Volume 9, Issue 1

Dependently-typed programming languages provide a powerful tool for establishing code correctness. However, it can be hard for newcomers to learn how to employ the advanced type system of such languages effectively. For simply-typed languages, several techniques have been devised to generate helpful error messages and suggestions for the programmer. We adapt these techniques to dependently-typed languages, to facilitate their more widespread adoption. In particular, we modify a higher-order unification algorithm that is used to resolve and type-check implicit arguments. We augment this algorithm with replay graphs, allowing for a global heuristic analysis of a unification problem-set, error-tolerant typing, which allows type-checking to continue after errors are found, and counter-factual unification, which makes error messages less affected by the order in which types are checked. A formalization of our algorithm is presented with an outline of its correctness. We implement replay graphs, and compare the generated error messages to those from existing languages, highlighting the improvements we achieved.

Domain-specific languages (DSLs) permeate current programming practices. An important kind of DSLs includes those developed and integrated within a host language, which we call embedded or internal DSLs. Unfortunately, embedded DSLs usually fall short on domain-specific error diagnosis, that is, they do not give control to DSL authors over how errors are reported to the programmer. As a consequence, implementation details of the DSL leak through in error messages, and programmers need to understand the internals of the DSL implementation to fix their code in a productive way. This paper addresses the challenge of building a compiler with integrated support for domain-specific error diagnosis. We assume that the type system is described using a constraint-based approach, and constraint solving is specified using rewrite rules. Domain information can then be injected at constraint gathering time via type rules, during constraint solving via specialized rules and axioms, and finally at blaming and reparation time via transformations. Furthermore, we define error contexts as a way to control the order in which solving and blaming proceeds. We engineer domain-specific error diagnosis in such a way that the compiler can also reuse the techniques for improving general error diagnosis.

Providing IDE support for a programming language or a DSL (Domain Specific Language) helps the users of the language to be more productive and to have an immediate feedback on possible errors in a program. Static types can drive IDE mechanisms such as the content assist to propose sensible completions in a given program context. Types can also be used to enrich other typical IDE parts such as the Outline and theHovering pop-ups. In this paper, we focus on statically typed imperative languages, adopting some form of type inference. We present a few general patterns for implementing efficient type systems, focusing on type error recovery. This way, the type system is able to type as many parts of the program as possible, keeping a good IDE experience. Type error messages will be placed on the important parts of the program, avoiding cascading errors that can confuse the user. We show two case studies:we apply the presented patterns to implement the type system of two statically typed DSLs, a simple expression language and a reduced Java-like language, with OOP features. We use Xtext as the language workbench for implementing the compiler and the IDE support andXsemantics, a DSL for implementing type systems using a syntax that mimics formal systems. The patterns shown in the paper can be reused also for implementing languages with other language frameworks.

Modern search is heavily powered by knowledge bases, but users still query using keywords or natural language. As search becomes increasingly dependent on the integration of text and knowledge, novel approaches for a unified representation of combined data present the opportunity to unlock new ranking strategies. We have previously proposed the graph-of-entity as a purely graph-based representation and retrieval model, however this model would scale poorly. We tackle the scalability issue by adapting the model so that it can be represented as a hypergraph. This enables a significant reduction of the number of (hyper)edges, in regard to the number of nodes, while nearly capturing the same amount of information. Moreover, such a higher-order data structure, presents the ability to capture richer types of relations, including n ary connections such as synonymy, or subsumption. We present the hypergraph-of-entity as the next step in the graph-of-entity model, where we explore a ranking approach based on biased random walks. We evaluate the approaches using a subset of the INEX 2009 Wikipedia Collection. While performance is still below the state of the art, we were, in part, able to achieve a MAP score similar to TF-IDF and greatly improve indexing efficiency over the graph-of-entity.

