention module, reacting to a global feature correlation map of a scene, extracts contextual information. This module then computes channel and spatial attention weights for adjusting the target embedding, thus emphasizing the relevant feature channels and spatial segments of the target object. Large-scale visual tracking datasets were used to evaluate our tracking algorithm. Our results show improved performance relative to the baseline algorithm, and competitive real-time speed. Through further ablation experiments, the effectiveness of the proposed module is ascertained, demonstrating that our tracking algorithm performs better across various challenging aspects of visual tracking.
Heart rate variability (HRV) parameters are useful in clinical settings, such as sleep cycle identification, and ballistocardiograms (BCGs) allow for a non-intrusive quantification of these parameters. Selleck Harmine Electrocardiography is the established clinical method for estimating heart rate variability (HRV), however, bioimpedance cardiography (BCG) and electrocardiograms (ECGs) show contrasting heartbeat interval (HBI) estimations, impacting the computed HRV parameters. This research project assesses the usability of BCG-based heart rate variability (HRV) metrics to identify sleep stages, determining how timing variations impact the parameters of interest. A set of artificial time offsets were incorporated to simulate the distinctions in heartbeat intervals between BCG and ECG methods, and the generated HRV features were subsequently utilized for sleep stage identification. Selleck Harmine Subsequently, we delineate the connection between the mean absolute error in HBIs and the resultant accuracy of sleep stage identification. Our previous contributions concerning heartbeat interval identification algorithms are extended to demonstrate the similarity between our simulated timing jitters and the errors in heartbeat interval measurements. Sleep staging using BCG data displays accuracy comparable to ECG-based methods; a 60-millisecond increase in HBI error can translate into a 17% to 25% rise in sleep-scoring error, as seen in one of our investigated cases.
A fluid-filled Radio Frequency Micro-Electro-Mechanical Systems (RF MEMS) switch is the subject of this current investigation, and its design is presented here. The proposed RF MEMS switch's operating principle was analyzed using air, water, glycerol, and silicone oil as dielectric fluids, examining their effect on drive voltage, impact velocity, response time, and switching capacity. Insulating liquid, when used to fill the switch, leads to a reduction in both the driving voltage and the impact velocity of the upper plate colliding with the lower plate. The elevated dielectric constant of the filling medium is associated with a diminished switching capacitance ratio, which correspondingly affects the switch's operational capabilities. By assessing the threshold voltage, impact velocity, capacitance ratio, and insertion loss of the switch filled with different media, including air, water, glycerol, and silicone oil, the ultimate choice fell upon silicone oil as the ideal liquid filling medium for the switch.