No abstract

Decision support systems (DSSs) are playing an increasingly important role in the characterization of suspicious activities in an area of interest given their proved ability to turn vast amounts of raw data into actionable intelligence that is easy to understand by human operators. Although risk management is an integral component of the decision making process that directly contributes towards improved situational awareness and response assessment, an active end-to-end consideration of the underlying risk sources in the environment is still an important feature that most DSSs currently lack. Additionally, deciding on an appropriate course of action (COA) to mitigate emerging threats in the system is a challenging task even for domain experts given that (1) the number of potential responses to analyze could be overwhelmingly large; (2) seldom are those responses judged in terms of the risks associated with their enactment and (3) assessing the effectiveness of the potential responses in the real world is usually time-consuming and simulation-driven.In this paper, we formalize the adaptation of a recently proposed Risk Management Framework to account for behavioral intents associated with the objects of interest (OOIs) in the monitoring environment and their link to automatic response generation. The intent of the objects is inferred from high-level cognitive and behavioral knowledge in the form of anomalies. When an OOI has crossed a permissible risk threshold, we demonstrate how responses to that situation can be automatically elicited by the COA recommendation module of a risk-aware DSS. Multicriteria decision analysis (MCDA) is used to judge a diverse set of plausible responses according to different operational objectives. We illustrate the application of the proposed framework in the context of maritime surveillance operations by triggering a corporate search for a missing vessel. To the best of our knowledge, this is the first time that risk features are synthesized from anomalies and integrated into a more comprehensive RMF engine for knowledge (response) elicitation.Index Terms-situational awareness; risk management; intent and threat assessment; course of action recommendation; decision support systems; multicriteria decision making; high-level information fusion

Photonic crystal(PhC) waveguides are used for a wide range of applications with diverse performance metrics. A waveguide optimised for one application may not be suitable for others and no one-size-fits-all solution exists. Therefore each application requires a specialised waveguide design, a computationally and time intensive process. Here, we present a hybrid, multi-objective optimisation routine for PhC waveguides, to efficiently guide the device design. The algorithm can be configured to optimise for a wide range of performance metrics and applications. We demonstrate optimisations for three different applications: slow light performance, propagation loss due to fabrication disorder and delay line applications. For each optimisation target, our routine quickly finds practical waveguide designs (<48 h, on a laptop computer) that match or exceed the performance of state-of-the-art devices designed by the community over the last 10 years. This is also the first time that scattering loss from fabrication disorder has been incorporated into an optimisation algorithm, ensuring realistic predictions of a PhC waveguide design's practical performance.

No abstract

This paper explores the recently proposed Graph Convolutional Network architecture proposed in (Kipf & Welling, 2016) The key points of their work is summarized and their results are reproduced. Graph regularization and alternative graph convolution approaches are explored. I find that explicit graph regularization was correctly rejected by (Kipf & Welling, 2016). I attempt to improve the performance of GCN by approximating a k-step transition matrix in place of the normalized graph laplacian, but I fail to find positive results. Nonetheless, the performance of several configurations of this GCN variation is shown for the Cora, Citeseer, and Pubmed datasets.