Search

Remaining useful life (RUL) of an asset or system is defined as the length from the current time and operating state to the end of the useful life. It is of paramount importance for safety-critical industries such as aviation and lies in the heart of prognostics and health management (PHM). This paper investigates the usage of automated machine learning (AutoML) for RUL estimation, based on using classical machine learning algorithms for regression. The data is pre-processed by extracting statistical features from expanding windows of the signal in order to uncover the degradation that has been accumulating from the early life of the system or after an overhaul. We evaluate our methodology on the widely-used C-MAPSS dataset and compare our approach to the state-of-the-art deep neural networks (DNNs) and classical machine learning algorithms. The experimental results show that AutoML outperforms or is comparable to traditional machine learning techniques and standard neural networks, while being outperformed by specifically designed neural networks on datasets with multiple fault mode and operating conditions. These results show that with the correct pre-processing automated machine learning is able to accurately estimate the RUL, which implies that such approaches can be industrially deployed.

Geocast is the concept of sending data packets to nodes in a specified geographical area instead of nodes with a specific address. This forwarding method is valuable in situations where any number of nodes inside a geographic area need to be reached, such as vehicular networking scenarios. To facilitate large scale geocast, a wired network geographic routing algorithm is needed that can route packets efficiently towards a destination area. Our goal is to design an algorithm that can deliver shortest path tree like geographic forwarding while relying purely on distributed data without central knowledge. In this paper we present and implement two algorithms for geographic routing. One algorithm is based purely on distance-vector data. Another, more complicated algorithm is based on path data. We show that our purely distance-vector-based algorithm can come close to the number of links used by a shortest path tree when a small number of routers are present in the destination area. We also show that our path-based algorithm can come close to the link usage of a shortest path tree in almost all geocast situations. We also show that the algorithms converge relatively quickly following link drops.

Remote sensing (RS) of biophysical variables plays a vital role in providing the information necessary for understanding spatio-temporal dynamics in ecosystems. The hybrid approach to retrieve biophysical variables from RS by combining Machine Learning (ML) algorithms with surrogate data generated by Radiative Transfer Models (RTM). The susceptibility of the ill-posed solutions to noise currently constrains further application of hybrid approaches. Here, we explored how noise affects the performance of ML algorithms for biophysical trait retrieval. We focused on synthetic Sentinel-2 (S2) data generated using the PROSAIL RTM and four commonly applied ML algorithms: Gaussian Processes (GPR), Random Forests (RFR), and Artificial Neural Networks (ANN) and Multi-task Neural Networks (MTN). After identifying which biophysical variables can be retrieved from S2 using a Global Sensitivity Analysis, we evaluated the performance loss of each algorithm using the Mean Absolute Percentage Error (MAPE) with increasing noise levels. We found that, for S2 data, Carotenoid concentrations are uniquely dependent on band 2, Chlorophyll is almost exclusively dependent on the visible ranges, and Leaf Area Index, water, and dry matter contents are mostly dependent on infrared bands. Without added noise, GPR was the best algorithm (20%) even at low levels of added noise (≈5%). Overall, both neural networks performed significantly better than GPR and RFR when noise was added with the MTN being slightly better when compared to the ANN. Our results imply that the performance of the commonly used algorithms in hybrid-RTM inversion are pervasively sensitive to noise. The implication is that more advanced models or approaches are necessary to minimize the impact of noise to improve near real-time and accurate RS monitoring of biophysical trait retrieval.

Early time series classification (EarlyTSC) involves the prediction of a class label based on partial observation of a given time series. Most EarlyTSC algorithms consider the trade-off between accuracy and earliness as two competing objectives, using a single dedicated hyperparameter. To obtain insights into this trade-off requires finding a set of non-dominated (Pareto efficient) classifiers. So far, this has been approached through manual hyperparameter tuning. Since the trade-off hyperparameters only provide indirect control over the earliness-accuracy trade-off, manual tuning is tedious and tends to result in many sub-optimal hyperparameter settings. This complicates the search for optimal hyperparameter settings and forms a hurdle for the application of EarlyTSC to real-world problems. To address these issues, we propose an automated approach to hyperparameter tuning and algorithm selection for EarlyTSC, building on developments in the fast-moving research area known as automated machine learning (AutoML). To deal with the challenging task of optimising two conflicting objectives in early time series classification, we propose MultiETSC, a system for multi-objective algorithm selection and hyperparameter optimisation (MO-CASH) for EarlyTSC. MultiETSC can potentially leverage any existing or future EarlyTSC algorithm and produces a set of Pareto optimal algorithm configurations from which a user can choose a posteriori. As an additional benefit, our proposed framework can incorporate and leverage time-series classification algorithms not originally designed for EarlyTSC for improving performance on EarlyTSC; we demonstrate this property using a newly defined, “naïve” fixed-time algorithm. In an extensive empirical evaluation of our new approach on a benchmark of 115 data sets, we show that MultiETSC performs substantially better than baseline methods, ranking highest (avg. rank 1.98) compared to conceptually simpler single-algorithm (2.98) and single-objective alternatives (4.36).

