SANTM: Efficient Self-attention-driven Network for Text Matching

Self-attention mechanisms have recently been embraced for a broad range of text-matching applications. Self-attention model takes only one sentence as an input with no extra information, i.e., one can utilize the final hidden state or pooling. However, text-matching problems can be interpreted either in symmetrical or asymmetrical scopes. For instance, paraphrase detection is an asymmetrical task, while textual entailment classification and question-answer matching are considered asymmetrical tasks. In this article, we leverage attractive properties of self-attention mechanism and proposes an attention-based network that incorporates three key components for inter-sequence attention: global pointwise features, preceding attentive features, and contextual features while updating the rest of the components. Our model follows evaluation on two benchmark datasets cover tasks of textual entailment and question-answer matching. The proposed efficient Self-attention-driven Network for Text Matching outperforms the state of the art on the Stanford Natural Language Inference and WikiQA datasets with much fewer parameters.

A Framework for Reversible Data Embedding into Base45 and Other Non-Base64 Encoded Strings

Printable string encodings are widely used in several applications that cannot deal with binary data, the most known example being the mail system. In this paper, we investigate the potential of some of the proposed encodings to hide and carry extra information. We describe a framework for reversibly embedding data in printable string encodings, like Base45. The method leverages the characteristic of some encodings that are not surjective by using illegal configurations to embed one bit of information. With the assumption of uniformly distributed binary input data, an estimation of the expected payload can be computed easily. Results are reported for Base45 and Base85 encodings.

Quantifying Non-Stationarity with Information Theory

We introduce an index based on information theory to quantify the stationarity of a stochastic process. The index compares on the one hand the information contained in the increment at the time scale τ of the process at time t with, on the other hand, the extra information in the variable at time t that is not present at time t−τ. By varying the scale τ, the index can explore a full range of scales. We thus obtain a multi-scale quantity that is not restricted to the first two moments of the density distribution, nor to the covariance, but that probes the complete dependences in the process. This index indeed provides a measure of the regularity of the process at a given scale. Not only is this index able to indicate whether a realization of the process is stationary, but its evolution across scales also indicates how rough and non-stationary it is. We show how the index behaves for various synthetic processes proposed to model fluid turbulence, as well as on experimental fluid turbulence measurements.

Imputation on the Food, Nutrition and Environment Surveys 2007 and 2009 data

<p>The Food Nutrition Environment Survey (FNES) is a survey of New Zealand early childhood centres and schools and the food and nutritional services that they provide for their pupils. The 2007 and 2009 FNES surveys were managed by the Ministry of Health. Like all the other social surveys, the FNES has the common problem of unit and item non-responses. In other words, the FNES has missing data. In this thesis, we have surveyed a wide variety of missing data handling techniques and applied most of them to the FNES datasets. This thesis can be roughly divided into two parts. In the first part, we have studied and investigated the different nature of missing data (i.e. missing data mechanisms), and all the common and popular imputation methods, using the Synthetic Unit Record File (SURF) which has been developed by the Statistics New Zealand for educational purposes. By comparing all those different imputation methods, Bayesian Multiple Imputation (MI) method is the preferred option to impute missing data in terms of reducing non-response bias and properly propagating imputation uncertainty. Due to the overlaps in the samples selected for the 2007 and 2009 FNES surveys, we have discovered that the Bayesian MI can be improved by incorporating the matched dataset. Hence, we have proposed a couple of new approaches to utilize the extra information from the matched dataset. We believe that adapting the Bayesian MI to use the extra information from the matched dataset is a preferable imputation strategy for imputing the FNES missing data. This is because the use of the matched dataset provides more prediction power to the imputation model.</p>

Imputation on the Food, Nutrition and Environment Surveys 2007 and 2009 data

<p>The Food Nutrition Environment Survey (FNES) is a survey of New Zealand early childhood centres and schools and the food and nutritional services that they provide for their pupils. The 2007 and 2009 FNES surveys were managed by the Ministry of Health. Like all the other social surveys, the FNES has the common problem of unit and item non-responses. In other words, the FNES has missing data. In this thesis, we have surveyed a wide variety of missing data handling techniques and applied most of them to the FNES datasets. This thesis can be roughly divided into two parts. In the first part, we have studied and investigated the different nature of missing data (i.e. missing data mechanisms), and all the common and popular imputation methods, using the Synthetic Unit Record File (SURF) which has been developed by the Statistics New Zealand for educational purposes. By comparing all those different imputation methods, Bayesian Multiple Imputation (MI) method is the preferred option to impute missing data in terms of reducing non-response bias and properly propagating imputation uncertainty. Due to the overlaps in the samples selected for the 2007 and 2009 FNES surveys, we have discovered that the Bayesian MI can be improved by incorporating the matched dataset. Hence, we have proposed a couple of new approaches to utilize the extra information from the matched dataset. We believe that adapting the Bayesian MI to use the extra information from the matched dataset is a preferable imputation strategy for imputing the FNES missing data. This is because the use of the matched dataset provides more prediction power to the imputation model.</p>

