Stochastic Layer Research Articles

Multi‐scale feature combination for person re‐identification

Person re-identification (Re-ID) is an instance-level task of image retrieval, and its identification accuracy depends on the distinguishable features extracted from people. However, most identification methods based on deep learning only mechanically extract distinguishable features of person images, and some important details are frequently overlooked. For scenes with substantial background differences or occlusions, the Re-ID efficiency is not high and the network scalability is not good. Here, the authors propose a multi-scale feature combination network (MFC-Net) model that combines structural feature information with global comprehensive feature information of the person images through a convolution neural network that can effectively retain distinguishing character detail information. The authors also propose a Gaussian stochastic pooling layer to solve the defects of the pooling layer. For the problem of many network parameters and weak performance, the authors propose an attentive feature convolutions layer. The authors perform many comparative experiments on three benchmark datasets. The results prove that our MFC-Net model performs well in person Re-ID and that its identification accuracy is higher than that of other investigated models.

Confinement improvement during detached phase with RMP application in deuterium plasmas of LHD

In order to explore the compatibility of good core plasma performance with divertor heat load mitigation, the interaction between cold edge plasma and core plasma transport, including the edge transport barrier (ETB), has been analysed in the divertor detachment discharges of deuterium plasmas in LHD with resonant magnetic perturbation (RMP) field application. The RMP application introduces a widened edge stochastic layer and sharp boundary in the magnetic field structure between the confinement region and the edge stochastic layer. The widened edge stochastic layer enhances impurity radiation and provides stable detachment operation as compared with the case without RMP. It is found that ETB is formed at the confinement boundary at the onset of detachment transition. However, as the detachment deepens, the resistive pressure gradient-driven MHD mode is excited, which degrades the ETB. At the same time, however, the core transport decreases to keep global plasma stored energy (W p) unchanged, showing clear core-edge coupling. After a gradual increase of density fluctuation during the MHD activity, a spontaneous increase of W p and the recovery of ETB are observed while the detachment is maintained. Then, the coherent MHD mode ceases and ELM-like bursts appear. In the improved mode, impurity decontamination occurs, and the divertor heat load increases slightly. Key controlling physical processes in the interplay between core and cold edge plasma are discussed. A comparison between deuterium and hydrogen plasmas shows that hydrogen plasmas exhibit similar features to the deuterium ones in terms of density and magnetic fluctuations, impurity decontamination towards higher confinement, etc. But most of the features are modest in the hydrogen plasmas and thus no clear confinement mode transition with clear ETB formation is defined. Better global confinement is obtained in the deuterium plasmas than the hydrogen ones at a higher radiation level.

Improving the Behaviour of Vision Transformers with Token-Consistent Stochastic Layers

Improving the Behaviour of Vision Transformers with Token-Consistent Stochastic Layers

Conservative generalized bifurcation diagrams and phase space properties for oval-like billiards

In this paper, we study some dynamic properties for oval-like billiards. These billiards have two control parameters, named ϵ, which controls the deformation of the boundary, and p, which changes the number of inflection points. The particle’s position (X,Y) uses Cartesian coordinates, and the angle μ gives us the particle’s direction. Here we consider a Poincare section, where we calculate the position X (in the horizontal axis) and angle μ every time a particle crosses Y=0 (in the vertical axis). We compute the phase space and the conservative generalized bifurcation diagrams (CGBD). These diagrams are obtained when changing the initial position X and the control parameter ϵ. We plot the respective maximum Lyapunov exponent for each combination of the control parameter and initial condition, which uses a customized color palette. These diagrams show how complex billiards dynamics are, where one can find the direct and inverse parabolic bifurcations. Moreover, one can highlight periodic, quasi-periodic, and chaotic regions. We found a fractal behavior (self-similar structure), where we verified the existence of period-adding structures logical sequences (periodic orbits) in the CGBD. These sequences accumulate in different regions depending on the control parameters, following the main body’s period and accumulating in different regions. When we set the control parameter p to 1, we observe that chaos dominates for a high enough value of the control parameter ϵ (which controls our billiard’s deformation). We also studied some orbits embedded in stochastic layers that appear near saddle points, which obey another period-adding logical sequence. These stochastic layers play a crucial role in the dynamics of billiard systems because that chaos grows in such regions, near saddle points, after increasing the control parameter’s value.

