X. Rosalind Wang

Quantifying and Tracing Information Cascades in Swarms

We propose a novel, information-theoretic, characterisation of cascades within the spatiotemporal dynamics of swarms, explicitly measuring the extent of collective communications. This is complemented by dynamic tracing of collective memory, as another element of distributed computation, which represents capacity for swarm coherence. The approach deals with both global and local information dynamics, ultimately discovering diverse ways in which an individual’s spatial position is related to its information processing role. It also allows us to contrast cascades that propagate conflicting information with waves of coordinated motion. Most importantly, our simulation experiments provide the first direct information-theoretic evidence (verified in a simulation setting) for the long-held conjecture that the information cascades occur in waves rippling through the swarm. Our experiments also exemplify how features of swarm dynamics, such as cascades’ wavefronts, can be filtered and predicted. We observed that maximal information transfer tends to follow the stage with maximal collective memory, and principles like this may be generalised in wider biological and social contexts.

Wireless Sensor Network Anomalies: Diagnosis and Detection Strategies

Wireless Sensor Networks (WSNs) can experience problems (anomalies) during deployment, due to dynamic environmental factors or node hardware and software failures. These anomalies demand reliable detection strategies for supporting long term and/or large scale WSN deployments. Several strategies have been proposed for detecting specific subsets of WSN anomalies, yet there is still a need for more comprehensive anomaly detection strategies that jointly address network, node, and data level anomalies. This chapter examines WSN anomalies from an intelligent-based system perspective, covering anomalies that arise at the network, node and data levels. It generalizes a simple process for diagnosing anomalies in WSNs for detection, localization, and root cause determination. A survey of existing anomaly detection strategies also reveals their major design choices, including architecture and user support, and yields guidelines for tailoring new anomaly detection strategies to specific WSN application requirements.

Spatiotemporal anomaly detection in gas monitoring sensor networks

In this paper, we use Bayesian Networks as a means for unsupervised learning and anomaly (event) detection in gas monitoring sensor networks for underground coal mines. We show that the Bayesian Network model can learn cyclical baselines for gas concentrations, thus reducing false alarms usually caused by flatline thresholds. Further, we show that the system can learn dependencies between changes of concentration in different gases and at multiple locations. We define and identify new types of events that can occur in a sensor network. In particular, we analyse joint events in a group of sensors based on learning the Bayesian model of the system, contrasting these events with merely aggregating single events.We demonstrate that anomalous events in individual gas data might be explained if considered jointly with the changes in other gases. Vice versa, a network-wide spatiotemporal anomaly may be detected even if individual sensor readings were within their thresholds. The presented Bayesian approach to spatiotemporal anomaly detection is applicable to a wide range of sensor networks.

Fisher information at the edge of chaos in random boolean networks

We study the order-chaos phase transition in random Boolean networks (RBNs), which have been used as models of gene regulatory networks. In particular we seek to characterize the phase diagram in information-theoretic terms, focusing on the effect of the control parameters (activity level and connectivity). Fisher information, which measures how much system dynamics can reveal about the control parameters, offers a natural interpretation of the phase diagram in RBNs. We report that this measure is maximized near the order-chaos phase transitions in RBNs, since this is the region where the system is most sensitive to its parameters. Furthermore, we use this study of RBNs to clarify the relationship between Shannon and Fisher information measures.

Towards Quantifying Interaction Networks in a Football Match

We present several novel methods quantifying dynamic interactions in simulated football games. These interactions are captured in directed networks that represent significant coupled dynamics, detected information-theoretically. The model-free approach measures information dynamics of both pair-wise players’ interactions as well as local tactical contests produced during RoboCup 2D Simulation League games. This analysis involves computation of information transfer and storage, relating the information transfer to responsiveness of the players and the team, and the information storage within the team to the team’s rigidity and lack of tactical flexibility. The resultant directed networks (interaction diagrams) and the measures of responsiveness and rigidity reveal implicit interactions, across teams, that may be delayed and/or long-ranged. The analysis was verified with a number of experiments, identifying the zones of the most intense competition and the extent of interactions.

