Complexity Metrics Research Articles

Permutation entropy variations in seismic noise before and after eruptive activity at Shinmoedake volcano, Kirishima complex, Japan

Permutation entropy (PE) is a complexity metric that encodes a time series into sequences of symbols and can be used to decipher between deterministic and stochastic behavior. This study investigates PE variations in seismic noise during three eruption cycles in 2011, 2017, and 2018 at Shinmoedake volcano, Japan. The volcano is monitored by a dedicated seismic network and by infrasound microphones that recorded continuously during the aforementioned eruptions. The frequency range 1–7 Hz was used in order to infer temporal changes of PE in seismic noise and minimize any human contributions. The results showed that PE values decreased before the occurrence of each eruption. By combining these results with other observations we can attribute this decrease in PE to two reasons: first, to the occurrence of volcanic tremor that is a deterministic signal, and second, to magma migration at shallower depth beneath Shinmoedake which can attenuate high-frequency seismic waves and thus result in a less stochastic signal. PE also exhibited a spike-like increase just before the onset of the three eruptions. In 2011 and 2017, this feature was probably associated with bubble growth and collapse due to the interaction between the aquifer and high temperature magma. In 2018 the aquifer had mostly evaporated; hence, the spike in PE values was likely generated by fracturing of solidified magma within the conduit as fresh magma was pushing its way upwards. These results show that PE is a potentially useful tool for monitoring seismic noise at volcanoes and can contribute toward forecasting volcanic eruptions in conjunction with other widely used methodologies.Graphical

Patient-specific quality assurance prediction models based on machine learning for novel dual-layered MLC linac.

Patient-specific quality assurance (PSQA) is an indispensable and essential procedure in radiotherapy workflow, and several studies have been done to develop prediction models based on the conventional C-arm linac of single-layered multileaf collimator (MLC) with machine learning (ML) and deep learning techniques to predict PSQA results and improve efficiency. Recently, a newly designed O-ring gantry linac "Halcyon" equipped with unique jawless stacked-and-staggered dual-layered MLC was released. However, few studies have focused on developing PSQA prediction models for this novel dual-layered MLC system. To evaluate the performance of ML to predict PSQA results of fixed field intensity-modulated radiation therapy (FF-IMRT) plans for linac equipped with dual-layered MLC. A total of 213 FF-IMRT treatment plans, including 1383 beams from various treatment sites, were selected and delivered with portal dosimetry verification on Halcyon linac. Gamma analysis was performed using 1%/1, 2%/2, and 3%/2mm criteria with a 10% threshold. The training set (TS) of ML models consisted of 1106 beams, and an independent evaluation set (ES) consisted of 277 beams. For each beam, 33 complexity metrics were extracted as input data for training models. Three ML algorithms (gradient boosting decision tree/GBDT, random forest/RF, and Poisson Lasso/PL) were utilized to build the models and predict gamma passing rates (GPRs). To improve the prediction accuracy in the rare region, a method of reweighting for TS has been performed and compared to the unweighted results. The importance of complexity metrics was studied by permuted interesting features. The GBDT model had the best performance in this study. In ES, the minimal mean prediction error for unweighted results was 1.93%, 1.16%, 0.78% under 1%/1, 2%/2, and 3%/2mm criteria, respectively, from GBDT model. Comparing to the unweighted results, the models after reweighting gained up to 30% improvement in the rare region, whereas the overall prediction error was slightly worse depending on the kind of models. For feature importance, 2 tree-based models (GBDT and RF) had in common the top 10 most important metrics as well as the same metric with the largest impact. For linac equipped with novel dual-layered MLC, the ML model based on GBDT algorithm shows a certain degree of accuracy for GPRs prediction. The specific ML model for dual-layered MLC configuration could be a useful tool for physicists detecting PSQA measurement failures.

Marked distinctions in syntactic complexity: A case of second language university learners' and native speakers' syntactic constructions.