Collisionless plasma transport mechanisms in stochastic open magnetic field lines in tokamaks

The collisionless plasma transport in stochastic magnetic fields connecting to wall boundary has been studied to understand the underlying mechanisms of thermal quench physics in tokamak disruption. In this study, we present a comprehensive understanding of the plasma dynamics in three-dimensional stochastic open magnetic field lines, taking into account the consistent coupling of electron and ion dynamics through the ambipolar electric field. The open stochastic field lines act as a 3D magnetic mirror consisting of the magnetic well and hill regions. It was found that the magnetic hill regions play a critical role in enhancing electron thermal transport. When the plasma collapses to the wall, the 3D ambipolar potential arises in the stochastic layer to maintain the quasi-neutrality of the plasma. The E × B vortices induced by the 3D ambipolar potential mix the plasma across stochastic field lines and enhance the radial transport. Particularly, the E × B mixing between the magnetic well and hill regions provides a collisionless detrapping mechanism that plays a major role in the loss of high-energy trapped electrons. As a result, the electron temperature steadily decreases at a rate of about −0.5 keV ms−1, comparable to the rate observed in thermal quench experiments.

The Stability Boundary of the Distant Scattered Disk

Abstract The distant scattered disk is a vast population of trans-Neptunian minor bodies that orbit the Sun on highly elongated, long-period orbits. The orbital stability of scattered-disk objects (SDOs) is primarily controlled by a single parameter—their perihelion distance. While the existence of a perihelion boundary that separates chaotic and regular motion of long-period orbits is well established through numerical experiments, its theoretical basis as well as its semimajor axis dependence remain poorly understood. In this work, we outline an analytical model for the dynamics of distant trans-Neptunian objects and show that the orbital architecture of the scattered disk is shaped by an infinite chain of exterior 2:j resonances with Neptune. The widths of these resonances increase as the perihelion distance approaches Neptune’s semimajor axis, and their overlap drives chaotic motion. Within the context of this theoretical picture, we derive an analytic criterion for instability of long-period orbits, and demonstrate that rapid dynamical chaos ensues when the perihelion drops below a critical value, given by q crit = a N ln ( ( 24 2 / 5 ) ( m N / M ⊙ ) a / a N 5 / 2 ) 1 / 2 . This expression constitutes an analytic boundary between the “detached” and actively “scattering” subpopulations of distant trans-Neptunian minor bodies. Additionally, we find that within the stochastic layer, the Lyapunov time of SDOs approaches the orbital period, and show that the semimajor axis diffusion coefficient is approximated by  a ∼ ( 8 / ( 5 π ) ) ( m N / M ⊙ )  M ⊙ a N exp ​ − q / a N 2 / 2 . We confirm our results with direct N-body simulations and highlight the connections between scattered-disk dynamics and the Chirikov Standard Map. Implications of our results for the long-term evolution of minor bodies in the distant solar system are discussed.

Chance Constraints and Machine Learning integration for uncertainty management in Virtual Power Plants operating in simultaneous energy markets

The management of uncertainty is one of the most important issues affecting the optimal operation of Distributed Energy Resources (DERs). Virtual Power Plants (VPPs) aggregate different Energy Nodes (ENs) to enable them to participate in energy markets in an aggregated way. This participation requires very accurate forecast services to trade off between the revenue at bidding time, by making more aggresive bids, and the reduction of penalties at operation time due to deviations from the commitment. In this paper, a stochastic optimization layer is built over a Model Predictive Control (MPC) kernel to define a Stochastic Model Predictive Control (SMPC) scheme by combining Chance-Constrained (CC) and Machine Learning (ML) to handle the uncertainty related to generation and load profiles at optimization time.This technique can be applied to participate simultaneously in different energy markets depending on the configuration or capacity of the VPP, and according to the business model of the system operator. More accurate optimizations allow more profitable operations and more flexibility to redistribute the energy allocation to offer different energy services. The results are satisfactory as this scheme works better than the deterministic approach in terms of penalty reduction for most of the cases.