This book contains a selection of the best papers of the 32nd Benelux Conference on Artificial Intelligence, BNAIC/Benelearn 2020, held in Leiden, The Netherlands, in November 2020. Due to the COVID-19 pandemic the conference was held online. The 12 papers presented in this volume were carefully reviewed and selected from 41 regular submissions. They address various aspects of artificial intelligence such as natural language processing, agent technology, game theory, problem solving, machine learning, human-agent interaction, AI and education, and data analysis.

How do we make sure that all citizens in a city have access to enough green space? An increasing part of the world’s population lives in urban areas, where contact with nature is largely reduced to street trees and parks. As optional tree planting sites and financial resources are limited, determining the best planting site can be formulated as an optimization problem with constraints. Can we locate these sites based on the popularity of nearby venues? How can we ensure that we include groups of people who tend to spend time in tree deprived areas?

From detecting skin cancer to translating languages and to forecasting electricity consumption, machine learning is enabling advanced capabilities of computer systems across a broad range of important real-world applications. In this work, we present machine learning models for forecasting the consumption of electricity. Short-term electric load forecasting has been a fundamental concern in power operation systems for over a century. Energy load forecasting is of even greater importance, due to applications in the planning of demand side management, smart electric vehicles and other smart grid technologies. We use two state-of-the-art automated machine learning systems (auto-sklearn and TPOT), which automate model selection and hyperparameter optimization, to achieve maximum prediction accuracy, and compare their performance for the task of load prediction using two benchmark problems. These benchmarks are derived from real world load consumption tasks, namely household consumption from the UCI data repository and consumption data from an industrial office building. Our experimental results indicate great potential for improving the accuracy of energy consumption prediction by using automated machine learning approaches.

Recently, Point of interest (POI) recommendation has gained ever-increasing importance in various Location-Based Social Networks (LBSNs). With the recent advances of neural models, much work has sought to leverage neural networks to learn neural embeddings in a pre-training phase that achieve an improved representation of POIs and consequently a better recommendation. However, previous studies fail to capture crucial information about POIs such as categorical information. In this paper, we propose a novel neural model that generates a POI embedding incorporating sequential and categorical information from POIs. Our model consists of a check-in module and a category module. The check-in module captures the geographical influence of POIs derived from the sequence of users' check-ins, while the category module captures the characteristics of POIs derived from the category information. To validate the efficacy of the model, we experimented with two large-scale LBSN datasets. Our experimental results demonstrate that our approach significantly outperforms state-of-the-art POI recommendation methods., Comment: 4 pages, 1 figures

There are multiple methods for tracking individuals, but the classical ones such as using GPS or video surveillance systems do not scale or have large costs. The need for large-scale tracking, for thousands or even millions of individuals, over large areas such as cities, requires the use of alternative techniques. WiFi tracking is a scalable solution that has gained attention recently. This method permits unobtrusive tracking of large crowds, at a reduced cost. However, extracting knowledge from the data gathered through WiFi tracking is not simple, due to the low positional accuracy and the dependence on signals generated by the tracked device, which are irregular and sparse. To facilitate further data analysis, we can partition individual trajectories into periods of stops and moves. This abstraction level is fundamental, and it opens the way for answering complex questions about visited locations or even social behavior. Determining stops and movements has been previously addressed for tracking data gathered using GPS. GPS trajectories have higher positional accuracy at a fixed, higher frequency as compared to trajectories obtained through WiFi. However, even with the increase in accuracy, the problem, of separating traces in periods of stops and movements, remains similar to the one we encountered for WiFi tracking. In this paper, we study three algorithms for determining stops and movements for GPS-based datasets and explore their applicability to WiFi-based data. We propose possible improvements to the best-performing algorithm considering the specifics of WiFi tracking data.

Intelligent Transportation Systems in the domain of vehicular networking, have recently been subject to rapid development. In vehicular ad hoc networks, data broadcast is one of the main communication types and its reliability is crucial for high performance applications. However, due to the lack of acknowledgment techniques in IEEE 802.11p standard, it is challenging to ensure communication reliability. In this work, we analytically model a receiver-oriented reliability mechanism, with the objective of on-demand error recovery for multi-hop broadcast vehicular communication. In particular, using absorbing Markov modeling and probabilistic graphical modeling, we analyze its performance in terms of relevant indicators, such as overhead and delivery ratio. Further, the model is validated using simulations. The results are useful in tuning the influential parameters and accordingly adjusting trade-offs and meet performance requirements in various circumstances.