Existing research on syntactic complexity tends to examine diversity and complexity embedded in grammatical employments which may well be witnessed in formulations of different syntactic structures. Conceivably, the subject syntactic complexity seems to be exercised mostly by non-native speakers and writers. The present study employs pseudo-longitudinal data: essays written by undergraduate students in different levels collected at the same time. Hence, it aims at investigating the mechanism involved in the L2 production of L2 learners in terms of syntactic complexity by analyzing a corpus of non-native English-speaking learners. The research inquiry is mainly focused on investigation of significant differences in terms of syntactic complexity between writing of Japanese university students and that of native speakers, probing further into the particular dimension and level of difference. The study also traces marked variations in syntactic complexity employed by Japanese university students in different grades. Nagoya Interlanguage Corpus of English (NICE) developed by Sugiura in 2015 was employed to conduct the entire research proceedings. The corpus of the subject study comprises 339 essays written by L2 EFL learners studying in a Japanese university employing a judicious selection of quantitative measures of syntactic complexity. The results exhibited a considerable statistical difference between the writing of Japanese learners and native English writers. The findings of this study provide meaningful pedagogical implications for English teachers and textbook compilers. Japanese university students in higher grades are found to be employing more complicated and diverse syntactic structures. Conforming to the already conducted research studies with almost similar objectives, this study demonstrates the significance of using both general and more particular complexity metrics to assess syntactic development in L2.

Comparison of metrics for measuring Wikipedia ecology: characteristics of self-consistent metrics for editor scatteredness and article complexity

Comparison of metrics for measuring Wikipedia ecology: characteristics of self-consistent metrics for editor scatteredness and article complexity

An independent Monte Carlo-based IMRT QA tool for a 0.35T MRI-guided linear accelerator.

To develop an independent log file-based intensity-modulated radiation therapy (IMRT) quality assurance (QA) tool for the 0.35T magnetic resonance-linac (MR-linac) and investigate the ability of various IMRT plan complexity metrics to predict the QA results. Complexity metrics related to tissue heterogeneity were also introduced. The tool for particle simulation (TOPAS) Monte Carlo code was utilized with a previously validated linac head model. A cohort of 29 treatment plans was selected for IMRT QA using the developed QA tool and the vendor-supplied adaptive QA (AQA) tool. For 27 independent patient cases, various IMRT plan complexity metrics were calculated to assess the deliverability of these plans. A correlation between the gamma pass rates (GPRs) from the AQA results and calculated IMRT complexity metrics was determined using the Pearson correlation coefficients. Tissue heterogeneity complexity metrics were calculated based on the gradient of the Hounsfield units. The median and interquartile range for the TOPAS GPRs (3%/3mm criteria) were 97.24% and 3.75%, respectively, and were 99.54% and 0.36% for the AQA tool, respectively. The computational time for TOPAS ranged from 4 to 8h to achieve a statistical uncertainty of <1.5%, whereas the AQA tool had an average calculation time of a few minutes. Of the 23 calculated IMRT plan complexity metrics, the AQA GPRs had correlations with 7 out of 23 of the calculated metrics. Strong correlations (|r|>0.7) were found between the GPRs and the heterogeneity complexity metrics introduced in this work. An independent MC and log file-based IMRT QA tool was successfully developed and can be clinically deployed for offline QA. The complexity metrics will supplement QA reports and provide information regarding plan complexity.

Complexity in sign languages

Abstract Sign languages are human communication systems that are equivalent to spoken language in their capacity for information transfer, but which use a dynamic visual signal for communication. Thus, linguistic metrics of complexity, which are typically developed for linear, symbolic linguistic representation (such as written forms of spoken languages) do not translate easily into sign language analysis. A comparison of physical signal metrics, on the other hand, is complicated by the higher dimensionality (spatial and temporal) of the sign language signal as compared to a speech signal (solely temporal). Here, we review a variety of approaches to operationalizing sign language complexity based on linguistic and physical data, and identify the approaches that allow for high fidelity modeling of the data in the visual domain, while capturing linguistically-relevant features of the sign language signal.

In-network placement of delay-constrained computing tasks in a softwarized intelligent edge

Future sixth-generation (6G) networks will rely on the synergies of edge computing and machine learning (ML) to build an intelligent edge, where communication and computing resources will be jointly orchestrated. In this work, we leverage ML algorithms to judiciously orchestrate the placement of delay-constrained computing tasks in a softwarized edge domain. A set of popular supervised learning algorithms, i.e., Decision Tree (DT), Bagged Trees (BTs), Multi-Layer Perceptron (MLP), Support Vector Machine (SVM), have been leveraged to this purpose. They are trained off-line through the results of an optimization problem targeting the minimization of the edge network resource usage while respecting the tasks’ delay constraints. Extensive simulation results are reported to showcase the performance of the considered techniques in terms of model accuracy, complexity and network-related metrics, e.g., amount of exchanged data in the edge domain. Among the compared techniques, DT and MLP are shown to be the most efficient solutions in terms of algorithm execution time, by achieving almost the same performance.

