End-to-End Pedestrian Trajectory Forecasting with Transformer Network

Analysis of pedestrians’ motion is important to real-world applications in public scenes. Due to the complex temporal and spatial factors, trajectory prediction is a challenging task. With the development of attention mechanism recently, transformer network has been successfully applied in natural language processing, computer vision, and audio processing. We propose an end-to-end transformer network embedded with random deviation queries for pedestrian trajectory forecasting. The self-correcting scheme can enhance the robustness of the network. Moreover, we present a co-training strategy to improve the training effect. The whole scheme is trained collaboratively by the original loss and classification loss. Therefore, we also achieve more accurate prediction results. Experimental results on several datasets indicate the validity and robustness of the network. We achieve the best performance in individual forecasting and comparable results in social forecasting. Encouragingly, our approach achieves a new state of the art on the Hotel and Zara2 datasets compared with the social-based and individual-based approaches.

Method for estimating the frequency-time parameters of signals in the system of simultaneously and independently functioning generators.

The issues of assessing the current frequency of each of the total of simultaneously and independently functioning generators within the electronic system are considered. The most common case is considered, in which the deviation of the true values of the nominal parameters of generators from the estimated values of these parameters has an unknown distribution law, and random deviations of frequency from the true nominal value are subject to the normal distribution law with the corresponding variance. The proposed method of obtaining parameters estimates of both independently functioning generators (true nominal frequency values and long-term instability, accidental deviation of frequency caused by short-term instability of the generator) is based on the measurement of the signals phases of each of the generators during several measuring intervals. The first step in obtaining estimates is to line the relationship between deviations of generator parameters from nominal values and the random deviation of the generator frequency by deviation of the measured phase value from the intended nominal value. This allows the formation of a system of linear algebraic equations relative to unknown values of frequency par deviations and relative instability, random deviation of frequency. Minimally necessary number of measuring intervals estimated in the materials of the article. The method proposed in the article is a further generalization of the previously proposed method of statistical frequency stabilization of the system of simultaneously and independently functioning generators, in which, based on finding the extremum of the multidimensional likelihood function, formed under the assumption of a normal distribution law for each of the arguments, the current duration of the measurement time interval, the nominal frequency and the relative instability of each of the generators. The suggested generalization is as follows: formation of an overdetermined system of equations, the unknowns in which are the deviation of the duration of the measurement time interval from the nominal value, the deviation of the values​​of the nominal frequencies of each of the generators from the corresponding assumed values, the deviation of the values ​​of the relative instabilities of each of the generators from the corresponding assumed values; formation of a system of normal equations, based on the results of the solution of which, a joint assessment of the duration of each of the measurement time intervals, frequency deviations and relative instabilities of each of the generators from the corresponding assumed values ​​is carried out; measurements of the signal phases of each of the generators at each time interval of the measurement. The proposed generalization allows us to abandon the following assumptions: exactly known values ​​of the nominal frequency and relative instability of each of the generators; about the known distribution law of the deviation of the values​​of the specified parameters from the assumed values. At the same time, the use of the proposed method leads to the need to measure the phases of the signals of each of the generators not in one, but in several measurement intervals. On the basis of numerical modeling, an increase in the accuracy of estimating the duration of time intervals of measurements and, accordingly, the frequency of generators is analyzed. It is shown that the refinement of the values ​​of the relative frequency instability of each of the generators does not lead to a practically significant increase in the accuracy of estimates of the current values ​​of the generator frequency. A transition to a simplified likelihood function is proposed, in which only the duration of the measurement time interval and the nominal frequency of each of the generators are estimated.

Statistical analysis of Swarm satellite data for assessing the effectiveness of ionospheric precursors of earthquakes

<p>Analysing ionospheric electron density and magnetic field data from several years of the <em>Swarm</em> three-satellite mission we define a dataset of anomalies statistically.  We then use a superposed epoch approach to study the possible relation with a corresponding dataset of earthquakes occurred in the same space-time domain. Two statistical quantities <em>d</em> and <em>n</em> are then established comparing the statistics of the real analyses with simulations to assess the effectiveness of the largest concentrations of anomalies as ionospheric precursors. In detail, <em>d</em> would show how much the real maximum concentration is above the expected typical maximum concentration of a random anomaly distribution; while <em>n</em> value measures how much the largest concentration deviates with respect a typical random deviation: the larger are the <em>d</em> and <em>n</em> values, the more the results of the analysis applied to real data deviate from randomness. The best cases for which the real analyses are well distinct from random simulations are selected when <em>d</em>≥1.5, because the anomaly density is equal to or larger than 50% of random distribution, and <em>n</em>≥4, because the probability to be random is equal to or less than 0.1%.  This is the case of Y magnetic field component with a search in the Dobrovolsky area around each considered earthquake epicentre. The electron density is slightly less effective in the correlation with earthquakes, but still better than a homogeneous random distribution of anomalies.</p>

