Bayesian model averaging of load demand forecasts from neural network models

Creating a set of a number of neural network (NN) models in an ensemble and accumulating them can achieve better overview capability as compared to single neural network. Neural network ensembles are designed to provide solutions to particular problems. Many researchers and academicians have adopted this NN ensemble technique, especially in machine learning, and has been applied in various fields of engineering, medicine and information technology. This paper present a robust aggregation methodology for load demand forecasting based on Bayesian Model Averaging of a set of neural network models in an ensemble. This paper estimate a vector of coefficient for individual NN models' forecasts using validation data-set. These coefficients, also known as weights, are equal to posterior probabilities of the models generating the forecasts. These BMA weights are then used in combining forecasts generated from NN models with test data-set. By comparing the Bayesian results with the Simple Averaging method, it was observed that benefits are obtained by utilizing an advanced method like BMA for forecast combinations.

Network traffic clustering using random forest proximities

The recent years have seen extensive work on statistics-based network traffic classification using machine learning (ML) techniques. In the particular scenario of learning from unlabeled traffic data, some classic unsupervised clustering algorithms (e.g. K-Means and EM) have been applied but the reported results are unsatisfactory in terms of low accuracy. This paper presents a novel approach for the task, which performs clustering based on Random Forest (RF) proximities instead of Euclidean distances. The approach consists of two steps. In the first step, we derive a proximity measure for each pair of data points by performing a RF classification on the original data and a set of synthetic data. In the next step, we perform a K-Medoids clustering to partition the data points into K groups based on the proximity matrix. Evaluations have been conducted on real-world Internet traffic traces and the experimental results indicate that the proposed approach is more accurate than the previous methods.

An audio signal based model for condition monitoring of sheet metal stamping process

Tool condition monitoring is an important factor in ensuring manufacturing efficiency and product quality. Audio signal based methods are a promising technique for condition monitoring. However, the influence of interfering signals and background noise has hindered the use of this technique in production sites. Blind signal separation (BSS) has the potential to solve this problem by recovering the signal of interest out of the observed mixtures, given that the knowledge about the BSS model is available. In this paper, we discuss the development of the BSS model for sheet metal stamping with a mechanical press system, so that the BSS techniques based on this model can be developed in future. This involves conducting a set of specially designed machine operations and developing a novel signal extraction technique. Also, the link between stamping process conditions and the extracted audio signal associated with stamping was successfully demonstrated by conducting a series of trials with different lubrication conditions and levels of tool wear.

Adding a new anticoagulant or antiplatelet agent for patient receiving aspirin after an acute coronary syndrome? - Results from a pairwise and network meta-analysis of randomized-controlled trials

 Objectives: To synthesize the efficacy and safety outcomes from randomized-controlled trials (RCTs) regarding new oral anticoagulant, protease-activated receptor-1 (PAR-1) antagonist, and warfarin adjunctive to aspirin for patients after acute coronary syndrome (ACS) via pair-wise and network meta-analyses.
Methods: A comprehensive literature search was performed in Embase, Medline, Cochrane Library Web of Knowledge, and Scopus. The pair-wise meta-analysis was undertaken respectively to each agent/treatment category via Revmen 5.1. In order to estimate the relative efficacy of each agent/treatment category whilst preserving the randomized comparisons within each trial, a Bayesian network meta-analysis was conducted in WinBUGS using both fixed- and random-effects model. Covariate analysis was performed to explore the effects of length of follow-up and age of subject on the final results.
Results: In total, 23 RCTs were included in the meta-analysis. As shown by the results (OR,95%CI) for the pair-wise meta-analysis, new oral anticoagulants (0.85, [0.78, 0.93] and 3.04, [2.21, 4.19]), PAR-1 antagonists (0.80, [0.52, 1.22] and 1.55, [1.25, 1.93]) and warfarin (0.87, [0.74, 1.02] and 1.77, [1.46, 2.14]) might be able to provide better outcome in the incidences of major adverse events (MAE) but with higher bleeding risk comparing to aspirin treatment alone. Based on the model fit assessment, the random-effects model was adopted. The network meta-analysis (treatment effect comparing to aspirin lone) identified ximelagatran (-0.3044, [-0.8601, 0.2502]), dabigatran (-0.2144, [-0.8666, 0.4525]), rivoroxaban (-0.2179, [-0.5986, 0.1628]) and vorapaxar (-0.2272, [-0.81, 0.1664]) produced better improvements in MAE incidences whereas vorapaxar (0.3764, [-0.4444, 1.124]), warfarin (0.663, [0.3375, 1.037]), ximelagatran (0.7509, [-0.4164, 2.002]) and apixaban (0.8594, [-0.0049, 1.7]) produced less major bleeding events. The indirect comparisons among drug category (difference in incidence comparing to aspirin lone) showed new oral anticoagulants (-0.1974, [-0.284, -0.111]) and PAR-1 antagonists (-0.1239, [-0.215, -0.033]) to besuperior to warfarin (-0.1004, [-0.166, -0.035]) in the occurrences of MAE whereas PAR-1 antagonists (0.4292, [0.2123, 0.6476]) afforded better outcomes in major bleeding events against warfarin (0.5742, [0.3889, 0.7619]) and new oral anticoagulants (1.169, [0.8667, 1.485]).
Conclusion: Based on the study results, we cannot recommend the routine administration of new oral anticoagulant as add-on treatment for patients after ACS. However, for ACS patients comorbid with atrial fibrillation, new oral anticoagulant might be superior to warfarin in both efficacy and safety outcomes.

