Time-evolving Patterns Research Articles

Deep Structured State Learning for Next-Period Recommendation.

User activities in real systems are usually time-sensitive. But, most of the existing sequential models in recommender systems neglect the time-related signals. In this article, we find that users' temporal behaviors tend to be driven by their regularly changing states, which provides a new perspective on learning users' dynamic preference. However, since the individual state is usually latent, the event space is high dimensional, and meanwhile, temporal dependency of states is personalized and complex; it is challenging to represent, model, and learn the time-evolving patterns of user's state. Focusing on these challenges, we propose a deep structured state learning (DSSL) framework, which is able to learn the representation of temporal states and the complex state dependency for time-sensitive recommendation. Extensive experiments demonstrate that the DSSL achieves competitive results on four real-world recommendation datasets. Furthermore, experiments also show some interesting rules for designing the state dependency network.

Data-driven model reduction of agent-based systems using the Koopman generator.

The dynamical behavior of social systems can be described by agent-based models. Although single agents follow easily explainable rules, complex time-evolving patterns emerge due to their interaction. The simulation and analysis of such agent-based models, however, is often prohibitively time-consuming if the number of agents is large. In this paper, we show how Koopman operator theory can be used to derive reduced models of agent-based systems using only simulation data. Our goal is to learn coarse-grained models and to represent the reduced dynamics by ordinary or stochastic differential equations. The new variables are, for instance, aggregated state variables of the agent-based model, modeling the collective behavior of larger groups or the entire population. Using benchmark problems with known coarse-grained models, we demonstrate that the obtained reduced systems are in good agreement with the analytical results, provided that the numbers of agents is sufficiently large.

A Network-Design Analysis of Airline Business Model Adaptation in the Face of Competition and Consolidation

By focusing on the intrinsic relationship between business models and network configurations in the airline industry, this paper develops a two-stage methodology to estimate the strategic drivers of network design of the major carriers in Brazil. The empirical approach decomposes their domestic network-building rationales into the ones adopted by virtual archetypical carriers. We consider the previously conceived low-cost, full-service, and regional carrier archetypes. Our main contribution is the development of a model that allows airlines’ networks to be strategically designed in a time-evolving pattern, reflecting a dynamically chosen blend of these archetypes. Moreover, we also consider the effects that mergers and acquisitions may have had in inducing changes in these blends. Our results suggest that all analyzed airlines have repositioned themselves through their trajectories to adopt a hybrid configuration, aiming at the intersection of at least two archetypical network-design rationales. Besides, the effects of consolidations point to certain diversions of the acquiring airlines’ domestic network-building rationales toward the ones of the acquired carriers, providing evidence that the consolidations may have served as stepping stones for market-repositioning moves.

Time-varying volatility spillover between Chinese fuel oil and stock index futures markets based on a DCC-GARCH model with a semi-nonparametric approach

This paper reveals some new evidence on the volatility spillover between fuel oil and stock index futures markets in China by considering its time-variant feature. Time-varying effect is specified by the Legendre polynomials in a DCC GARCH model. Moreover, the semi-nonparametric (SNP) approach accounting for marginal excess kurtosis and asymmetry is applied. We find different time-evolving patterns of volatility spillover across multiple periods due to the structural breaks. The direction of mean volatility spillover is found to be bilateral. Moreover, the strength of mean volatility spillover shows that the effect from stock index to fuel oil futures is much stronger than the other way around. This implies that the stock index futures market plays a leading role in information processing. These findings have important implications for the fuel oil stakeholders as well as risk management.

A model for phased evacuations for disasters with spatio-temporal randomness

ABSTRACTThis research presents an operable zoning approach for phased evacuations adapted to disasters with spatio-temporal randomness. As a criterion for prioritizing evacuation order, evacuation risk is formulated by taking into consideration the estimated residual evacuation horizon associated with the characteristics of the disaster, the estimated time-dependent capacities of outbound lanes related to network supply, and the time-dependent evacuation demand of an evacuation unit. The modeling of the subzone determined for phased evacuation is based on rescue demand, the characteristics of the disaster, and network supply, and is labeled as a high-risk evacuation zone (HEZ). The range of HEZ features a time-evolving pattern in accordance with phased evacuation. The zone partition paradigm can be seamlessly applied to different types of disasters, especially those with high spatio-temporal randomness. It also provides a generalizable approach for subzone partitioning in phased evacuation by minimizing evacuation risk. The proposed approach is examined on numerical experiments through the road network of Xi’an, China, the results of which highlight its strength in increased adaptability to the dynamics of disaster impact and improved performance in evacuation operation.

