Documents and Dependencies: an Exploration of Vector Space Models for Semantic Composition

In most previous research on distributional semantics, Vector Space Models (VSMs) of words are built either from topical information (e.g., documents in which a word is present), or from syntactic/semantic types of words (e.g., dependency parse links of a word in sentences), but not both. In this paper, we explore the utility of combining these two representations to build VSM for the task of semantic composition of adjective-noun phrases. Through extensive experiments on benchmark datasets, we find that even though a type-based VSM is effective for semantic composition, it is often outperformed by a VSM built using a combination of topic- and type-based statistics. We also introduce a new evaluation task wherein we predict the composed vector representation of a phrase from the brain activity of a human subject reading that phrase. We exploit a large syntactically parsed corpus of 16 billion tokens to build our VSMs, with vectors for both phrases and words, and make them publicly available.

An optimisation of Gaussian mixture models for integer processing units

This paper investigates sub-integer implementations of the adaptive Gaussian mixture model (GMM) for background/foreground segmentation to allow the deployment of the method on low cost/low power processors that lack Floating Point Unit (FPU). We propose two novel integer computer arithmetic techniques to update Gaussian parameters. Specifically, the mean value and the variance of each Gaussian are updated by a redefined and generalised "round'' operation that emulates the original updating rules for a large set of learning rates. Weights are represented by counters that are updated following stochastic rules to allow a wider range of learning rates and the weight trend is approximated by a line or a staircase. We demonstrate that the memory footprint and computational cost of GMM are significantly reduced, without significantly affecting the performance of background/foreground segmentation.

When does information about causal structure improve statistical reasoning?

Base rate neglect on the mammography problem can be overcome by explicitly presenting a causal basis for the typically vague false-positive statistic. One account of this causal facilitation effect is that people make probabilistic judgements over intuitive causal models parameterized with the evidence in the problem. Poorly defined or difficult-to-map evidence interferes with this process, leading to errors in statistical reasoning. To assess whether the construction of parameterized causal representations is an intuitive or deliberative process, in Experiment 1 we combined a secondary load paradigm with manipulations of the presence or absence of an alternative cause in typical statistical reasoning problems. We found limited effects of a secondary load, no evidence that information about an alternative cause improves statistical reasoning, but some evidence that it reduces base rate neglect errors. In Experiments 2 and 3 where we did not impose a load, we observed causal facilitation effects. The amount of Bayesian responding in the causal conditions was impervious to the presence of a load (Experiment 1) and to the precise statistical information that was presented (Experiment 3). However, we found less Bayesian responding in the causal condition than previously reported. We conclude with a discussion of the implications of our findings and the suggestion that there may be population effects in the accuracy of statistical reasoning.

Integrating geophysical data and upscaling techniques to support regional groundwater flow modelling: A practical example of the Lagan Valley, Northern Ireland

When studying heterogeneous aquifer systems, especially at regional scale, a degree of generalization is anticipated. This can be due to sparse sampling regimes, complex depositional environments or lack of accessibility to measure the subsurface. This can lead to an inaccurate conceptualization which can be detrimental when applied to groundwater flow models. It is important that numerical models are based on observed and accurate geological information and do not rely on the distribution of artificial aquifer properties. This can still be problematic as data will be modelled at a different scale to which it was collected. It is proposed here that integrating geophysics and upscaling techniques can assist in a more realistic and deterministic groundwater flow model. In this study, the sedimentary aquifer of the Lagan Valley in Northern Ireland is chosen due to intruding sub-vertical dolerite dykes. These dykes are of a lower permeability than the sandstone aquifer. The use of airborne magnetics allows the delineation of heterogeneities, confirmed by field analysis. Permeability measured at the field scale is then upscaled to different levels using a correlation with the geophysical data, creating equivalent parameters that can be directly imported into numerical groundwater flow models. These parameters include directional equivalent permeabilities and anisotropy. Several stages of upscaling are modelled in finite element. Initial modelling is providing promising results, especially at the intermediate scale, suggesting an accurate distribution of aquifer properties. This deterministic based methodology is being expanded to include stochastic methods of obtaining heterogeneity location based on airborne geophysical data. This is through the Direct Sample method of Multiple-Point Statistics (MPS). This method uses the magnetics as a training image to computationally determine a probabilistic occurrence of heterogeneity. There is also a need to apply the method to alternate geological contexts where the heterogeneity is of a higher permeability than the host rock.