Hierarchical Concept-Driven Language Model

For guiding natural language generation, many semantic-driven methods have been proposed. While clearly improving the performance of the end-to-end training task, these existing semantic-driven methods still have clear limitations: for example, (i) they only utilize shallow semantic signals (e.g., from topic models) with only a single stochastic hidden layer in their data generation process, which suffer easily from noise (especially adapted for short-text etc.) and lack of interpretation; (ii) they ignore the sentence order and document context, as they treat each document as a bag of sentences, and fail to capture the long-distance dependencies and global semantic meaning of a document. To overcome these problems, we propose a novel semantic-driven language modeling framework, which is a method to learn a Hierarchical Language Model and a Recurrent Conceptualization-enhanced Gamma Belief Network, simultaneously. For scalable inference, we develop the auto-encoding Variational Recurrent Inference, allowing efficient end-to-end training and simultaneously capturing global semantics from a text corpus. Especially, this article introduces concept information derived from high-quality lexical knowledge graph Probase, which leverages strong interpretability and anti-nose capability for the proposed model. Moreover, the proposed model captures not only intra-sentence word dependencies, but also temporal transitions between sentences and inter-sentence concept dependence. Experiments conducted on several NLP tasks validate the superiority of the proposed approach, which could effectively infer meaningful hierarchical concept structure of document and hierarchical multi-scale structures of sequences, even compared with latest state-of-the-art Transformer-based models.

IB-GAN: Disentangled Representation Learning with Information Bottleneck Generative Adversarial Networks

We propose a new GAN-based unsupervised model for disentangled representation learning. The new model is discovered in an attempt to utilize the Information Bottleneck (IB) framework to the optimization of GAN, thereby named IB-GAN. The architecture of IB-GAN is partially similar to that of InfoGAN but has a critical difference; an intermediate layer of the generator is leveraged to constrain the mutual information between the input and the generated output. The intermediate stochastic layer can serve as a learnable latent distribution that is trained with the generator jointly in an end-to-end fashion. As a result, the generator of IB-GAN can harness the latent space in a disentangled and interpretable manner. With the experiments on dSprites and Color-dSprites dataset, we demonstrate that IB-GAN achieves competitive disentanglement scores to those of state-of-the-art β-VAEs and outperforms InfoGAN. Moreover, the visual quality and the diversity of samples generated by IB-GAN are often better than those by β-VAEs and Info-GAN in terms of FID score on CelebA and 3D Chairs dataset.

A Physically Based Stochastic Boundary Layer Perturbation Scheme. Part I: Formulation and Evaluation in a Convection-Permitting Model

AbstractWe present a simple, physically consistent stochastic boundary layer scheme implemented in the Met Office’s Unified Model. It is expressed as temporally correlated multiplicative Poisson noise with a distribution that depends on physical scales. The distribution can be highly skewed at convection-permitting scales (horizontal grid lengths around 1 km) when temporal correlation is far more important than spatial. The scheme is evaluated using small ensemble forecasts of two case studies of severe convective storms over the United Kingdom. Perturbations are temporally correlated over an eddy-turnover time scale, and may be similar in magnitude to or larger than the mean boundary layer forcing. However, their mean is zero and hence they, in practice, they have very little impact on the energetics of the forecast, so overall domain-averaged precipitation, for example, is essentially unchanged. Differences between ensemble members grow; after around 12 h they appear to be roughly saturated; this represents the time scale to achieve a balance between addition of new perturbations, perturbation growth, and dissipation, not just saturation of initial perturbations. The scheme takes into account the area chosen to average over, and results are insensitive to this area at least where this remains within an order of magnitude of the grid scale.

A Physically Based Stochastic Boundary Layer Perturbation Scheme. Part II: Perturbation Growth within a Superensemble Framework

AbstractConvection-permitting forecasts have improved the forecasts of flooding from intense rainfall. However, probabilistic forecasts, generally based upon ensemble methods, are essential to quantify forecast uncertainty. This leads to a need to understand how different aspects of the model system affect forecast behavior. We compare the uncertainty due to initial and boundary condition (IBC) perturbations and boundary layer turbulence using a superensemble (SE) created to determine the influence of 12 IBC perturbations versus 12 stochastic boundary layer (SBL) perturbations constructed using a physically based SBL scheme. We consider two mesoscale extreme precipitation events. For each, we run a 144-member SE. The SEs are analyzed to consider the growth of differences between the simulations, and the spatial structure and scales of those differences. The SBL perturbations rapidly spin up, typically within 12 h of precipitation commencing. The SBL perturbations eventually produce spread that is not statistically different from the spread produced by the IBC perturbations, though in one case there is initially increased spread from the IBC perturbations. Spatially, the growth from IBC occurs on larger scales than that produced by the SBL perturbations (typically by an order of magnitude). However, analysis across multiple scales shows that the SBL scheme produces a random relocation of precipitation up to the scale at which the ensemble members agree with each other. This implies that statistical postprocessing can be used instead of running larger ensembles. Use of these statistical postprocessing techniques could lead to more reliable probabilistic forecasts of convective events and their associated hazards.

