Convective Storm VIL and Lightning Nowcasting Using Satellite and Weather Radar Measurements Based on Multi-Task Learning Models

Convective storms and lightning are among the most important weather phenomena that are challenging to forecast.In this study,a novel multi-task learning(MTL)encoder-decoder U-net neural network was developed to forecast convective storms and lightning with lead times for up to 90 min,using GOES-16 geostationary satellite infrared brightness temperatures(IRBTs),lightning flashes from Geostationary Lightning Mapper(GLM),and vertically integrated liquid(VIL)from Next Generation Weather Radar(NEXRAD).To cope with the heavily skewed distribution of lightning data,a spatiotemporal exponent-weighted loss function and log-transformed lightning normalization approach were developed.The effects of MTL,single-task learning(STL),and IRBTs as auxiliary input features on convection and lightning nowcasting were investigated.The results showed that normalizing the heavily skew-distributed lightning data along with a log-transformation dramatically outperforms the min-max normalization method for nowcasting an intense lightning event.The MTL model significantly outperformed the STL model for both lightning nowcasting and VIL nowcasting,particularly for intense lightning events.The MTL also helped delay the lightning forecast performance decay with the lead times.Furthermore,incorporating satellite IRBTs as auxiliary input features substantially improved lightning nowcasting,but produced little difference in VIL forecasting.Finally,the MTL model performed better for forecasting both lightning and the VIL of organized convective storms than for isolated cells.

A Multi-Task Motion Generation Model that Fuses a Discriminator and a Generator

The human motion generation model can extract structural features from existing human motion capture data,and the generated data makes animated characters move.The 3D human motion capture sequences contain complex spatial-temporal structures,and the deep learning model can fully describe the potential semantic structure of human motion.To improve the authenticity of the generated human motion sequences,we propose a multi-task motion generation model that consists of a discriminator and a generator.The discriminator classifies motion sequences into different styles according to their similarity to the mean spatial-temporal templates from motion sequences of 17 crucial human joints in three-freedom degrees.And target motion sequences are created with these styles by the generator.Unlike traditional related works,our model can handle multiple tasks,such as identifying styles and generating data.In addition,by extracting 17 crucial joints from 29 human joints,our model avoids data redundancy and improves the accuracy of model recognition.The experimental results show that the discriminator of the model can effectively recognize diversified movements,and the generated data can correctly fit the actual data.The combination of discriminator and generator solves the problem of low reuse rate of motion data,and the generated motion sequences are more suitable for actual movement.

Bi-Level Programming for the Optimal Nonlinear Distance-Based Transit Fare Structure Incorporating Principal-Agent Game

The urban transit fare structure and level can largely affect passengers’travel behavior and route choices.The commonly used transit fare policies in the present transit network would lead to the unbalanced transit assignment and improper transit resources distribution.In order to distribute transit passenger flow evenly and efficiently,this paper introduces a new distance-based fare pattern with Euclidean distance.A bi-level programming model is developed for determining the optimal distance-based fare pattern,with the path-based stochastic transit assignment(STA)problem with elastic demand being proposed at the lower level.The upper-level intends to address a principal-agent game between transport authorities and transit enterprises pursing maximization of social welfare and financial interest,respectively.A genetic algorithm(GA)is implemented to solve the bi-level model,which is verified by a numerical example to illustrate that the proposed nonlinear distance-based fare pattern presents a better financial performance and distribution effect than other fare structures.

MDTCNet:Multi-Task Classifications Network and TCNN for Direction of Arrival Estimation

The direction-of-arrival(DoA) estimation is one of the hot research areas in signal processing. To overcome the DoA estimation challenge without the prior information about signal sources number and multipath number in millimeter wave system,the multi-task deep residual shrinkage network(MTDRSN) and transfer learning-based convolutional neural network(TCNN), namely MDTCNet, are proposed. The sampling covariance matrix based on the received signal is used as the input to the proposed network. A DRSN-based multi-task classifications model is first introduced to estimate signal sources number and multipath number simultaneously. Then, the DoAs with multi-signal and multipath are estimated by the regression model. The proposed CNN is applied for DoAs estimation with the predicted number of signal sources and paths. Furthermore, the modelbased transfer learning is also introduced into the regression model. The TCNN inherits the partial network parameters of the already formed optimization model obtained by the CNN. A series of experimental results show that the MDTCNet-based DoAs estimation method can accurately predict the signal sources number and multipath number under a range of signal-to-noise ratios. Remarkably, the proposed method achieves the lower root mean square error compared with some existing deep learning-based and traditional methods.

SHORT COMMUNICATION Principal-agent Relationships: A Note on Biomass Depletion

Public authorities frequently mandate public or private agencies to manage their renewable natural resources.Contrary to the agency,which is an expert in renewable natural resource management,public authorities usually ignore the sustainable level of harvest.In this note,we first model the contractual relationship between a principal,who owns the renewable natural resource,and an agent,who holds private information on its sustainable level of harvest.We then look for the Pareto-optimal allocations.In the situation of an imperfect information setting,we find that the Pareto-optimal contracting depends on the probability that the harvesting level stands outside the sustainability interval.The information rent held by the agent turns out to be unavoidable,such that stepping outside the sustainability interval implies the possibility of depletion of the renewable natural resource.This,in turn,compromises the maintenance of the ecological balance in natural ecosystems.