Impact of Plan Quality and Complexity Metrics on Gamma Passing Rate SIMT SRS Technique

<h3>Purpose/Objective(s)</h3> Multiple studies have reported different outcomes regarding the effect of plan complexity and quality metrics on gamma passing rate. However, there is very little information in literature regarding a single isocenter multitarget (SIMT) SRS technique. This study aims to evaluate the effect of different plan complexity metrics and quality metrics on gamma passing rate (GPR). Moreover, we aim to analyze the impact of distance to isocenter (DIS) & target size on plan quality metrics for the SIMT SRS technique. <h3>Materials/Methods</h3> 40 previously treated patient scans with 4–11 targets, an average volume of 1.2 cc and DIS range of 1.45 to 8.75 cm were used to create SIMT SRS in a treatment planning system (TPS) for a Linac using 6MV-FFF beam energy. In-house programs were developed to compute complexity metrics such as Plan averaged beam Area (PA), Plan averaged beam Irregularity (PI), Total MU, Leaf Travel (LT), mean Dose Rate variation (DR), mean Gantry Speed (GS) variation. Also, target volume, DIS, Paddick Conformity (PCI), and RTOG Conformity (RTOG_CI) indices were calculated per target & per plan. Paddick Gradient index (PGI) only was computed per plan. Patient QAs were then performed on a 4-D detector phantom using SNC software for comparison between measurements and calculated dose with objectives of 3%2mm & 10% threshold as per AAPM TG-218 recommendation. Spearman correlation was used for correlation and T-test for determining whether there are any statistically significant differences between the means of calculation per target vs per plan for PCI and RTOG_CI. <h3>Results</h3> The following results were observed: no correlation between GPR and PGI/RTOG_CI/number of targets/ target size, a positive but weak correlation between PCI & GPR (0.322, p-value = 0.023) and a negative weak correlation between DIS & GPR (-0.38, p-value = 0.01). However, a strong correlation was observed between target size & PCI/RTOG_CI/PGI (0.793, 0.719 & 0.636 respectively (p-value<0.001)), but no correlation between plan quality metrics and DIS. There was a positive weak correlation between GPR & DR/PA and a negative weak correlation between PI & GPR (0.35, 0.302 & -0.33 respectively). A moderate correlation was observed between GS & GPR (0.46, p-value= 0.0025), while no correlation was found between GPR & LT/total MU. Moreover, there was significant difference in mean PCI, RTOG_CI between per target and per plan. <h3>Conclusion</h3> GPR of SRS/SRT using single-isocenter VMAT for multiple targets is not associated with DIS, LT, total MU, and plan quality metrics. Also, DIS does not have any impact on plan quality metrics. Instead, there is a strong correlation between tumor volume & plan quality metrics, indicating volume is an important predictor for plan quality metrics. Moreover, DIS has a moderate impact on GPR, showing that a larger distance between target location and isocenter can result in lower GPR. For plan quality evaluation, calculation of PCI & RTOG_CI should be done per tumor.

The evolution of primate vocal communication: a social route

There is no doubt that different factors determine the evolutionary trajectory of a species' communication abilities. Beyond habitat, social life is likely a selection pressure. The hypothesis positing that living in a complex social system requires complex communication skills has gained ground. However, the measures used classically to tackle this question (i.e. number of call types and group size) fail to capture some of the subtleties of the evolution of communication systems. Integrative approaches comparing closely related species to highlight the possible relationship between the characteristics of their social systems, their ecological niche and associated vocal communication systems are required to further our understanding of this evolutionary puzzle. We first addressed this topic in two sympatric guenons, Diana monkeys and Campbell’s monkeys. Interspecies competition, and the niche specialisations this creates, appeared to be a key evolutionary driver of their repertoires. While most of their vocal units were shared, we found a diversification of alarm calls and socially meaningful vocal combinations, as well as a differential use of inconspicuous (less detectable) call structures. This was linked to differences in group size and degree of exposition to predators. Secondly, we compared the acoustic structures in macaques belonging to four species with contrasting social styles. We found that tolerant (Tonkean and crested) macaques displayed higher levels of vocal diversity (number of vocal units) and flexibility (degree of gradation) than intolerant (Japanese and rhesus) macaques in agonistic and affiliative (but not neutral) contexts. Lastly, because the core of communication is represented not only by what is expressed by an isolated caller, but also by the way vocal interactions are structured, we explored ‘conversational rules’ in apes. We found no relationship between classical social (group size, interaction rates) and vocal (repertoire size, call rates) complexity metrics. However, close call interaction patterns differed in line with the nature of societies. Orang-utan, the most solitary species, preferentially used repeated and isolated calls apparently outside of any vocal interactions. Chimpanzee, the most competitive species displayed a high proportion of overlapping vocalisations. Vocal turn-taking predominated in the most tolerant species (bonobo and gorilla). Even if primate vocal repertoires change slowly over evolutionary time, making them good phylogenetic indicators, socio-ecological niches may induce considerable local divergence at different levels: vocal repertoire composition, acoustic structure and contextual use of calls, and vocal interaction patterns.