Power-Capped DVFS and Thread Allocation with ANN Models on Modern NUMA Systems

Abstract—Power capping is an essential function for efficient power budgeting and cost management on modern server systems. Contemporary server processors operate under power caps by using dynamic voltage and frequency scaling (DVFS). However, these processors are often deployed in non-uniform memory
access (NUMA) architectures, where thread allocation between cores may significantly affect performance and power consumption. This paper proposes a method which maximizes performance under power caps on NUMA systems by dynamically optimizing two knobs: DVFS and thread allocation. The method selects the optimal combination of the two knobs with models based on artificial neural network (ANN) that captures the nonlinear effect of thread allocation on performance. We implement
the proposed method as a runtime system and evaluate it with twelve multithreaded benchmarks on a real AMD Opteron based NUMA system. The evaluation results show that our method outperforms a naive technique optimizing only DVFS by up to
67.1%, under a power cap.

Weaving behavior into feature models for embedded system families

Product Line software Engineering depends on capturing the commonality and variability within a family of products, typically using feature modeling, and using this information to evolve a generic reference architecture for the family. For embedded systems, possible variability in hardware and operating system platforms is an added complication. The design process can be facilitated by first exploring the behavior associated with features. In this paper we outline a bidirectional feature modeling scheme that supports the capture of commonality and variability in the platform environment as well as within the required software. Additionally, 'behavior' associated with features can be included in the overall model. This is achieved by integrating the UCM path notation in a way that exploits UCM's static and dynamic stubs to capture behavioral variability and link it to the feature model structure. The resulting model is a richer source of information to support the architecture development process.

Coupling ground and airborne geophysical data with upscaling techniques for regional groundwater modeling of heterogeneous aquifers: Case study of a sedimentary aquifer intruded by volcanic dykes in Northern Ireland

In highly heterogeneous aquifer systems, conceptualization of regional groundwater flow models frequently results in the generalization or negligence of aquifer heterogeneities, both of which may result in erroneous model outputs. The calculation of equivalence related to hydrogeological parameters and applied to upscaling provides a means of accounting for measurement scale information but at regional scale. In this study, the Permo-Triassic Lagan Valley strategic aquifer in Northern Ireland is observed to be heterogeneous, if not discontinuous, due to subvertical trending low-permeability Tertiary dolerite dykes. Interpretation of ground and aerial magnetic surveys produces a deterministic solution to dyke locations. By measuring relative permeabilities of both the dykes and the sedimentary host rock, equivalent directional permeabilities, that determine anisotropy calculated as a function of dyke density, are obtained. This provides parameters for larger scale equivalent blocks, which can be directly imported to numerical groundwater flow models. Different conceptual models with different degrees of upscaling are numerically tested and results compared to regional flow observations. Simulation results show that the upscaled permeabilities from geophysical data allow one to properly account for the observed spatial variations of groundwater flow, without requiring artificial distribution of aquifer properties. It is also found that an intermediate degree of upscaling, between accounting for mapped field-scale dykes and accounting for one regional anisotropy value (maximum upscaling) provides results the closest to the observations at the regional scale.

Entanglement properties of spin models in triangular lattices

The different quantum phases appearing in strongly correlated systems as well as their transitions are closely related to the entanglement shared between their constituents. In 1D systems, it is well established that the entanglement spectrum is linked to the symmetries that protect the different quantum phases. This relation extends even further at the phase transitions where a direct link associates the entanglement spectrum to the conformal field theory describing the former. For 2D systems much less is known. The lattice geometry becomes a crucial aspect to consider when studying entanglement and phase transitions. Here, we analyze the entanglement properties of triangular spin lattice models by also considering concepts borrowed from quantum information theory such as geometric entanglement.

Building Information Modelling Energy Performance Assessment on Domestic Dwellings: A Comparative Study

Building Information Modelling (BIM) is growing in pace, not only in design and construction stages, but also in the analysis of facilities throughout their life cycle. With this continued growth and utilisation of BIM processes, comes the possibility to adopt such procedures, to accurately measure the energy efficiency of buildings, to accurately estimate their energy usage. To this end, the aim of this research is to investigate if the introduction of BIM Energy Performance Assessment in the form of software analysis, provides accurate results, when compared with actual energy consumption recorded. Through selective sampling, three domestic case studies are scrutinised, with baseline figures taken from existing energy providers, the results scrutinised and compared with calculations provided from two separate BIM energy analysis software packages. Of the numerous software packages available, criterion sampling is used to select two of the most prominent platforms available on the market today. The two packages selected for scrutiny are Integrated Environmental Solutions - Virtual Environment (IES-VE) and Green Building Studio (GBS). The results indicate that IES-VE estimated the energy use in region of ±8% in two out of three case studies while GBS estimated usage approximately ±5%. The findings indicate that the introduction of BIM energy performance assessment, using proprietary software analysis, is a viable alternative to manual calculations of building energy use, mainly due to the accuracy and speed of assessing, even the most complex models. Given the surge in accurate and detailed BIM models and the importance placed on the continued monitoring and control of buildings energy use within today’s environmentally conscious society, this provides an alternative means by which to accurately assess a buildings energy usage, in a quick and cost effective manner.