Deep Learning Based End-to-End Wireless Communication Systems Without Pilots

The recent development in machine learning, especially in deep neural networks (DNN), has enabled learning-based end-to-end communication systems, where DNNs are employed to substitute all modules at the transmitter and receiver. In this article, two end-to-end frameworks for frequency-selective channels and multi-input and multi-output (MIMO) channels are developed, where the wireless channel effects are modeled with an untrainable stochastic convolutional layer. The end-to-end framework is trained with mini-batches of input data and channel samples. Instead of using pilot information to implicitly or explicitly estimate the unknown channel parameters as in current communication systems, the transmitter DNN learns to transform the input data in a way that is robust to various channel conditions. The receiver consists of two DNN modules used for channel information extraction and data recovery, respectively. A bilinear production operation is employed to combine the features extracted from the channel information extraction module and the received signals. The combined features are further utilized in the data recovery module to recover the transmitted data. Compared with the conventional communication systems, performance improvement has been shown for frequency-selective channels and MIMO channels. Furthermore, the end-to-end system can automatically leverage the correlation in the channels and in the source data to improve the overall performance.

TEST PARTICLE SIMULATIONS FOR NON-MAXWELLIAN PLASMA TRANSPORT: DISCRETIZED COLLISIONAL OPERATOR

The plasma observed in modern fusion devices is very often characterized by strongly non Maxwellian distribution function. That is the direct result of inevitable application of plasma heating techniques, such as neutral beam injection (NBI) and ion/electron cyclotron resonance frequency (ICRF/ECRF) heating, which induce the non Maxwellian fast ions. Another cause of transfer from Maxwellian to non Maxwellian is the reconnection of magnetic field lines followed by formation of magnetic resonant structures like magnetic islands and stochastic layers. One of the basic approaches used to simulate fusion plasma is test particle approach based on a solution of the equations of test particle motion. To make this approach more comprehensive one should take care of plasma particle interactions, i.e. Coulomb collisions in non Maxwellian environment. In present paper the expressions for the discretized collision operator of a general Monte Carlo equivalent form in terms of expectation values and standard deviation for an arbitrary non Maxwellian bulk distribution function are derived. The modification of transport coefficients of impurity ions caused by the transition from the background Maxwellian to non Maxwellian plasma is studied by means of this discretized collision operator. On this purpose, the set of monoenergetic neon test impurities is followed in a toroidal plasma consisting of bulk deuterons and electrons. The non Maxwellian distribution of the bulk is obtained by adding a fraction of energetic particles of the same species. It is demonstrated that a change of collision frequencies of impurities takes place in presence of this energetic fraction leading to a change of impurity neoclassical transport regime.

Dynamics of a pair of point vortices and a foil with parametric excitation in an ideal fluid

In this paper we obtain equations of motion for a vortex pair and a circular foil with parametric excitation due to the periodic motion of a material point. Undoubtedly, such problems are, on the one hand, model problems and cannot be used for an exact quantitative description of real trajectories of the system. On the other hand, in many cases such 2D models provide a sufficiently accurate qualitative picture of the dynamics and, due to their simplicity, an estimate of the influence of different parameters. We describe relative equilibria that generalize Föppl solutions and collinear configurations when the material point does not move. We show that a stochastic layer forms in the neighborhood of relative equilibria in the case of periodic motion of the foil's center of mass.

Assessing energy dependence of the transport of relativistic electrons in perturbed magnetic fields with orbit-following simulations

Experimental observations, as well as theoretical predictions, indicate that the transport of energetic electrons decreases with energy. This reduction in transport is attributed to finite orbit width (FOW) effects. Using orbit-following simulations in perturbed tokamak magnetic fields that have an ideal homogeneous stochastic layer at the edge, we quantify the energy dependence of energetic electrons transport and confirm previous theoretical estimates. However, using magnetic configurations characteristic of JET disruptions, we find no reduction in runaway electron transport at higher energies, which we attribute to the mode widths being comparable to the minor radius, making the FOW effects negligible. Instead, the presence of islands and non-uniform magnetic perturbations are found to be more important. The diffusive-advective transport coefficients calculated in this work, based on simulations for electron energies 10 keV–100 MeV, can be used in integrated disruption modelling to account for the transport due to the magnetic field perturbations.