Metric for Structural Complexity Assessment of Space Systems Modeled Using the System Modeling Language

A complexity metric is proposed for the quantification of system complexity using information about the composition of a system and its interactions depicted in a System Modelling Language (SysML) model. The proposed metric is adapted from the complexity metric developed for design structure matrix (DSM) applications and was modified to allow the metric to be applied at different decomposition levels and to accommodate the inclusion of external interactions. The metric was applied to three case studies: a Mars lander, a CubeSat and a spacecraft thermal control system. The proposed metric attributed a higher amount of complexity due to the interactions compared to the DSM metric. This variance resulted in instances where the results differed for the two metrics. Despite these differences, both metrics behaved similarly to changes in component or interaction complexity.

Complexity trade-offs and equi-complexity in natural languages: a meta-analysis.

In linguistics, there is little consensus on how to define, measure, and compare complexity across languages. We propose to take the diversity of viewpoints as a given, and to capture the complexity of a language by a vector of measurements, rather than a single value. We then assess the statistical support for twocontroversial hypotheses: the trade-off hypothesis and the equi-complexity hypothesis. We furnish meta-analyses of 28 complexity metrics applied to texts written in overall 80 typologically diverse languages. Thetrade-off hypothesis is partially supported, in the sense that around one third of the significant correlations between measures are negative. The equi-complexity hypothesis, on the other hand, is largely confirmed. While we find evidence for complexity differences in the domains of morphology and syntax, the overall complexity vectors of languages turn out virtually indistinguishable.

A new buffering theory of social support and psychological stress.

A dynamical model linking stress, social support, and health has been recently proposed and numerically analyzed from a classical point of view of integer-order calculus. Although interesting observations have been obtained in this way, the present work conducts a fractional-order analysis of that model. Under a periodic forcing of an environmental stress variable, the perceived stress has been analyzed through bifurcation diagrams and two well-known metrics of entropy and complexity, such as spectral entropy and C0 complexity. The results obtained by numerical simulations have shown novel insights into how stress evolves with frequency and amplitude of the perturbation, as well as with initial conditions for the system variables. More precisely, it has been observed that stress can alternate between chaos, periodic oscillations, and stable behaviors as the fractional order varies. Moreover, the perturbation frequency has revealed a narrow interval for the chaotic oscillations, while its amplitude may present different values indicating a low sensitivity regarding chaos generation. Also, the perceived stress has been noted to be highly sensitive to initial conditions for the symptoms of stress-related ill-health and for the social support received from family and friends. This work opens new directions of research whereby fractional calculus might offer more insight into psychology, life sciences, mental disorders, and stress-free well-being.

Knowledge distillation technique enabled hardware efficient OSNR monitoring from directly detected PDM-QAM signals

Deep neural network (DNN) enabled optical SNR (OSNR) monitoring is vital to facilitate the management of agile optical networks. However, both the scale size and the implementation complexity of DNNs prevent their practical application. Here, we experimentally demonstrate the knowledge distillation technique (KDT) for DNN-based OSNR monitoring from the directly detected polarization division multiplexed 16 quadrature amplitude modulation (PDM-16QAM) and PDM-64QAM signals by the use of amplitude histograms. First, both floating-point operations (FLOPs) and parameters (Params) are introduced into the fiber optic communications as the evaluation metrics for the computational complexity of DNNs. Then, the pre-trained teacher network and the to-be-used student network constitute a KDT network, while the task of the pre-trained teacher network transfers the knowledge to the student network. From the experimental results, we learn that the KDT benefits the student network and realizes a rms error (RMSE) of 0.98 dB over OSNR ranges of 14–24 dB and 23–34 dB for 10 Gbaud PDM-16QAM and PDM-64QAM signals, respectively. The Params of the student network are only 0.004M, and the FLOPs are 0.25M, which are less than that of existing OSNR monitoring schemes. Meanwhile, an RMSE improvement of 0.23 dB arising in the OSNR monitoring has been secured to the student network with the help of KDT. Therefore, we believe our proposed KDT is promising for optical performance monitoring with a lightweight DNN.