Modelling persistence in conditional volatility of asset returns

Studies on volatility forecasting models indicate superior performance of generalised autoregressive conditional heteroscedasticity (GARCH) type models in the modelling conditional variance of asset returns. The utility of GARCH parameters lies in their ability in explaining the persistence of the conditional variance. The estimate of persistence provides a quantitative measure of the impact of a sudden significant change in the asset return on its future volatility. This study attempts to analyse the magnitude and time-evolving pattern in the persistence of conditional volatility using data on S&P CNX NIFTY 50 (henceforth, Nifty) benchmark index. The GARCH (1, 1) model is fitted on daily returns and a simple iterative scheme is used to re-estimate GARCH parameters on samples of different sizes and different time periods. The GARCH estimates obtained through repeated estimations furnish empirical evidence on the nature and consistency of the persistence parameter. Findings of the study confirm high persistence in the volatility process and indicate a positive relationship between the conditional volatility and volatility persistence.

Modelling persistence in conditional volatility of asset returns

Studies on volatility forecasting models indicate superior performance of generalised autoregressive conditional heteroscedasticity (GARCH) type models in the modelling conditional variance of asset returns. The utility of GARCH parameters lies in their ability in explaining the persistence of the conditional variance. The estimate of persistence provides a quantitative measure of the impact of a sudden significant change in the asset return on its future volatility. This study attempts to analyse the magnitude and time-evolving pattern in the persistence of conditional volatility using data on S&P CNX NIFTY 50 (henceforth, Nifty) benchmark index. The GARCH (1, 1) model is fitted on daily returns and a simple iterative scheme is used to re-estimate GARCH parameters on samples of different sizes and different time periods. The GARCH estimates obtained through repeated estimations furnish empirical evidence on the nature and consistency of the persistence parameter. Findings of the study confirm high persistence in the volatility process and indicate a positive relationship between the conditional volatility and volatility persistence.

Improvisation and the self-organization of multiple musical bodies.

Understanding everyday behavior relies heavily upon understanding our ability to improvise, how we are able to continuously anticipate and adapt in order to coordinate with our environment and others. Here we consider the ability of musicians to improvise, where they must spontaneously coordinate their actions with co-performers in order to produce novel musical expressions. Investigations of this behavior have traditionally focused on describing the organization of cognitive structures. The focus, here, however, is on the ability of the time-evolving patterns of inter-musician movement coordination as revealed by the mathematical tools of complex dynamical systems to provide a new understanding of what potentiates the novelty of spontaneous musical action. We demonstrate this approach through the application of cross wavelet spectral analysis, which isolates the strength and patterning of the behavioral coordination that occurs between improvising musicians across a range of nested time-scales. Revealing the sophistication of the previously unexplored dynamics of movement coordination between improvising musicians is an important step toward understanding how creative musical expressions emerge from the spontaneous coordination of multiple musical bodies.

Adjoint sensitivities of sub-ice-shelf melt rates to ocean circulation under the Pine Island Ice Shelf, West Antarctica

AbstractWe investigate the sensitivity of sub-ice-shelf melt rates under Pine Island Ice Shelf, West Antarctica, to changes in the oceanic state using an adjoint ocean model that is capable of representing the flow in sub-ice-shelf cavities. The adjoint code is based on algorithmic differentiation (AD) of the Massachusetts Institute of Technology’s ocean general circulation model (MITgcm). The adjoint model was extended by adding into the AD process the corresponding sub-ice-shelf cavity code, which implements a three-equation thermodynamic melt-rate parameterization to infer heat and freshwater fluxes at the ice-shelf/ocean boundary. The inferred sensitivities reveal dominant timescales of 30–60 days over which the shelf exit is connected to the deep interior via advective processes. They exhibit rich three-dimensional time-evolving patterns that can be understood in terms of a combination of the buoyancy forcing by inflowing water masses, the cavity geometry and the effect of rotation and topography in steering the flow in the presence of prominent features in the bedrock bathymetry. Dominant sensitivity pathways are found over a sill, as well as ‘shadow regions’ of very low sensitivities. To the extent that these transient patterns are robust they carry important information for decision-making in observation deployment and monitoring.

