Federated learning is a distributed machine learning approach in which clients train models locally with their own data and upload them to a server so that their trained results are shared between them without uploading raw data to the server. There are some challenges in federated learning, such as communication size reduction and client heterogeneity. The former can mitigate the communication overheads, and the latter can allow the clients to choose proper models depending on their available compute resources. To address these challenges, in this paper, we utilize Neural ODE based models for federated learning. The proposed flexible federated learning approach can reduce the communication size while aggregating models with different iteration counts or depths. Our contribution is that we experimentally demonstrate that the proposed federated learning can aggregate models with different iteration counts or depths. It is compared with a different federated learning approach in terms of the accuracy. Furthermore, we show that our approach can reduce communication size by up to 89.4% compared with a baseline ResNet model using CIFAR-10 dataset.
Abstract Purpose This study aimed to investigate the frequency of COVID-19 vaccine-induced reactive change and potential factors correlated with increased FDG uptake on positron emission tomography (PET)/computed tomography (CT). Materials and methods We evaluated 285 patients who underwent PET/CT between June and September 2021 and had a known history of COVID-19 vaccination. Information on the injection site, vaccine type, and adverse reactions was obtained. We visually assessed the presence or absence of accumulation in the axillary and supraclavicular lymph nodes and the deltoid muscles. We measured the maximum standardized uptake value (SUVmax) using semi-quantitative analysis. Results Our study included 158 males and 127 females aged 16 to 94 years. The median time between vaccination and PET/CT was nine and 42 days for patients who had received their first and second doses, respectively. We observed axillary lymph node accumulation, supraclavicular lymph node accumulation, and deltoid muscle accumulation in 99 (SUVmax 1.07–25.1), nine (SUVmax 2.28–14.5), and 33 cases (SUVmax 0.93–7.42), respectively. In cases with axillary lymph node (P = 0.0066) or deltoid muscle (P = 0.047) accumulation, the shorter the time since vaccination, the higher the FDG accumulation. Patients with axillary lymph node accumulation were significantly younger (P < 0.0001) and had a significantly higher frequency of adverse reactions such as fever (P < 0.0001) and myalgia (P = 0.001). Logistic regression analysis also showed that age, sex, days since vaccination, and adverse reactions such as fever and myalgia were important factors for axillary lymph node accumulation. Conclusion Our study found that FDG accumulation in the axillary lymph nodes and deltoid muscle was higher within a shorter time after vaccination, and axillary lymph node accumulation was higher in young patients, females, and those with adverse reactions of fever and myalgia. Confirming the vaccination status, time since vaccination, and the presence of adverse reactions before PET may reduce false positives.
A transformer is an emerging neural network model that employs an attention mechanism. It has been adopted to various tasks and achieved a favorable accuracy compared to CNNs (Convolutional Neural Networks) and RNNs (Recurrent Neural Networks). Although the attention mechanism is recognized as a general-purpose component, many of the transformer models require a significant number of parameters and thus they are not suited to low-cost edge devices. Recently, a resource-efficient hybrid model that uses ResNet as a backbone architecture and replaces a part of its convolutional layers with an MHSA (Multi-Head Self-Attention) mechanism was proposed. In this paper, we significantly reduce the parameter size of this approach by using Neural ODE as a backbone architecture for the MHSA mechanism. The proposed hybrid model reduces the parameter size by 97.3% compared to the original model without degrading the accuracy. Since the model size is quite small, it is implemented on Xilinx ZCU104 FPGA (Field Programmable Gate Array) board so that it can fully exploit on-chip BRAM/URAM resources. The FPGA implementation is evaluated in terms of resource utilization, accuracy, performance, and power consumption. The results demonstrate that it speeds up the model by up to 2.63 times compared to a software execution without accuracy degradation.
Human CYP3A is the most abundant P450 isozyme present in the human liver and small intestine, and metabolizes around 50% of medical drugs on the market. The human CYP3A subfamily comprises four members (CYP3A4, CYP3A5, CYP3A7, CYP3A43) encoded on human chromosome 7. However, transgenic mouse lines carrying the entire human CYP3A cluster have not been constructed because of limitations in conventional cloning techniques. Here, we show that the introduction of a human artificial chromosome (HAC) containing the entire genomic human CYP3A locus recapitulates tissue- and stage-specific expression of human CYP3A genes and xenobiotic metabolism in mice. About 700 kb of the entire CYP3A genomic segment was cloned into a HAC (CYP3A-HAC), and trans-chromosomic (Tc) mice carrying a single copy of germline-transmittable CYP3A-HAC were generated via a chromosome-engineering technique. The tissue- and stage-specific expression profiles of CYP3A genes were consistent with those seen in humans. We further generated mice carrying the CYP3A-HAC in the background homozygous for targeted deletion of most endogenous Cyp3a genes. In this mouse strain with 'fully humanized' CYP3A genes, the kinetics of triazolam metabolism, CYP3A-mediated mechanism-based inactivation effects and formation of fetal-specific metabolites of dehydroepiandrosterone observed in humans were well reproduced. Thus, these mice are likely to be valuable in evaluating novel drugs metabolized by CYP3A enzymes and in studying the regulation of human CYP3A gene expression. Furthermore, this system can also be used for generating Tc mice carrying other human metabolic genes.
This paper presents the suppression method of anti-resonance and resonance caused by use of air springs, which are the actuator of a pneumatic vibration isolator. Although the relative displacement derivative (RDD) positive feedback control is effective for the suppression of the anti-resonance and resonance, a control system is not causal under the condition that the time delay of response of air springs is present. To overcome this issue, the RDD positive feedback control combined with Smith prediction is used. The effectiveness of the proposed approach is shown by simulation and experiment.
The factor that influences the residual stress distribution generated by a quenching process was clarified and an analysis technique for estimating the residual stress distribution was established accurately even if the diameter of the cylinder is changed. First of all, a method of the numerical analysis was established through the comparative study of the experimental and the numerical analysis values of the residual stress distribution by using the steel material where the phase transformation is not generated during a quenching process. Next, using the established method of the numerical analysis, the influences of material constants and mechanical properties on the residual stress distribution were examined followed by clarifying the factor to estimate the residual stress distribution when the diameter of the cylindrical material is changed. Non-dimensional heat dissipation 'H' is important as for the factor of temperature, and each temperature dependency of linear expansion coefficient, Young's modulus and yield stress are important as for the factor of materials to analyze the residual stress distribution. It was confirmed that the residual stress distribution was able to be predicted by using Non-dimensional heat dissipation 'H' provided that the material was fixed and the diameter of cylinder was changed.
