Bayesian Federated Learning with Hamiltonian Monte Carlo: Algorithm and Theory
Not provided.
Engineering topics
Publications and source records attributed to Deng, Wei.
Not provided.
In the presence of energetic particles (EPs), the long-lived mode (LLM) frequency multiplication with n = 1, 2, 3, or higher is often observed on HL-2A, where n is the toroidal mode number. Hybrid kinetic-MHD model simulations of the energetic particle (EP) driven kink/fishbone modes on a static HL-2A-like tokamak using NIMROD code find that when the background plasma pressure is relatively high, and the EP pressure and the beam energy are relatively low, the mode frequency increases almost linearly with EP pressure, and the frequency is proportional to n (‘frequency multiplication’), even in the absence of any equilibrium plasma rotation. In addition, the frequency multiplication persists as the safety factor at the magnetic axis q 0 varies. In the absence of EPs, the growth rate of the 1/1 mode is the largest; however, as the EP pressure increases, the growth rate of 2/2 modes or 3/3 modes becomes dominant, suggesting that higher-n modes are more vulnerable to EPs. These results may shed light on the understanding of the toroidal mode number dependence of kink/fishbone modes in the advanced scenarios of tokamaks with weak or reversed central magnetic shear.
Not provided.
Bayesian approaches have been successfully integrated into training deep neural networks. One popular family is stochastic gradient Markov chain Monte Carlo methods (SG-MCMC), which have gained increasing interest due to their ability to handle large datasets and the potential to avoid overfitting. Although standard SG-MCMC methods have shown great performance in a variety of problems, they may be inefficient when the random variables in the target posterior densities have scale differences or are highly correlated. Here, we present an adaptive Hessian approximated stochastic gradient MCMC method to incorporate local geometric information while sampling from the posterior. The idea is to apply stochastic approximation (SA) to sequentially update a preconditioning matrix at each iteration. The preconditioner possesses second-order information and can guide the random walk of a sampler efficiently. Instead of computing and saving the full Hessian of the log posterior, we use limited memory of the samples and their stochastic gradients to approximate the inverse Hessian-vector multiplication in the updating formula. Moreover, by smoothly optimizing the preconditioning matrix via SA, our proposed algorithm can asymptotically converge to the target distribution with a controllable bias under mild conditions. To reduce the training and testing computational burden, we adopt a magnitude-based weight pruning method to enforce the sparsity of the network. Our method is user-friendly and demonstrates better learning results compared to standard SG-MCMC updating rules. The approximation of inverse Hessian alleviates storage and computational complexities for large dimensional models. Numerical experiments are performed on several problems, including sampling from 2D correlated distribution, synthetic regression problems, and learning the numerical solutions of heterogeneous elliptic PDE. The numerical results demonstrate great improvement in both the convergence rate and accuracy.
Deep neural networks have been successfully employed in an extensive variety of research areas, including solving partial differential equations. Despite its significant success, there are some challenges in effectively training DNN, such as avoiding overfitting in over-parameterized DNNs and accelerating the optimization in DNNs with pathological curvature. Here, we propose a Bayesian type sparse deep learning algorithm. The algorithm utilizes a set of spike-and-slab priors for the parameters in the deep neural network. The hierarchical Bayesian mixture will be trained using an adaptive empirical method. That is, one will alternatively sample from the posterior using preconditioned stochastic gradient Langevin Dynamics (PSGLD), and optimize the latent variables via stochastic approximation. The sparsity of the network is achieved while optimizing the hyperparameters with adaptive searching and penalizing. A popular SG-MCMC approach is Stochastic gradient Langevin dynamics (SGLD). However, considering the complex geometry in the model parameter space in nonconvex learning, updating parameters using a universal step size in each component as in SGLD may cause slow mixing. To address this issue, we apply a computationally manageable preconditioner in the updating rule, which provides a step-size parameter to adapt to local geometric properties. Moreover, by smoothly optimizing the hyperparameter in the preconditioning matrix, our proposed algorithm ensures a decreasing bias, which is introduced by ignoring the correction term in the preconditioned SGLD. According to the existing theoretical framework, we show that the proposed algorithm can asymptotically converge to the correct distribution with a controllable bias under mild conditions. Numerical tests are performed on both synthetic regression problems and learning solutions of elliptic PDE, which demonstrate the accuracy and efficiency of the present work.