Addressing Answer Selection (AS) tasks with complex neural networks typically requires a large amount of annotated data to increase the accuracy of the models. In this work, we are interested in simple models that can potentially give good performance on datasets with no or few annotations. First, we propose new unsupervised baselines that leverage distributed word and sentence representations. Second, we compare the ability of our neural architectures to learn from few annotated examples in a weakly supervised scheme and we demonstrate how these methods can benefit from a pre-training on an external dataset. With an emphasis on results reproducibility, we show that our simple methods can reach or approach state-of-the-art performances on four common AS datasets.

In machine learning, hyperparameter tuning is the problem of choosing a set of optimal hyperparameters for a learning algorithm. Several approaches have been widely adopted for hyperparameter tuning, which is typically a time consuming process. We propose an efficient technique to speed up the process of hyperparameter tuning with Grid Search. We applied this technique on text categorization using kNN algorithm with BM25 similarity, where three hyperparameters need to be tuned. Our experiments show that our proposed technique is at least an order of magnitude faster than conventional tuning.

Ontologies have become a key element since many decades in information systems such as in epidemiological surveillance domain. Building domain ontologies requires the access to domain knowledge owned by domain experts or contained in knowledge sources. However, domain experts are not always available for interviews. Therefore, there is a lot of value in using ontology learning which consists in automatic or semi-automatic extraction of ontological knowledge from structured or unstructured knowledge sources such as texts, databases, etc. Many techniques have been used but they all are limited in concepts, properties and terminology extraction leaving behind axioms and rules. Source code which naturally embed domain knowledge is rarely used. In this paper, we propose an approach based on Hidden Markov Models (HMMs) for concepts, properties, axioms and rules learning from Java source code. This approach is experimented with the source code of EPICAM, an epidemiological platform developed in Java and used in Cameroon for tuberculosis surveillance. Domain experts involved in the evaluation estimated that knowledge extracted was relevant to the domain. In addition, we performed an automatic evaluation of the relevance of the terms extracted to the medical domain by aligning them with ontologies hosted on Bioportal platform through the Ontology Recommender tool. The results were interesting since the terms extracted were covered at 82.9% by many biomedical ontologies such as NCIT, SNOWMEDCT and ONTOPARON.

Query auto-completion helps users to formulate their information needs by providing suggestion lists at every typed key. This task is commonly addressed by exploiting query logs and the approaches proposed in the literature fit well in web scale scenarios, where usually huge amounts of past user queries can be analyzed to provide reliable suggestions. However, when query logs are not available, e.g. in enterprise or desktop search engines, these methods are not applicable at all. To face these challenging scenarios, we present a novel corpus-based approach which exploits the textual content of an indexed document collection in order to dynamically generate query completions. Our method extracts informative text fragments from the corpus and it combines them using a probabilistic graphical model in order to capture the relationships between the extracted concepts. Using this approach, it is possible to automatically complete partial queries with significant suggestions related to the keywords already entered by the user without requiring the analysis of the past queries. We evaluate our system through a user study on two different real-world document collections. The experiments show that our method is able to provide meaningful completions outperforming the state-of-the art approach.

Secure data distribution is critical for data accountability. Surveillance caused privacy breaching incidents have already questioned existing personal data collection techniques. Organizations assemble a huge amount of personally identifiable information (PII) for data-driven market analysis and prediction. However, the limitation of data tracking tools restricts the detection of exact data breaching points. Blockchain technology, an ‘immutable’ distributed ledger, can be leveraged to establish a transparent data auditing platform. However, Art. 42 and Art. 25 of general data protection regulation (GDPR) demands ‘right to forget’ and ‘right to erase’ of personal information, which goes against the immutability of blockchain technology. This paper proposes a GDPR complied decentralized and trusted PII sharing and tracking scheme. Proposed blockchain based personally identifiable information management system (BcPIIMS) demonstrates data movement among GDPR entities (user, controller and processor). Considering GDPR limitations, BcPIIMS used off-the-chain data storing architecture. A prototype was created to validate the proposed architecture using multichain. The use of off-the-chain storage reduces individual block size. Additionally, private blockchain also limits personal data leaking by collecting fast approval from restricted peers. This study presents personal data sharing, deleting, modifying and tracking features to verify the privacy of proposed blockchain based personally identifiable information management system.