A view similarity-based shape complexity metric to guide part selection for additive manufacturing

PurposeThis paper aims to propose a view similarity-based shape complexity metric to guide part selection for additive manufacturing (AM) and advance the goals of design for AM. The metric helps to improve the selection process by objectively screening a large number of parts and identifying the parts most suited for AM and enabling experts to prioritize parts from a smaller set based on relevant subjective/contextual factors.Design/methodology/approachThe methodology involves calculating a part’s shape complexity based on the concept of view similarity, that is, the similarity of different views of the outer shape and internal cross-sectional geometry. The combined shape complexity metric (weighted sum of the external shape and internal structure complexity) has been used to rank various three dimensional (3D) models. The metric has been tested for its sensitivity to various input parameters and thresholds are suggested for effective results. The proposed metric’s applicability for part selection has also been investigated and compared with the existing metric-based part selection.FindingsThe proposed shape complexity metric can distinguish the parts of different shapes, sizes and parts with minor design variations. The method is also efficient regarding the amount of data and computation required to facilitate the part selection. The proposed method can detect differences in the mass properties of a 3D model without evaluating the modified parameters. The proposed metric is effective in initial screening of a large number of parts in new product development and for redesign using AM.Research limitations/implicationsThe proposed metric is sensitive to input parameters, such as the number of viewpoints, design orientation, image resolution and different lattice structures. To address this issue, this study suggests thresholds for each input parameter for optimum results.Originality/valueThis paper evaluates shape complexity using view similarity to rank parts for prototyping or redesigning with AM.

EEG-based methods for recovery prognosis of patients with disorders of consciousness: A systematic review

ObjectiveDisorders of consciousness (DoC) are acquired conditions of severely altered consciousness. Electroencephalography (EEG)-derived biomarkers have been studied as clinical predictors of consciousness recovery. Therefore, this study aimed to systematically review the methods, features, and models used to derive prognostic EEG markers in patients with DoC in a rehabilitation setting. MethodsWe conducted a systematic literature search of EEG-based strategies for consciousness recovery prognosis in five electronic databases. ResultsThe search resulted in 2964 papers. After screening, 15 studies were included in the review. Our analyses revealed that simpler experimental settings and similar filtering cut-off frequencies are preferred. The results of studies were categorised by extracting qualitative and quantitative features. The quantitative features were further classified into evoked/event-related potentials, spectral measures, entropy measures, and graph-theory measures. Despite the variety of methods, features from all categories, including qualitative ones, exhibited significant correlations with DoC prognosis. Moreover, no agreement was found on the optimal set of EEG-based features for the multivariate prognosis of patients with DoC, which limits the computational methods applied for outcome prediction and correlation analysis to classical ones. Nevertheless, alpha power, reactivity, and higher complexity metrics were often found to be predictive of consciousness recovery. ConclusionsThis study’s findings confirm the essential role of qualitative EEG and suggest an important role for quantitative EEG. Their joint use could compensate for their reciprocal limitations. SignificanceThis study emphasises the need for further efforts toward guidelines on standardised EEG analysis pipeline, given the already proven role of EEG markers in the recovery prognosis of patients with DoC.

Kolmogorov complexity metrics in assessing L2 proficiency: An information-theoretic approach.

Based on 774 argumentative writings produced by Chinese English as a foreign language (EFL) learners, this study examined the extent to which Kolmogorov complexity metrics can distinguish the proficiency levels of beginner, lower-intermediate, and upper-intermediate second language (L2) English learners. Kolmogorov complexity metric is a holistic information-theoretic approach, which measures three facets of linguistic complexity, i.e., overall, syntactic, and morphological complexity simultaneously. To assess its validity in distinguishing L2 proficiency, Kolmogorov complexity metric is compared with traditional syntactic and morphological complexity metrics as well as fine-grained syntactic complexity metrics. Results showed that Kolmogorov overall and syntactic complexity could significantly distinguish any adjacent pair of L2 levels, serving as the best separators explored in the present study. Neither Kolmogorov morphological complexity nor other complexity metrics at both the syntactic and morphological levels can distinguish between all pairs of adjacent levels. Results of correlation analysis showed that Kolmogorov syntactic complexity was not or weakly correlated with all the fine-grained syntactic complexity metrics, indicating that they may address distinct linguistic features and can complement each other to better predict different proficiency levels.