Electronic properties of organic semiconductor (OSC) thin films are largely determined by their morphologies and crystallinities. However, solution-processed conjugated small-molecule OSC thin films usually exhibit abundant grain boundaries and impure grain orientations because of complex fluid dynamics during solution coating. Here, a novel methodology, water-surface drag coating, is demonstrated to fabricate high-quality OSC thin films with greatly enhanced charge transport properties. This method utilizes the water surface to alter the evaporation dynamics of solution to enlarge the grain size, and a unique drag-coating process to achieve the unidirectional growth of organic crystals. Furthermore, using 2,8-difluoro-5,11-bis(triethylsilylethynyl)anthradithiophene (Dif-TES-ADT) as an example, thin films with millimeter-sized single-crystal domains and pure crystallographic orientations are achieved, revealing a significant enhancement (4.7 times) of carrier mobility. More importantly, the resulting film can be directly transferred onto any desired flexible substrates, and flexible transistors based on the Dif-TES-ADT thin films show a mobility as high as 16.1 cm 2 V -1 s -1 , which represents the highest mobility value for the flexible transistors reported thus far. The method is general for the growth of various high-quality OSC thin films, thus opening up opportunities for high-performance organic flexible electronics.
Highlights: • The apoptosis of peripheral blood CD4+ T cell was reduced in lupus-prone mice. • UC-MSCs ameliorated disease progression of lupus mice. • UC-MSCs promoted CD4+ T lymphocyte apoptosis of lupus mice in vivo and vitro. Umbilical cord-derived mesenchymal stem cell transplantation (UCMSCT) has been used to treat human autoimmune diseases like lupus for example, but little is known about its effect on cell apoptosis. Here we evaluated the efficacy of UCMSCT for lupus treatment and explored the mechanism by which mesenchymal stem cells (MSCs) modulate T cell apoptosis in lupus mice. 1 × 10{sup 6} human umbilical cord-derived mesenchymal stem cells (UC-MSCs) were injected into B6.MRL-Fas{sup lpr} (B6.lpr) mice via tail vein. 6 h, 24 h or 4 weeks later, the mice were sacrificed and the apoptosis of lymphocytes in peripheral blood and spleen were detected by flow cytometry. The immune cell subpopulations in spleen were also measured at 6 h and 24 h, respectively. The therapeutic effects were assessed after 4 weeks. The frequency of peripheral blood CD4+ T cell apoptosis was reduced in lupus-prone B6.lpr mice. UCMSCT alleviated the disease phenotypes in B6.lpr mice, decreased the ratio of Th1 as well as Th2 cells, and increased percentages of apoptotic CD4+ T cells in vivo and vitro. Collectively, our findings unravel that UCMSCT alleviate lupus disease and reverse immune imbalance possibly by promoting T cell apoptosis in B6.lpr mice.