A distributed gateway selection algorithm for UAV networks

In recent years, unmanned aerial vehicle (UAV) has been widely adopted in military and civilian applications. For small UAVs, cooperation based on communication networks can effectively expand their working area. Although the UAV networks are quite similar to the traditional mobile ad hoc networks, the special characteristics of the UAV application scenario have not been considered in the literature. In this paper, we propose a distributed gateway selection algorithm with dynamic network partition by taking into account the application characteristics of UAV networks. In the proposed algorithm, the influence of the asymmetry information phenomenon on UAVs' topology control is weakened by dividing the network into several subareas. During the operation of the network, the partition of the network can be adaptively adjusted to keep the whole network topology stable even though UAVs are moving rapidly. Meanwhile, the number of gateways can be completely controlled according to the system requirements. In particular, we define the stability of UAV networks, build a network partition model, and design a distributed gateway selection algorithm. Simulation results show using our proposed scheme that the faster the nodes move in the network, the more stable topology can be found, which is quite suitable for UAV applications.

The relation between local and global influence of individuals in scale-free networks

Large-degree nodes in scale-free networks are normally responsible for large cascades of epidemics. However, recent research shows small-degree nodes can also produce large-scale epidemics in the real world. In this letter, we investigate the relation between local and global influence of individuals in scale-free network in order to theoretically explain this real-world phenomenon. The local influence of an individual corresponds to the node degree, and the global influence of an individual reflects the expected number of individuals directly or indirectly influenced by this individual in epidemics. We formalize the later as the novel epidemic betweenness concept, to mathematically estimate the global influence of individuals. Our analysis shows that the global influence follows power-law distributions in scale-free networks. We also observe that the average global influence of individuals is power-law to the degree of nodes, which well explains the reason why large-degree nodes are more likely to produce large cascades of epidemics. In addition, we discover that some smalldegree nodes also possess large global influence in terms of epidemics betweenness. This well explains the counter-intuitive phenomenon in recent research.

Unknown pattern extraction for statistical network protocol identification

The past decade has seen a lot of research on statistics-based network protocol identification using machine learning techniques. Prior studies have shown promising results in terms of high accuracy and fast classification speed. However, most works have embodied an implicit assumption that all protocols are known in advance and presented in the training data, which is unrealistic since real-world networks constantly witness emerging traffic patterns as well as unknown protocols in the wild. In this paper, we revisit the problem by proposing a learning scheme with unknown pattern extraction for statistical protocol identification. The scheme is designed with a more realistic setting, where the training dataset contains labeled samples from a limited number of protocols, and the goal is to tell these known protocols apart from each other and from potential unknown ones. Preliminary results derived from real-world traffic are presented to show the effectiveness of the scheme.

Complex network theoretical analysis on information dissemination over vehicular networks

How to enhance the communication efficiency and quality on vehicular networks is one critical important issue. While with the larger and larger scale of vehicular networks in dense cities, the real-world datasets show that the vehicular networks essentially belong to the complex network model. Meanwhile, the extensive research on complex networks has shown that the complex network theory can both provide an accurate network illustration model and further make great contributions to the network design, optimization and management. In this paper, we start with analyzing characteristics of a taxi GPS dataset and then establishing the vehicular-to-infrastructure, vehicle-to-vehicle and the hybrid communication model, respectively. Moreover, we propose a clustering algorithm for station selection, a traffic allocation optimization model and an information source selection model based on the communication performances and complex network theory.