Quantifying long-range interactions and coherent structure in multi-agent dynamics

We develop and apply several novel methods quantifying dynamic multi-agent team interactions. These interactions are detected information-theoretically and captured in two ways: via (i) directed networks (interaction diagrams) representing significant coupled dynamics between pairs of agents, and (ii) state-space plots (coherence diagrams) showing coherent structures in Shannon information dynamics. This model-free analysis relates, on the one hand, the information transfer to responsiveness of the agents and the team, and, on the other hand, the information storage within the team to the teams rigidity and lack of tactical flexibility. The resultant interaction and coherence diagrams reveal implicit interactions, across teams, that may be spatially long-range. The analysis was verified with a statistically significant number of experiments (using simulated football games, produced during RoboCup 2D Simulation League matches), identifying the zones of the most intense competition, the extent and types of interactions, and the correlation between the strength of specific interactions and the results of the matches.

Measuring Information Dynamics in Swarms

We propose a novel, information theoretic characterization of dynamics within swarms, through explicitly measuring the extent of collective communications and tracing collectivememory. These elements of distributed computation provide complementary views into the capacity for swarm coherence and reorganization. The approach deals with both global and local information dynamics ultimately discovering diverse ways in which an individual’s location within the group is related to its information processing role.

Using Echo State Networks for Anomaly Detection in Underground Coal Mines

We investigate the problem of identifying anomalies in monitoring critical gas concentrations using a sensor network in an underground coal mine. In this domain, one of the main problems is a provision of mine specific anomaly detection, with cyclical (moving) instead offlatline (static) alarm threshold levels. An additional practical difficulty in modelling a specific mine is the lack of fully labelled data of normal and abnormal situations. We present an approach addressing these difficulties based on echo state networks learning mine specific anomalies when only normal data is available. Echo state networks utilize incremental updates driven by new sensor readings, thus enabling a detection of anomalies at any time during the sensor network operation. We evaluate this approach against a benchmark - Bayesian network based anomaly detection, and observe that the quality of the overall predictions is comparable to the benchmark. However, the echo state networks maintain the same level of predictive accuracy for data from multiple sources. Therefore, the ability of echo state networks to model dynamical systems make this approach more suitable for anomaly detection and predictions in sensor networks.

Feature selection for chemical sensor arrays using mutual information.

We address the problem of feature selection for classifying a diverse set of chemicals using an array of metal oxide sensors. Our aim is to evaluate a filter approach to feature selection with reference to previous work, which used a wrapper approach on the same data set, and established best features and upper bounds on classification performance. We selected feature sets that exhibit the maximal mutual information with the identity of the chemicals. The selected features closely match those found to perform well in the previous study using a wrapper approach to conduct an exhaustive search of all permitted feature combinations. By comparing the classification performance of support vector machines (using features selected by mutual information) with the performance observed in the previous study, we found that while our approach does not always give the maximum possible classification performance, it always selects features that achieve classification performance approaching the optimum obtained by exhaustive search. We performed further classification using the selected feature set with some common classifiers and found that, for the selected features, Bayesian Networks gave the best performance. Finally, we compared the observed classification performances with the performance of classifiers using randomly selected features. We found that the selected features consistently outperformed randomly selected features for all tested classifiers. The mutual information filter approach is therefore a computationally efficient method for selecting near optimal features for chemical sensor arrays.

Optimising Sensor Layouts for Direct Measurement of Discrete Variables

An optimal sensor layout is attained when a limited number of sensors are placed in an area such that the cost of the placement is minimized while the value of the obtained information is maximized. In this paper, we discuss the optimal sensor layout design problem from first principles, show how an existing optimization criterion (maximum entropy of the measured variables) can be derived, and compare the performance of this criterion with three others that have been reported in the literature for a specific situation for which we have detailed experimental data available. This is achieved by firstly learning a spatial model of the environment using a Bayesian Network, then predicting the expected sensor data in the rest of the space, and finally verifying the predicted results with the experimental measurements. The development of rigorous techniques for optimizing sensor layouts is argued to be an essential requirement for reconfigurable and self-adaptive networks.