Abstract not provided
Highlights: • lncRNA-ANCR upregulation decreased bone formation of PDLSCs. • miR-758 upregulation promoted osteogenic differentiation of PDLSCs. • miR-758 is the direct target of lncRNA ANCR. • lncRNA-ANCR/miR-758/Notch2-Wnt/β-catenin signaling pathway involve in bone formation. Long noncoding RNAs (lncRNAs) were proposed to be important regulators influencing various differentiation processes. Yet, the molecular mechanisms of lncRNAs governing osteogenic differentiation of Periodontal Ligament Stem Cells (PDLSCs) remain unclear. Here, PDLSCs were isolated from normal periodontal ligament of human (PDL) whereas P-PDLSCs were isolated from periodontitis affected PDL. Quantitative real-time PCR (qRT-PCR) was performed to examine the relative expression level of lncRNA-ANCR and of Osterix (OSX), Alkaline Phosphatase (ALP) as well as Runt-related transcription factor 2 (RUNX2) in PDLSCs. Gain- and loss-of- function experiments was performed to study the role of lncRNA-ANCR. Alizarin Red staining was used to evaluate the function of lncRNA-ANCR and miRNA-758 on osteogenic differentiation. In addition, via dual luciferase reporter assay and RNA immunoprecipitation the microRNA sponge potential of lncRNA-ANCR was assessed. A luciferase reporter assay identified the correlation between miR-758 and Notch2. Our results showed that the expression of ALP, RUNX2 and OSX were increased whereas lncRNA-ANCR was decreased during the process of differentiation in PDLSCs. Overexpression of lncRNA-ANCR decreased the expression of ALP, RUNX2 and OSX as confirmed by Alizarin red staining. Overexpression of lncRNA-ANCR resulted in reduction of the miR-758 expression level. Furthermore, RNA immunoprecipitation proved that lncRNA-ANCR targets miR-758 directly. The results of dual luciferase reporter assay also demonstrated that miR-758 regulated Notch2 expression by targeting 3′-UTR of Notch2. In conclusion, the novel pathway lncRNA-ANCR/miR-758/Notch2 plays an important role in the process of regulating osteogenic differentiation of PDLSCs.
Highlights: • β-mercaptoethanol (BME) and monothiolglycerol (MTG) reduce cardiac differentiation. • BME and MTG treatments decrease TNNI3/TNNI1 and MLC2V/MLC2A expression ratios. • BME and MTG treatments reduce beating rates and contraction/relaxation velocities. • Trolox, a strong ROS scavenger, does not inhibit cardiac differentiation. Reactive oxygen species (ROS) scavengers such as beta-mercaptoethanol (BME) and monothiol glycerol (MTG) are extensively used in stem cell research to prevent cellular oxidative stress. However, how these antioxidant supplements impact stem cell cardiac differentiation, a process regulated by redox-signaling remains unknown. In this study, we found that removal of BME from the conventional high-glucose, serum-based differentiation medium improved cardiac differentiation efficiency by 2-3 fold. BME and MTG treatments during differentiation significantly reduced mRNA expression of cardiac progenitor markers (NKX2.5 and ISL1) as well as sarcomeric markers (MLC2A, MLC2V, TNNI3, MYH6 and MYH7), suggesting reduced cardiomyogenesis by BME or MTG. Moreover, BME and MTG altered the expression ratios between the sarcomeric isoforms. In particular, TNNI3 to TNNI1 ratio and MLC2V to MLC2A ratio were significantly lower in BME or MTG treated cells than untreated cells, implying altered cardiomyocyte phenotype and maturity. Lastly, BME and MTG treatments resulted in less frequent beating, slower contraction and relaxation velocities than untreated cells. Interestingly, none of the above-mentioned effects was observed with Trolox, a non-thiol based antioxidant, despite its strong antioxidant activity. This work demonstrates that commonly used antioxidant supplements may cause considerable changes to cellular redox state and the outcome of differentiation.
Not provided.
The early optical afterglow emission of several gamma-ray bursts (GRBs) shows a high linear polarization degree (PD) of tens of percent, suggesting an ordered magnetic field in the emission region. The light curves are consistent with being of a reverse shock (RS) origin. However, the magnetization parameter, σ , of the outflow is unknown. If σ is too small, an ordered field in the RS may be quickly randomized due to turbulence driven by various perturbations so that the PD may not be as high as observed. Here we use the “Athena++” relativistic MHD code to simulate a relativistic jet with an ordered magnetic field propagating into a clumpy ambient medium, with a focus on how density fluctuations may distort the ordered magnetic field and reduce PD in the RS emission for different σ values. For a given density fluctuation, we discover a clear power-law relationship between the relative PD reduction and the σ value of the outflow. Such a relation may be applied to estimate σ of the GRB outflows using the polarization data of early afterglows.