Towards Liveable Urban Densities Using a GIS-Based Assessment Methodology

Urban density and densi fication are hotly debated topics of sustainable urban development. On the one hand, international landmark agreements such as the Sustainable Development Goals and the New Urban Agenda highlight the importance of good quality urban environments with low pollution, easy access to green spaces, walkability, and active mobility. On the other hand, densi fication can increase pollution, decrease available green spaces, and degrade walkability by concentrating vehicles and their operations. The result can be a degradation of urban liveability, de fined as a city's capacity to promote the wellbeing of residents. Yet higher urban density can also result in more effi cient urban infrastructures and networks and provide more housing (including more aff ordable housing) in more appropriate mixed-use locations. The challenge, then, is to maintain the liveable quality of this denser urban fabric. The GIS-based approach presented in this paper uses a basic liveability assessment by calculating connectivity, greenery, and urban form complexity metrics to be employed in the context of densi fication, aiming to optimize sustainability and liveability aspects. Using our study area in Salzburg, Austria, we demonstrate how such a GIS-based liveability assessment, relying on spatial data, can aid urban planners in quantifying and achieving both urban density and liveability.

Design of fractional comprehensive learning PSO strategy for optimal power flow problems

Optimal reactive power dispatch (ORPD) is one of the paramount issue for the researchers during the investigation of power system performance in dynamic load scenarios. In this paper, a new nature inspired computing paradigm based on fractional order comprehensive learning particle swarm optimization (FO-CLPSO) is designed and implemented for solving the reactive power dispatch problems. The objective of the study is to improve the power system efficiency by reducing line losses, enhancing bus voltage profiles and reducing the operating cost of the system for different load factors. The decision variables for the fitness evaluation are the tap changer settings, generator bus voltages, fixed capacitors and flexible AC transmission systems (FACTS). The operation, validity and scalability of the FO-CLPSO are tested on standard IEEE 30 bus and IEEE-57 bus systems. The exploitation and exploration for FO-CLPSO are further extended using different fractional orders for minimization problems in ORPD to critically analyze the performance by comparing with several state of art counterpart methodologies. The stability, consistency, robustness and reliability of FO-CLPSO for the solution of ORPD problems is also substantiated through detailed statistical analyses including the development of empirical cumulative distribution functions, probability plots, box plot illustrations and histograms both for precision and complexity metrics.

Measuring redundancy: the relation between concord and complexity

AbstractSyntagmatic redundancy involves constructions in which a single meaning relates to more than one form in a clause or phrase. Redundancy has been claimed to facilitate language processing and acquisition, and to be a linguistic universal, because of these advantages. At the same time, the repeated expression of meaning has been argued to make languages more complex because of the violation of principles of economy and transparency. This paper first investigates the relation between redundancy and complexity by examining how redundancy scores on well-known metrics of complexity. This assessment shows that redundancy makes languages more complex according to metrics of objective complexity, while it has contradictory effects according to metrics of subjective complexity. To empirically test whether redundancy makes acquisition and processing easier or harder, a cross-linguistically valid way of measuring redundancy first needs to be established. The second part of the paper involves an attempt in this direction. It presents a typological study on cross-linguistic variation in four types of concord, showing that languages differ with respect to the number and types of redundancy constructions they allow. The ranking of the 50 sample languages as to their redundancy-based complexity serves as a testable hypothesis for acquisition and processing research.

Robust and Responsive Learning of Spatiotemporal Urban Traffic Flow Relationships

Spatiotemporal models that incorporate information about the relationships between traffic flows over space and time have recently become the mainstream of urban traffic forecasting. Many algorithms have been developed to learn the structure of spatiotemporal relationships – the travel time-based, cross-correlation, graphical lasso, mutual information, and random forest methods, among several others. In this paper, we address two issues of spatiotemporal relationship learning. First, we suggest the application of ensemble techniques (ranking- and voting-based) that combine the advantages of individual algorithms and allow more robust structures to be obtained for learning spatiotemporal relationships. Second, we propose strategies that enable discovery of their dynamics and provide extensive information on traffic flows for the responsive learning of spatiotemporal relationships. In addition to the widely used forecasting performance (estimated for the spatially regularised vector autoregression model, the spatiotemporal k-nearest neighbour algorithm, and the high-order graph convolutional artificial neural network), we compare the estimated spatiotemporal relationship structures using their complexity and stability metrics and demonstrate the importance of these characteristics. A multitude of numerical experiments using a large citywide traffic data set are conducted, and the results demonstrate the preferences of ensemble learning and the strengths of the proposed learning strategies.