A Solution of the Junction Riemann Problem for 1D Hyperbolic Balance Laws in Networks including Supersonic Flow Conditions on Elastic Collapsible Tubes

The numerical modeling of one-dimensional (1D) domains joined by symmetric or asymmetric bifurcations or arbitrary junctions is still a challenge in the context of hyperbolic balance laws with application to flow in pipes, open channels or blood vessels, among others. The formulation of the Junction Riemann Problem (JRP) under subsonic conditions in 1D flow is clearly defined and solved by current methods, but they fail when sonic or supersonic conditions appear. Formulations coupling the 1D model for the vessels or pipes with other container-like formulations for junctions have been presented, requiring extra information such as assumed bulk mechanical properties and geometrical properties or the extension to more dimensions. To the best of our knowledge, in this work, the JRP is solved for the first time allowing solutions for all types of transitions and for any number of vessels, without requiring the definition of any extra information. The resulting JRP solver is theoretically well-founded, robust and simple, and returns the evolving state for the conserved variables in all vessels, allowing the use of any numerical method in the resolution of the inner cells used for the space-discretization of the vessels. The methodology of the proposed solver is presented in detail. The JRP solver is directly applicable if energy losses at the junctions are defined. Straightforward extension to other 1D hyperbolic flows can be performed.

Improving Welfare in One-Sided Matchings using Simple Threshold Queries

We study one-sided matching problems where each agent must be assigned at most one object. In this classic problem it is often assumed that agents specify only ordinal preferences over objects and the goal is to return a matching that satisfies some desirable property such as Pareto optimality or rank-maximality. However, agents may have cardinal utilities describing their preference intensities and ignoring this can result in welfare loss. We investigate how to elicit additional cardinal information from agents using simple threshold queries and use it in turn to design algorithms that return a matching satisfying a desirable matching property, while also achieving a good approximation to the optimal welfare among all matchings satisfying that property. Overall, our results show how one can improve welfare by even non-adaptively asking agents for just one bit of extra information per object.

Robust Visual Relationship Detection towards Sparse Images in Internet-of-Things

Visual relationship can capture essential information for images, like the interactions between pairs of objects. Such relationships have become one prominent component of knowledge within sparse image data collected by multimedia sensing devices. Both the latent information and potential privacy can be included in the relationships. However, due to the high combinatorial complexity in modeling all potential relation triplets, previous studies on visual relationship detection have used the mixed visual and semantic features separately for each object, which is incapable for sparse data in IoT systems. Therefore, this paper proposes a new deep learning model for visual relationship detection, which is a novel attempt for cooperating computational intelligence (CI) methods with IoTs. The model imports the knowledge graph and adopts features for both entities and connections among them as extra information. It maps the visual features extracted from images into the knowledge-based embedding vector space, so as to benefit from information in the background knowledge domain and alleviate the impacts of data sparsity. This is the first time that visual features are projected and combined with prior knowledge for visual relationship detection. Moreover, the complexity of the network is reduced by avoiding the learning of redundant features from images. Finally, we show the superiority of our model by evaluating on two datasets.

A Review of Remote Sensing Image Dehazing

Remote sensing (RS) is one of the data collection technologies that help explore more earth surface information. However, RS data captured by satellite are susceptible to particles suspended during the imaging process, especially for data with visible light band. To make up for such deficiency, numerous dehazing work and efforts have been made recently, whose strategy is to directly restore single hazy data without the need for using any extra information. In this paper, we first classify the current available algorithm into three categories, i.e., image enhancement, physical dehazing, and data-driven. The advantages and disadvantages of each type of algorithm are then summarized in detail. Finally, the evaluation indicators used to rank the recovery performance and the application scenario of the RS data haze removal technique are discussed, respectively. In addition, some common deficiencies of current available methods and future research focus are elaborated.

Revisiting the Food- and Nutrition-Related Curriculum in Healthcare Education: An Example for Pharmacy Education

Objective: This study aimed to obtain an objective overview of nutritional topics discussed in community pharmacies to adapt the nutrition-related course content in pharmacy education. Methods: We performed an observational study between July 2014 and April 2015 in 136 community pharmacies in Belgium. During four months, each pharmacy intern recorded the first two food- and nutrition-related cases with which they were confronted. Each case was classified into one of 18 categories. Results: 1004 cases were included by 135 pharmacy interns. The most often discussed subjects include “food supplements” (38%), “baby food” (19%), and “healthy food and nutritional recommendations” (11%). In 45% (447/1004) of all cases, pharmacy interns were able to immediately discuss the cases without searching for additional information. Eventually, after looking up extra information, 95% (958/1004) of cases could be answered. Conclusions: Food- and nutrition-related cases are discussed in primary healthcare. We recommend food- and nutrition-related courses in the curriculum of every healthcare profession.