Deep Autoencoding Topic Model With Scalable Hybrid Bayesian Inference.

To build a flexible and interpretable model for document analysis, we develop deep autoencoding topic model (DATM) that uses a hierarchy of gamma distributions to construct its multi-stochastic-layer generative network. In order to provide scalable posterior inference for the parameters of the generative network, we develop topic-layer-adaptive stochastic gradient Riemannian MCMC that jointly learns simplex-constrained global parameters across all layers and topics, with topic and layer specific learning rates. Given a posterior sample of the global parameters, in order to efficiently infer the local latent representations of a document under DATM across all stochastic layers, we propose a Weibull upward-downward variational encoder that deterministically propagates information upward via a deep neural network, followed by a Weibull distribution based stochastic downward generative model. To jointly model documents and their associated labels, we further propose supervised DATM that enhances the discriminative power of its latent representations. The efficacy and scalability of our models are demonstrated on both unsupervised and supervised learning tasks on big corpora.

Quantitative analysis of the gain in probability of escaping for ideal phototactic swimmers due to chaotic dynamics.

We study the dynamics of ideal phototactic swimmers in a steady two-dimensional model flow with transport barriers. We consider a distant light source, in which case the self-propulsion velocity of the swimmers is, at any instant, along a predetermined direction. The probability of transport along that direction emerges from the competing effects of the swimmers' self-propulsion and the flow's transport barriers. For swimmers bounded to have the same time average self-propulsion speed, temporal modulation of that speed increases the probability of escaping due to the formation of a stochastic layer which fosters transport. We use separatrix-map techniques to calculate the gain in the probability of escaping.

Development of the l = 2, m = 6 heliotron magnetic system with irregular helical coil pitch

Abstract In the present paper the numerical calculations of magnetic field properties of the l = 2, m = 6 heliotron with irregular helical coil pitch, the magnetic system model of which significantly differs from the previous model, is carried out. The new model has the doubled length helical coil section with reduced helical coil pitch (doubled length L-section). A vertical magnetic field compensation scheme, elaborated for the operating l = 2, m = 2 torsatron U-2M with the additional toroidal magnetic field coils, has been applied to compensate the vertical magnetic field of the l = 2, m = 6 heliotron with doubled length L-section. The calculation results show that due to the defined relation between the ampere-turns in the compensation and in the helical coils the developed l = 2, m = 6 heliotron magnetic system with irregular helical coil pitch can provide some increase of the magnetic surface existence region. Also the pinch of the closed magnetic surface region and of the outer boundary of the stochastic field line layer takes place within lengthened L-section.

Nonlinear modeling of the effect of n = 2 resonant magnetic field perturbation on peeling-ballooning modes in KSTAR

Using the nonlinear 3D MHD code JOREK with reduced MHD equations (visco-resistive MHD), we have successfully simulated a recent n = 2 resonant magnetic perturbation (RMP)-driven edge localized mode (ELM) suppression in KSTAR. We have found that such ELM suppression has been attributable not only to the degraded pedestal but also to the direct coupling between the peeling-ballooning mode (PBM) and RMP-driven plasma response. Notably, the pedestal pressure gradient is reduced as the radial transport is enhanced because of the formation of the stochastic layer and increased convection fluxes due to tearing and the kink-peeling mode driven by RMPs. The increased transport in the stochastic layer is due to the parallel transport across the stochastic fields, described by the Braginskii model in the simulation. While the linear stability of the PBM improves owing to the degraded pedestal, it is not a sole contributor to ELM suppression, in that the nonlinear mode coupling plays a more critical role. This outcome is consistent with previous studies where mode coupling affects the ELM mitigation or suppression. In addition, PBM locking has been numerically achieved during the ELM suppression phase, which may support the relationship between at the pedestal and the onset of ELM suppression. We suggest that PBM locking can enhance the mode interactions between RMPs and PBMs, which is significant for ELM suppression.

Motion of a Smooth Foil in a Fluid under the Action of External Periodic Forces. I

This paper considers the plane-parallel motion of an elliptic foil in a fluid with a nonzero constant circulation under the action of external periodic forces and torque. The existence of the first integral is shown for the case in which there is no external torque and an external force acts along one of the principal axes of the foil. It is shown that, in the general case, in the absence of friction, an extensive stochastic layer is observed for the period advance map. When dissipation is added to the system, strange attractors can arise from the stochastic layer.