Nowadays, the real life constraints necessitates controlling modern machines using human intervention by means of sensorial organs. The voice is one of the human senses that can control/monitor modern interfaces. In this context, Automatic Speech Recognition is principally used to convert natural voice into computer text as well as to perform an action based on the instructions given by the human. In this paper, we propose a general framework for Arabic speech recognition that uses Long Short-Term Memory (LSTM) and Neural Network (Multi-Layer Perceptron: MLP) classifier to cope with the nonuniform sequence length of the speech utterances issued fromboth feature extraction techniques, (1)Mel Frequency Cepstral Coefficients MFCC (static and dynamic features), (2) the Filter Banks (FB) coefficients. The neural architecture can recognize the isolated Arabic speech via classification technique. The proposed system involves, first, extracting pertinent features from the natural speech signal using MFCC (static and dynamic features) and FB. Next, the extracted features are padded in order to deal with the non-uniformity of the sequences length. Then, a deep architecture represented by a recurrent LSTM or GRU (Gated Recurrent Unit) architectures are used to encode the sequences of MFCC/FB features as a fixed size vector that will be introduced to a Multi-Layer Perceptron network (MLP) to perform the classification (recognition). The proposed system is assessed using two different databases, the first one concerns the spoken digit recognition where a comparison with other related works in the literature is performed, whereas the second one contains the spoken TV commands. The obtained results show the superiority of the proposed approach.

We propose a simple feature extraction algorithm for license plate location, which can reduce the occurrence of pseudo-licenses significantly. Our scheme arises from a novel L 0 -norm image smoothing, in which the multiple local textures in the complex backgrounds can be suppressed remarkably without changing the structures and edges of the license objects. Due to this “edgeaware” property, we then combine a feature filtering with an efficient binarized image, a simple multi-scale image analysis algorithm, to remove the potential false license plates. Finally, we extract license plates with a projection method. Experimental results show the proposed method provides a flexible and powerful way to the license plate location in complex backgrounds.

One of the most evident costs in cow farming is the identification of the animals. Classic identification processes are labour-intensive, prone to human errors and invasive for the animal. An automated alternative is an animal identification based on unique biometric patterns like iris recognition; in this context, correct segmentation of the region of interest becomes of critical importance. This work introduces a bovine iris segmentation pipeline that processes images taken in the wild, extracting the iris region. The solution deals with images taken with a regular visible-light camera in real scenarios, where reflections in the iris and camera flash introduce a high level of noise that makes the segmentation procedure challenging. Traditional segmentation techniques for the human iris are not applicable given the nature of the bovine eye; at this aim, a dataset composed of catalogued images and manually labelled ground truth data of Aberdeen-Angus has been used for the experiments and made publicly available. The unique ID number for each different animal in the dataset is provided, making it suitable for recognition tasks. Segmentation results have been validated with our dataset showing high reliability: with the most pessimistic metric (i.e. intersection over union), a mean score of 0.8957 has been obtained.

The use of data driven techniques is popular in smart manufacturing. Machine learning, statistics or a combination of both have been used to improve processes in electronic manufacturing. This paper presents the application of classical techniques to reduce production cycle time by compacting a production test sequence. This set of tests is run on stop-on-fail scenario for quality assurance of an electronical device. Data generated in the production test-set on stop-on-fail scenario challenges the traditional application of the data driven techniques, because of the missing data characteristic. The developed computational procedures handle this application-specific data attribute. The novelty of this work is in the algorithm developed, which applies classical techniques in an iterative environment, as a strategy to analyse incomplete datasets. Results show that the method can reduce a production test set with parametric and non-parametric tests by building an accurate prognostic model. The results can reduce production cycle time and costs. The paper details and provides discussions on the advantages and limitations of the proposed algorithms.