Nuclear mass based on the multi-task learning neural network method

The global nuclear mass based on the macroscopic-microscopic model was studied by applying a newly designed multi-task learning artificial neural network(MTL-ANN). First, the reported nuclear binding energies of 2095 nuclei(Z ≥ 8, N ≥ 8) released in the latest Atomic Mass Evaluation AME2020 and the deviations between the fitting result of the liquid drop model(LDM)and data from AME2020 for each nucleus were obtained.To compensate for the deviations and investigate the possible ignored physics in the LDM, the MTL-ANN method was introduced in the model. Compared to the single-task learning(STL) method, this new network has a powerful ability to simultaneously learn multi-nuclear properties,such as the binding energies and single neutron and proton separation energies. Moreover, it is highly effective in reducing the risk of overfitting and achieving better predictions. Consequently, good predictions can be obtained using this nuclear mass model for both the training and validation datasets and for the testing dataset. In detail, the global root mean square(RMS) of the binding energy is effectively reduced from approximately 2.4 MeV of LDM to the current 0.2 MeV, and the RMS of Sn, Spcan also reach approximately 0.2 MeV. Moreover, compared to STL, for the training and validation sets, 3-9% improvement can be achieved with the binding energy, and 20-30% improvement for S_(n), S_(p);for the testing sets, the reduction in deviations can even reach 30-40%, which significantly illustrates the advantage of the current MTL.

A multi-task deep learning model for short-term taxi demand forecasting considering spatiotemporal dependences

Short-term taxi demand forecasting is of great importance to incentivize vacant cars moving from over-supply regions to over-demand regions,which can minimize the wait time for passengers and drivers.With the consideration of spatiotemporal dependences,this study proposes a multi-task deep learning(MTDL)model to predict short-term taxi demand in multi-zone level.The nonlinear Granger causality test is applied to explore the causality relationships among various traffic zones,and long short-term memory(LSTM)is used as the core neural unit to construct the framework of the multi-task deep learning model.In addition,several hyperparameter optimization methods(e.g.,grid search,random search,Bayesian optimization,hyperopt)are used to tune the model.Using the taxi trip data in New York City for validation,the multi-task deep learning model considering spatiotemporal dependences(MTDL*)is compared with the single-task deep learning model(STDL),the full-connected multi-task deep learning model(MTDL#)and other benchmark algorithms(such as LSTM,support vector machine(SVM)and k-nearest neighbors(k-NN)).The experiment results show that the proposed MTDL model is promising to predict short-term taxi demand in multi-zone level,the nonlinear Granger causality analysis is able to capture the spatiotemporal correlations among various traffic zones,and the Bayesian optimization is superior to the other three methods,which verified the feasibility and adaptability of the proposed method.

Boosting Unsupervised Monocular Depth Estimation with Auxiliary Semantic Information

Learning-based multi-task models have been widely used in various scene understanding tasks,and complement each other,i.e.,they allow us to consider prior semantic information to better infer depth.We boost the unsupervised monocular depth estimation using semantic segmentation as an auxiliary task.To address the lack of cross-domain datasets and catastrophic forgetting problems encountered in multi-task training,we utilize existing methodology to obtain redundant segmentation maps to build our cross-domain dataset,which not only provides a new way to conduct multi-task training,but also helps us to evaluate results compared with those of other algorithms.In addition,in order to comprehensively use the extracted features of the two tasks in the early perception stage,we use a strategy of sharing weights in the network to fuse cross-domain features,and introduce a novel multi-task loss function to further smooth the depth values.Extensive experiments on KITTI and Cityscapes datasets show that our method has achieved state-of-the-art performance in the depth estimation task,as well improved semantic segmentation.

Value of subjective performance evaluation： A theoretical perspective

To break through the academic research tradition of isolating subjective performance evaluation （SPE） with objective evaluation, and enhance the application and explanation power of analytical research findings, this paper adopts a one-period principal-agent model, and solves programming problems by internalizing validity conditions for SPE as constraints according to revelation principle. We find that both SPE and its objectivization can achieve Pareto improvement for both the principal and the agent. Our findings have a general application beyond bonus contracts; it not only explains the pervasive existence of SPE, but also helps to interpret phenomena of the objectification of SPE.

Understanding the deterrence effect of punishment for marine information security policies non-compliance

In the organizational setting of marine engineering,a significant number of information security incidents have been arised from the employees’failure to comply with the information security policies(ISPs).This may be treated as a principal-agent problem with moral hazard between the employer and the employee for the practical compliance effort of an employee is not observable without high cost-.On the other hand,according to the deterrence theory,the employer and the employee are inherently self-interested beings.It is worth examining to what extent the employee is self-interested in the marine ISPs compliance context.Moreover,it is important to clarify the proper degree of severity of punishment in terms of the deterrent effect.In this study,a marine ISPs compliance game model has been proposed to evaluate the deterrence effect of punishment on the non-compliance behavior of employee individuals.It is found that in a non-punishment contract,the employee will decline to comply with the marine ISPs;but in a punishment contract,appropriate punishment will lead her to select the marine ISPs compliance effort level expected by the employer,and cause no potential backfire effect.

Understanding how heterogeneous agents affect Principal's returns:Perspectives from short-termism and Bayesian learning

We consider a general framework of optimal contract design under the heterogeneity and short-termism of agents.Our research shows that the optimal contract must weigh the agent's information rent,incentive cost,and benefit to overcome the contract's adverse selection and moral hazards.Agents with higher moral levels were more likely to choose higher effort and lower manipulation.Simultaneously,the principal offers lower incentives and receives more significant payoff.We also extend our model to investigate the benefits of Bayesian learning.Furthermore,we compare the principal's returns in general and learning models and find that the learning contract can bring more profit to the principal。