Backstepping Based Automatically Controlled Angiogenic Inhibition Therapy

In this work, we present an automatically controlled angiogenic inhibition therapy where the variation profile of the inhibitory agent is generated by a control law that is derived using a back-stepping based control methodology. The angiogenic inhibition is described by a second order model representing the dynamics of tumor and supporting vasculature volumes. Backstepping control recursively stabilizes that dynamics and generates automatic control laws that allows the tumor volume to stay at a desired value. The desired value will be kept at one thousandth of its initial value. This will lead to a very small injection at the steady state. This is important as zero injection may lead to regrow of the tumor. The results will be presented in tabular and graphical forms. Tabular results present the variation of maximum injection rate and setup time. Graphical results present the variation of tumor and supporting vasculature volumes, injection rate and the tracking error between the reference and actual tumor volumes. In addition, we will also perform a simulation to test the capability of the closed loop to accomodate the parametric uncertainties in the rate constants. The uncertainties are represented by a random deviation in the range ±10% times the nominal value of the effected parameter. The control laws will be kept the same and the simulations will be repeated by 1000 times and each result will be superimposed on the graph. The area occupied by the curves will show the relative capability of the designs.

Evacuation simulation of different flow ratios in low-density state

The relationship between the factors of formation mechanism of stratification and the pedestrian ratio in low-density state has not been analyzed by the existing human flow evacuation simulation method, so that the simulation effect is poor. Thus, the evacuation simulation method for different flow ratios in low-density state is proposed to analyze the walking characteristics of the opposite pedestrians. On the basis of the random deviation grid gas model, the view field of pedestrian is introduced as one parameter. Considering the preference characteristics of pedestrians for the movement of open areas within the view field, the improved random deviation grid gas model is constructed. Through the model, the stratification characteristics of the opposite pedestrian flow in the simple channel scene are simulated. The results show that the proposed method can reproduce the characteristics of non-layering phenomenon of opposite pedestrian flow in low-density state. According to the probability of layer formation, the density of the opposite pedestrian flow is divided into five intervals. The opposite pedestrian flow in the critical density region is metastable, and is susceptible to interference. These results are consistent with the dynamic evolution of the actual opposite pedestrian flow.

Эволюция картин оптической дифракции на неупорядоченных фотонных структурах типа поленница

Using three-dimensional direct laser writing, ordered and disordered photonic woodpile structures have been produced. An ideal woodpile is formed by layers of parallel “logs” turned through 90° with respect to logs of the previous layer. The disorder was specified by a random deviation in the angle with respect to their parallel arrangement in each layer of the woodpile. The quality of samples was tested by scanning electron microscopy. The optical diffraction patterns were studied experimentally on microsamples with different degrees of disorder and structure periods. With an increase in the degree of the disorder, the diffraction patterns changed qualitatively with the preservation of the zero diffraction order and formation of a speckle field pattern by higher diffraction orders.

Local Parametric Approach of Wireless Capsule Endoscope Localization Using Randomly Scattered Path Loss Based WCL

We propose scattered path loss based weighted centroid localization (WCL) algorithm for wireless video capsule endoscope (VCE). The main challenge in this approach is the random deviation in the measured received signal strength indicator (RSSI) caused by multipath propagation and shadowing effects of human body channel which in turn increases the localization error. To address this issue, we propose local parameter dependent path loss representation in the training phase and apply adaptive least square error (LSE) method to extract the parameters. Then, in the test phase, we estimate distance using the extracted parameters and the randomly scattered path loss. The position of capsule is estimated using non-degree based WCL followed by a calibration process. We propose suboptimal method of estimating the calibration coefficient and also compute the optimal value of coefficient analytically to set the benchmark. We develop a simulation platform using MATLAB to present the results and to verify the performance. We gradually increase the number of sensors and place them in different topologies using different dimensions. The obtained accuracy by our proposed suboptimal method of WCL is very close to the optimal benchmark for all cases. Our proposed approach also outperforms existing works in terms of localization accuracy.

The Field Confinement, Narrow Transmission Resonances, and Green Function of a Multilayered Microsphere with Metamaterial Defects

We numerically investigate the optical transmission through a compound spherical stack with conventional and metamaterial (MM) layers and also embedded MM defect. A formation of extremely narrow resonant peak with nearly complete transmittance in area of a band gap is found. We demonstrate that photon fields of certain frequencies can be strongly confined by a left-handed (LH) defect. The influence of a random deviation in the width of compound spherical layers as well the transit to the whispering gallery mode (WGM) is also discussed.