Anthropogenic Influence on Long Return Period Daily Temperature Extremes at Regional Scales

Abstract Observed 1961–2000 annual extreme temperatures, namely annual maximum daily maximum (TXx) and minimum (TNx) temperatures and annual minimum daily maximum (TXn) and minimum (TNn) temperatures, are compared with those from climate simulations of multiple model ensembles with historical anthropogenic (ANT) forcing and with combined anthropogenic and natural external forcings (ALL) at both global and regional scales using a technique that allows changes in long return period extreme temperatures to be inferred. Generalized extreme value (GEV) distributions are fitted to the observed extreme temperatures using a time-evolving pattern of location parameters obtained from model-simulated extreme temperatures under ANT or ALL forcing. Evaluation of the parameters of the fitted GEV distributions shows that both ANT and ALL influence can be detected in TNx, TNn, TXn, and TXx at the global scale over the land areas for which there are observations, and also regionally over many large land areas, with detection in more regions in TNx. Therefore, it is concluded that the influence of anthropogenic forcing has had a detectable influence on extreme temperatures that have impacts on human society and natural systems at global and regional scales. External influence is estimated to have resulted in large changes in the likelihood of extreme annual maximum and minimum daily temperatures. Globally, waiting times for extreme annual minimum daily minimum and daily maximum temperature events that were expected to recur once every 20 yr in the 1960s are now estimated to exceed 35 and 30 yr, respectively. In contrast, waiting times for circa 1960s 20-yr extremes of annual maximum daily minimum and daily maximum temperatures are estimated to have decreased to fewer than 10 and 15 yr, respectively.

Simulations of spiral galaxies with an active potential: molecular cloud formation and gas dynamics

We describe simulations of the response of a gaseous disc to an active spiral potential. The potential is derived from an N-body calculation and leads to a multi-armed time-evolving pattern. The gas forms long spiral arms typical of grand design galaxies, although the spiral pattern is asymmetric. The primary difference from a grand-design spiral galaxy, which has a consistent 2/4-armed pattern, is that instead of passing through the spiral arms, gas generally falls into a developing potential minimum and is released only when the local minimum dissolves. In this case, the densest gas is coincident with the spiral potential, rather than offset as in the grand-design spirals. We would there fore expect no offset between the spiral shock and star formation, and no obvious co-rotation radius. Spurs which occur in grand-design spirals when large clumps are sheared off leaving the spiral arms, are rare in the active, time-evolving spiral reported here. Instead, large branches are formed from spiral arms when the underlying spiral potential is dissolving due to the N-body dynamics. We find that the molecular cloud mass spectrum for the active potential is similar to that for clouds in grand design calculations, depending primarily on the ambient pressure rather than the nature of the potential. The largest molecular clouds occur when spiral arms collide, rather than by agglomeration within a spiral arm.

Computational and experimental study of turbulent flow in a 0.4-scale water model of a continuous steel caster

Single-phase turbulent flow in a 0.4-scale water model of a continuous steel caster is investigated using large eddy simulations (LES) and particle image velocimetry (PIV). The computational domain includes the entire submerged entry nozzle (SEN) starting from the tundish exit and the complete mold region. The results show a large, elongated recirculation zone in the SEN below the slide gate. The simulation also shows that the flow exiting the nozzle ports has a complex time-evolving pattern with strong cross-stream velocities, which is also seen in the experiments. With a few exceptions, which are probably due to uncertainties in the measurements, the computed flow field agrees with the measurements. The instantaneous jet is seen to have two typical patterns: a wobbling “stair-step” downward jet and a jet that bends upward midway between the SEN and the narrow face. A 51-second time average suppressed the asymmetries between the two halves of the upper mold region. However, the instantaneous velocity fields can be very different in the two halves. Long-term flow asymmetry is observed in the lower region. Interactions between the two halves cause large velocity fluctuations near the top surface. The effects of simplifying the computational domain and approximating the inlet conditions are presented.