An Intelligent Threat Prevention Framework with Heterogeneous Information

Three issues usually are associated with threat prevention intelligent surveillance systems. First, the fusion and interpretation of large scale incomplete heterogeneous information; second, the demand of effectively predicting suspects’ intention and ranking the potential threats posed by each suspect; third, strategies of allocating limited security resources (e.g., the dispatch of security team) to prevent a suspect’s further actions towards critical assets. However, in the literature, these three issues are seldomly considered together in a sensor network based intelligent surveillance framework. To address
this problem, in this paper, we propose a multi-level decision support framework for in-time reaction in intelligent surveillance. More specifically, based on a multi-criteria event modeling framework, we design a method to predict the most plausible intention of a suspect. Following this, a decision support model is proposed to rank each suspect based on their threat severity and to determine resource allocation strategies. Finally, formal properties are discussed to justify our framework.

Global Sensitivity Analysis for MAP Inference in Graphical Models

We study the sensitivity of a MAP configuration of a discrete probabilistic graphical model with respect to perturbations of its parameters. These perturbations are global, in the sense that simultaneous perturbations of all the parameters (or any chosen subset of them) are allowed. Our main contribution is an exact algorithm that can check whether the MAP configuration is robust with respect to given perturbations. Its complexity is essentially the same as that of obtaining the MAP configuration itself, so it can be promptly used with minimal effort. We use our algorithm to identify the largest global perturbation that does not induce a change in the MAP configuration, and we successfully apply this robustness measure in two practical scenarios: the prediction of facial action units with posed images and the classification of multiple real public data sets. A strong correlation between the proposed robustness measure and accuracy is verified in both scenarios.

Towards a general framework for information fusion

Depending on the representation setting, different combination rules have been proposed for fusing information from distinct sources. Moreover in each setting, different sets of axioms that combination rules should satisfy have been advocated, thus justifying the existence of alternative rules (usually motivated by situations where the behavior of other rules was found unsatisfactory). These sets of axioms are usually purely considered in their own settings, without in-depth analysis of common properties essential for all the settings. This paper introduces core properties that, once properly instantiated, are meaningful in different representation settings ranging from logic to imprecise probabilities. The following representation settings are especially considered: classical set representation, possibility theory, and evidence theory, the latter encompassing the two other ones as special cases. This unified discussion of combination rules across different settings is expected to provide a fresh look on some old but basic issues in information fusion.

Predicting forest attributes In southeast Alaska using artificial neural networks

Artificial neural network (ANN) methods are used to predict forest characteristics. The data source is the Southeast Alaska (SEAK) Grid Inventory, a ground survey compiled by the USDA Forest Service at several thousand sites. The main objective of this article is to predict characteristics at unsurveyed locations between grid sites. A secondary objective is to evaluate the relative performance of different ANNs. Data from the grid sites are used to train six ANNs: multilayer perceptron, fuzzy ARTMAP, probabilistic, generalized regression, radial basis function, and learning vector quantization. A classification and regression tree method is used for comparison. Topographic variables are used to construct models: latitude and longitude coordinates, elevation, slope, and aspect. The models classify three forest characteristics: crown closure, species land cover, and tree size/structure. Models are constructed using n-fold cross-validation. Predictive accuracy is calculated using a method that accounts for the influence of misclassification as well as measuring correct classifications. The probabilistic and generalized regression networks are found to be the most accurate. The predictions of the ANN models are compared with a classification of the Tongass national forest in southeast Alaska based on the interpretation of satellite imagery and are found to be of similar accuracy.

Game-­theoretic Resource Allocation with Real-­time Probabilistic Surveillance Information

Game-theoretic security resource allocation problems have generated significant interest in the area of designing and developing security systems. These approaches traditionally utilize the Stackelberg game model for security resource scheduling in order to improve the protection of critical assets. The basic assumption in Stackelberg games is that a defender will act first, then an attacker will choose their best response after observing the defender’s strategy commitment (e.g., protecting a specific asset). Thus, it requires an attacker’s full or partial observation of a defender’s strategy. This assumption is unrealistic in real-time threat recognition and prevention. In this paper, we propose a new solution concept (i.e., a method to predict how a game will be played) for deriving the defender’s optimal strategy based on the principle of acceptable costs of minimax regret. Moreover, we demonstrate the advantages of this solution concept by analyzing its properties.