Context Maritime Surveillance (MS) has received increased attention from a civilian perspective in recent years. Anomaly detection (AD) is one of the many techniques available for improving the safety and security in the MS domain. Maritime authorities utilize various confidential data sources for monitoring the maritime activities; however, a paradigm shift on the Internet has created new sources of data for MS. These newly identified data sources, which provide publicly accessible data, are the open data sources. Taking advantage of the open data sources in addition to the traditional sources of data in the AD process will increase the accuracy of the MS systems. Objectives The goal is to investigate the potential open data as a complementary resource for AD in the MS domain. To achieve this goal, the first step is to identify the applicable open data sources for AD. Then, a framework for AD based on the integration of open and closed data sources is proposed. Finally, according to the proposed framework, an AD system with the ability of using open data sources is developed and the accuracy of the system and the validity of its results are evaluated. Methods In order to measure the system accuracy, an experiment is performed by means of a two stage random sampling on the vessel traffic data and the number of true/false positive and negative alarms in the system is verified. To evaluate the validity of the system results, the system is used for a period of time by the subject matter experts from the Swedish Coastguard. The experts check the detected anomalies against the available data at the Coastguard in order to obtain the number of true and false alarms. Results The experimental outcomes indicate that the accuracy of the system is 99%. In addition, the Coastguard validation results show that among the evaluated anomalies, 64.47% are true alarms, 26.32% are false and 9.21% belong to the vessels that remain unchecked due to the lack of corresponding data in the Coastguard data sources. Conclusions This thesis concludes that using open data as a complementary resource for detecting anomalous behavior in the MS domain is not only feasible but also will improve the efficiency of the surveillance systems by increasing the accuracy and covering some unseen aspects of maritime activities.

The amount of software that hosts spyware has increased dramatically. To avoid legal repercussions, the vendors need to inform users about inclusion of spyware via end user license agreements (EULAs) during the installation of an application. However, this information is intentionally written in a way that is hard for users to comprehend. We investigate how to automatically discriminate between legitimate software and spyware associated software by mining EULAs. For this purpose, we compile a data set consisting of 996 EULAs out of which 9.6% are associated to spyware. We compare the performance of 17 learning algorithms with that of a baseline algorithm on two data sets based on a bag-of-words and a meta data model. The majority of learning algorithms significantly outperform the baseline regardless of which data representation is used. However, a non-parametric test indicates that bag-of-words is more suitable than the meta model. Our conclusion is that automatic EULA classification can be applied to assist users in making informed decisions about whether to install an application without having read the EULA. We therefore outline the design of a spyware prevention tool and suggest how to select suitable learning algorithms for the tool by using a multi-criteria evaluation approach.

Abstract-Machine learning algorithms are frequently applied in data mining applications. Many of the tasks in this domain concern high-dimensional data. Consequently, these tasks are often complex and computationally expensive. This paper presents a GPU-based parallel implementation of the Random Forests algorithm. In contrast to previous work, the proposed algorithm is based on the compute unified device architecture (CUDA). An experimental comparison between the CUDA-based algorithm (CudaRF), and state-of-the-art Random Forests algorithms (FastRF and LibRF) shows that CudaRF outperforms both FastRF and LibRF for the studied classification task.

No abstract

In the context of document image analysis, image binarization is an important preprocessing step for other document analysis algorithms, but also relevant on its own by improving the readability of images of historical documents. While historical document image binarization is challenging due to common image degradations, such as bleedthrough, faded ink or stains, achieving good binarization performance in a timely manner is a worthwhile goal to facilitate efficient information extraction from historical documents. In this paper, we propose a recurrent neural network based algorithm using Grid Long Short-Term Memory cells for image binarization, as well as a pseudo F-Measure based weighted loss function. We evaluate the binarization and execution performance of our algorithm for different choices of footprint size, scale factor and loss function. Our experiments show a significant trade-off between binarization time and quality for different footprint sizes. However, we see no statistically significant difference when using different scale factors and only limited differences for different loss functions. Lastly, we compare the binarization performance of our approach with the best performing algorithm in the 2016 handwritten document image binarization contest and show that both algorithms perform equally well.

No abstract

Large-scale data centers account for a significant share of the energy consumption in many countries. Machine learning technology requires intensive workloads and thus drives requirements for lots of power and cooling capacity in data centers. It is time to explore green machine learning. The aim of this paper is to profile a machine learning algorithm with respect to its energy consumption and to determine the causes behind this consumption. The first scalable machine learning algorithm able to handle large volumes of streaming data is the Very Fast Decision Tree (VFDT), which outputs competitive results in comparison to algorithms that analyze data from static datasets. Our objectives are to: (i) establish a methodology that profiles the energy consumption of decision trees at the function level, (ii) apply this methodology in an experiment to obtain the energy consumption of the VFDT, (iii) conduct a fine-grained analysis of the functions that consume most of the energy, providing an understanding of that consumption, (iv) analyze how different parameter settings can significantly reduce the energy consumption. The results show that by addressing the most energy intensive part of the VFDT, the energy consumption can be reduced up to a 74.3%.

No abstract