Organic semiconductor micro/nanocrystals (OSMCs) have attracted intensive research over the past few decades. Their unique electrical and mechanical properties afford them applications in a vast network of active flexible electronic and optoelectronic fields, such as organic field‐effect transistors (OFETs), photodetectors (PDs), and organic light‐emitting diodes (OLEDs) etc. However, the growth orientation and location of OSMCs are usually stochastic in nature, which hinders the further use of OSMCs in large‐area plastic electronic devices. On the other hand, traditional microelectronic processing techniques like photolithography may damage the plastic substrates, polymer insulators, and organic materials. Thus the alignment and patterning of OSMCs and development of new device construction technologies are prerequisite for their flexible device applications in practice. In this progress report,the important advancement for fabricating the aligned and patterned OSMCs, with a focus on their applications in next‐generation flexible electronic and optoelectronic devices, is highlighted. Also, the challenges and outlooks of the ordered and patterned OSMCs in the flexible device applications are presented.
In this article, a Bayesian tomography method using non-stationary Gaussian process for a prior has been introduced. The Bayesian formalism allows quantities which bear uncertainty to be expressed in the probabilistic form so that the uncertainty of a final solution can be fully resolved from the confidence interval of a posterior probability. Moreover, a consistency check of that solution can be performed by checking whether the misfits between predicted and measured data are reasonably within an assumed data error. In particular, the accuracy of reconstructions is significantly improved by using the non-stationary Gaussian process that can adapt to the varying smoothness of emission distribution. The implementation of this method to a soft X-ray diagnostics on HL-2A has been used to explore relevant physics in equilibrium and MHD instability modes. This project is carried out within a large size inference framework, aiming at an integrated analysis of heterogeneous diagnostics.
Rewarding experiences are often well remembered, and such memory formation is known to be dependent on dopamine modulation of the neural substrates engaged in learning and memory; however, it is unknown how and where in the brain dopamine signals bias episodic memory toward preceding rather than subsequent events. Here we found that photostimulation of channelrhodopsin-2–expressing dopaminergic fibers in the dentate gyrus induced a long-term depression of cortical inputs, diminished theta oscillations, and impaired subsequent contextual learning. Computational modeling based on this dopamine modulation indicated an asymmetric association of events occurring before and after reward in memory tasks. In subsequent behavioral experiments, preexposure to a natural reward suppressed hippocampus-dependent memory formation, with an effective time window consistent with the duration of dopamine-induced changes of dentate activity. Altogether, our results suggest a mechanism by which dopamine enables the hippocampus to encode memory with reduced interference from subsequent experience.
Here, this paper reports the systematic parameter study of a tristable nonlinear electromagnetic energy harvester for ambient low-frequency vibration. Numerical simulations and experimental investigations are performed on the harvester which consists of a cantilever beam, a tip coil, two tip magnets and two external side magnets. The external side magnets are deployed symmetrically along a concave surface parallel to the trajectory of the cantilever tip with a controllable distance so that the magnetic orientation of the tip magnets are matched with that of the side magnets. Therefore, instead of the ternary position parameters (d, h, α), a binary parameters pair (d 0 , d) is used to characterize the position of the side magnets and the performance of the energy harvester. The magnetic force and magnetic field on the cantilever tip therefore depend on the relative distance in the tip displacement direction between the tip magnets and side magnets, but is independent of the position of the side magnets on the concave surface. The magnetic force (field)-distance relationship is measured experimentally and curve fitted to obtain explicit expressions, in order to characterize the magnetic force (field) when the side magnets are placed at varied positions along the concave surface. Numerical simulation is, then, performed to predict the electromagnetic voltage output and the bandwidth of the energy harvester. The simulation results coincided with the measured data. Significant broadband response is obtained experimentally and the maximum RMS power output is 40.2 mW at 0.45 g of excitation. Furthermore, the proposed structure showcasing the matched magnetic orientation is characterized by the binary parameters pair (d 0 , d) and the systematic parametric approach could contribute to the design and study of nonlinear broadband energy harvesters.