Version Control Systems are commonly used by Information and communication technology professionals. These systems allow monitoring programmers activity working in a project. Thus, Version Control Systems are also used by educational institutions. The aim of this work is to evaluate if the academic success of students may be predicted by monitoring their interaction with a Version Control System. In order to do so, we have built a Machine Learning model which predicts student results in a specific practical assignment of the Operating Systems Extension subject, from the second course of the degree in Computer Science of the University of León, through their interaction with a Git repository. To build the model, several classifiers and predictors have been evaluated. In order to do so, we have developed Model Evaluator (MoEv), a tool to evaluate Machine Learning models in order to get the most suitable for a specific problem. Prior to the model development, a feature selection from input data is done. The resulting model has been trained using results from 2016–2017 course and later validated using results from 2017–2018 course. Results conclude that the model predicts students’ success with a success high percentage.

Over the last few years, Multi-label classification has received significant attention from researchers to solve many issues in many fields. The manual annotation of available datasets is time-consuming and need a huge effort from the expert, especially for Multi-label applications in which each example of learning is associated with many labels at once. To overcome the manual annotation drawback, and to take advantages from the large amounts of unlabeled data, many semi-supervised approaches were proposed in the literature to give more sophisticated and fast solutions to support the automatic labeling of the unlabeled data. In this paper, a Collaborative Bagged Multi-label K-Nearest-Neighbors ( CobMLKNN ) algorithm is proposed, that extend the co-Training paradigm by a Multi-label K-Nearest-Neighbors algorithm. Experiments on ten real-world Multi-label datasets show the effectiveness of CobMLKNN algorithm to improve the performance of MLKNN to learn from a small number of labeled samples by exploiting unlabeled samples.

Word embeddings trained on natural corpora (e.g., newspaper collections, Wikipedia or the Web) excel in capturing thematic similarity (“topical relatedness”) on word pairs such as ‘coffee’ and ‘cup’ or ’bus’ and ‘road’. However, they are less successful on pairs showing taxonomic similarity, like ‘cup’ and ‘mug’ (near synonyms) or ‘bus’ and ‘train’ (types of public transport). Moreover, purely taxonomy-based embeddings (e.g. those trained on a random-walk of WordNet’s structure) outperform natural-corpus embeddings in taxonomic similarity but underperform them in thematic similarity. Previous work suggests that performance gains in both types of similarity can be achieved by enriching natural-corpus embeddings with taxonomic information from taxonomies like Word-Net. This taxonomic enrichment can be done by combining natural-corpus embeddings with taxonomic embeddings (e.g. those trained on a random-walk of WordNet’s structure). This paper conducts a deep analysis of this assumption and shows that both the size of the natural corpus and of the random-walk coverage of the WordNet structure play a crucial role in the performance of combined (enriched) vectors in both similarity tasks. Specifically, we show that embeddings trained on medium-sized natural corpora benefit the most from taxonomic enrichment whilst embeddings trained on large natural corpora only benefit from this enrichment when evaluated on taxonomic similarity tasks. The implication of this is that care has to be taken in controlling the size of the natural corpus and the size of the random-walk used to train vectors. In addition, we find that, whilst the WordNet structure is finite and it is possible to fully traverse it in a single pass, the repetition of well-connected WordNet concepts in extended random-walks effectively reinforces taxonomic relations in the learned embeddings.

The quality of Neural Machine Translation (NMT), as a data-driven approach, massively depends on quantity, quality and relevance of the training dataset. Such approaches have achieved promising results for bilingually high-resource scenarios but are inadequate for low-resource conditions. Generally, the NMT systems learn from millions of words from bilingual training dataset. However, human labeling process is very costly and time consuming. In this paper, we describe a round-trip training approach to bilingual low-resource NMT that takes advantage of monolingual datasets to address training data bottleneck, thus augmenting translation quality. We conduct detailed experiments on English-Spanish as a high-resource language pair as well as Persian-Spanish as a low-resource language pair. Experimental results show that this competitive approach outperforms the baseline systems and improves translation quality.