Literacy into action: digital games as action and text in the English and literacy classroom

Using data gathered from a three-year research project exploring digital literacy and pedagogy with respect to video games, including classroom games-based pedagogy and curriculum and ethnographic research on students' digital game playing, this article locates and explores a key conceptual problem facing the incorporation of digital games into English and literacy classroom activities. This challenge is defined as "action" and refers to the non-visual and non-textual elements of gameplay. This challenge is explored both theoretically and through a practical discussion of various strategies developed by teachers in the project to approach this issue. The article draws on contemporary game studies in order to map out and highlight several key areas where action-based projects lead to critical reflection.

Platform technology towards the chemical fingerprinting methamphetamine from ephedrine pathways.

A platform technology based on two-dimensional chromatography that allows for the chemical fingerprinting of the synthetic routes of methamphetamine manufacture has been developed. Further novel clandestine synthetic chemistry pathways to methamphetamine have been investigated in collaboration with the Victoria Police Forensic Services Department.

Prediction of stress-strain behaviors in steels using an integrated constitutive, FEM and ANN model

Austenitic steels with a carbon content of 0.0037 to 0.79 wt% C are torsion tested and modeled using a physically based constitutive model and an Integrated Phenomenological and Artificial neural Network (IPANN) model. The prediction of both the constitutive and IPANN models on steel 0.017 wt% C is then evaluated using a finite element (FEM) code ABAQUS with different reduction in the thickness after rolling through one roll stand. It is found that during the rolling process, the prediction accuracy of the reaction force from FEM simulation for both constitutive and IPANN models depends on the strain achieved (average reduction in thickness). By integrating FEM into IPANN model and introducing the product of strain and stress as an input of the ANN model, the accuracy of this integrated FEM and IPANN model is higher than either the constitutive or IPANN model.

Cold rolling mill thickness control using the cascade-correlation neural network

The improvements in thickness accuracy of a steel strip produced by a tandem cold-roIling mill are of substantial interest to the steel industry. In this paper, we designed a direct model-reference adaptive control (MRAC)  scheme that exploits the natural level of excitation existing in the closed-loop with a dynamically constructed cascade-correlation neural network (CCNN) as a controller for cold roIling mill thickness control. Simulation results show that the combination of a such a direct MRAC scheme and the dynamically constructed CCNN significantly improves the thickness accuracy in the presence of disturbances and noise in comparison with to the conventional PID controllers.

A load balancing algorithm for web based server grids

Load balance is a critical issue in distributed systems, such as server grids. In this paper, we propose a Balanced Load Queue (BLQ) model, which combines the queuing theory and hydro-dynamic theory, to model load balance in server grids. Base on the BLQ model, we claim that if the system is in the state of global fairness, then the performance of the whole system is the best. We propose a load balanced algorithm based on the model: the algorithm tries its best to keep the system in the global fairness status using job deviation. We present three strategies: best node, best neighbour, and random selection, for job deviation. A number of experiments are conducted for the comparison of the three strategies, and the results show that the best neighbour strategy is the best among the proposed strategies. Furthermore, the proposed algorithm with best neighbour strategy is better than the traditional round robin algorithm in term of processing delay, and the proposed algorithm needs very limited system information and is robust.

A dynamic ratio-based model for signalling corporate collapse

The recognition of behavioural elements in finance has caused major shifts in the analytic framework pertaining to ratio-based modeling of corporate collapse. The modeling approach so far has been based on the classical rational theory in behavioural economics, which assumes that the financial ratios (i.e., the predictors of collapse) are static over time. The paper argues that, in the absence of rational economic theory, a static model is flawed, and that a suitable model instead is one that reflects the heuristic behavioural framework, which is what characterises behavioural attributes of company directors and in turn influences the accounting numbers used in calculating the financial ratios. This calls for a dynamic model: dynamic in the sense that it does not rely on a coherent assortment of financial ratios for signaling corporate collapse over multiple time periods. This paper provides empirical evidence, using a data set of Australian publicly listed companies, to demonstrate that a dynamic model consistently outperforms its static counterpart in signaling the event of collapse. On average, the overall predictive power of the dynamic model is 86.83% compared to an average overall predictive power of 69.35% for the static model.