This Paper focuses on multi-criteria decision making techniques (MCDMs), especially analytical networking process (ANP) algorithm to design a model in order to minimize the task scheduling cost during implementation using a queuing model in a cloud environment and also deals with minimization of the waiting time of the task. The simulated results of the algorithm give better outcomes as compared to other existing algorithms by 15 percent.

Security (privacy, confidentiality and integrity) of pre-electoral, electoral and post electoral phases of the electioneering process is fundamental to the success of Electronic Voting (E-Voting) Systems. Crystography, which is the combination of cryptography and steganography could be a fitting ‘tool kit’ for enhancing the security of sensitive election-related information transmitted over public networks, thereby also ensuring free, fair and credible election/voting. Most of the existing secure e-voting systems are based on public key cryptographic schemes like RSA and Elliptic Curve Cryptography (ECC), whose security depends on the difficulty of solving Integer Factorization Problem (IFP) and Discrete Logarithm problem (DLP) respectively. However, techniques for solving IFP and DLP problems, improves continually. One of such is the quantum algorithm discovered by Peter Shor in 1994, which can solve both IFP and DLP problems in polynomial time. Consequently, the existence of quantum computers in the range of 1000 bits would spell doom to systems based on those problems. This paper presents the development of a new crystographic system that combines Post Quantum Cryptography with steganography to ensure that the security of e-voting is maintained both in classical computing era as well as post-quantum computing era. Our experiments’ results shows that our proposed system performed better than existing ones.

An annotation consists of a portion of information that is associated with a piece of content in order to explain something about the content or to add more information. The use of annotations as a tool in the educational field has positive effects on the learning process. The usual way to use this instrument is to provide students with contents, usually textual, with which they must associate annotations. In most cases this task is performed in groups of students who work collaboratively. This process encourages analysis and understanding of the contents since they have to understand them in order to annotate them, and also encourages teamwork. To facilitate its use, computer applications have been developed in recent decades that implement the annotation process and offer a set of additional functionalities. One of these functionalities is the classification of the annotations made. This functionality can be exploited in various ways in the learning process, such as guiding the students in the annotation process, providing information to the student about how the annotation process is done and to the teacher about how the students write and how they understand the content, as well as implementing other innovative educational processes. In this sense, the classification of annotations plays a critical role in the application of the annotation in the educational field. There are many studies of annotations, but most of them consider the classification aspect marginally only. This paper presents an exploratory study of the classification mechanisms used in the annotation tools, identifying four types of cases: absence of classification mechanisms, classification based on controlled vocabularies, classification based on folksonomies, and classification based on ontologies.

Measuring emotions non-intrusively via affective computing provides a promising source of information for adaptive learning and intelligent tutoring systems. Using non-intrusive, simultaneous measures of emotions, such systems could steadily adapt to students emotional states. One drawback, however, is the lack of evidence on how such modern measures of emotions relate to traditional self-reports. The aim of this study was to compare a prominent area of affective computing, facial emotion recognition, to students’ self-reports of interest, boredom, and valence. We analyzed different types of aggregation of the simultaneous facial emotion recognition estimates and compared them to self-reports after reading a text. Analyses of 103 students revealed no relationship between the aggregated facial emotion recognition estimates of the software FaceReader and self-reports. Irrespective of different types of aggregation of the facial emotion recognition estimates, neither the epistemic emotions ( i.e. , boredom and interest), nor the estimates of valence predicted the respective self-report measure. We conclude that assumptions on the subjective experience of emotions cannot necessarily be transferred to other emotional components, such as estimated by affective computing. We advise to wait for more comprehensive evidence on the predictive validity of facial emotion recognition for learning before relying on it in educational practice.