A system using frequency offset based transmit- reference (TR) modulation allows multiple nodes to transmit simultaneously and asynchronously without any mutual timing coordination. Thus, such a system provides inherent capabilities for a multiple access in the medium access control (MAC) layer to coordinate the shared use of the common wireless medium among the nodes of the wireless sensor network (WSN). However, certain characteristics of a frequency offset based system limits its performance, for example, the number of available frequency offsets is limited as it depends on several system parameters, and the number of simultaneous communications using different frequency offsets is limited due to inter-user interference. In this paper, we introduce an extended version of the performance model of a basic slotted- Aloha system, that captures the basic phenomena of a multi- channel system with a limited set of channels and a limit to the number of simultaneously used channels. An analysis of this model reveals the potential of a MAC protocol for TR modulation with frequency offsets.

The Medium Access Control (MAC) layer can influence the energy consumption of a wireless sensor network (WSN) to a significant level. TR-MAC is an energy-efficient preamble sampling based MAC protocol for low power WSNs suitable for low data rate and low duty cycle scenario. However, low data rate is not always maintained in wireless sensor networks which often have to deal with event-driven scenarios where a sudden event rapidly increases traffic load within the network. In this paper we propose a traffic-adaptive duty cycle adaptation mechanism in order to provide responsiveness to traffic rate variations for TR-MAC protocol. This mechanism increases throughput and decreases packet delay while maintaining energy-efficiency without any extra information exchange among the sensor nodes in the network.

Rapid advancements in Micro-Electro-Mechanical Systems (MEMS), and wireless communications, have resulted in a surge in data generation. Mobility data is one of the various forms of data, which are ubiquitously collected by different location sensing devices. Extensive knowledge about the behavior of humans and wildlife is buried in raw mobility data. This knowledge can be used for realizing numerous viable applications ranging from wildlife movement analysis, to various location-based recommendation systems, urban planning, and disaster relief. With respect to what mentioned above, in this thesis, we mainly focus on providing data analytics for understanding the behavior and interaction of mobile entities (humans and animals). To this end, the main research question to be addressed is: How can behaviors and interactions of mobile entities be determined from mobility data acquired by (mobile) wireless sensor nodes in an accurate and efficient manner? To answer the above-mentioned question, both application requirements and technological constraints are considered in this thesis. On the one hand, applications requirements call for accurate data analytics to uncover hidden information about individual behavior and social interaction of mobile entities, and to deal with the uncertainties in mobility data. Technological constraints, on the other hand, require these data analytics to be efficient in terms of their energy consumption and to have low memory footprint, and processing complexity.

Ubiquity of portable location-aware devices and popularity of online location-based services, have recently given rise to the collection of datasets with high spatial and temporal resolution. The subject of analyzing such data has consequently gained popularity due to numerous opportunities enabled by understanding objects' (people and animals, among others) mobility patterns. In this paper, we propose a hidden semi-Markov-based model to understand the behavior of mobile entities. The hierarchical state structure in our model allows capturing spatio-temporal associations in the locational history both at stay-points and on the paths connecting them. We compare the accuracy of our model with a number of other spatio-temporal models using two real datasets. Furthermore, we perform sensitivity analysis on our model to evaluate its robustness in presence of common issues in mobility datasets such as existence of noise and missing values. Results of our experiments show superiority of the proposed scheme compared with the other models.

With ever-increasing advancements of body area networks and participatory sensing, various healthcare applications have emerged. These applications are used to collect physiological and activity data from people and to evaluate their health situation in an online manner. The ultimate goal here is to ensure health and well-being of participants. Although various researches have been performed on underlying system of these applications from the technical point of view, very few have focused on ways to encourage people to use them and to increase their acceptability. In this paper, we present a taxonomy for motivations and incentives used in body area network based healthcare systems. Furthermore, we identify incentives' features and output expectancies and use them in a model to determine important factors that persuade people to accept and use a health monitoring system. Finally, we evaluate each incentive using Analytical Hierarchy Process (AHP) technique to identify the most effective elements of each incentive having a specific goal in mind.

Wireless sensor networks are composed of sensors with low computational and energy resources. Transmission of data is one of the most energy consuming operations in such networks. In-network data aggregation is a popular technique which is performed to reduce data transmissions. However, by aggregating multiple data into one, the security of them is no longer guaranteed. While concealed data aggregation methods have been recently proposed to provide energy efficient, end-to-end confidentiality, not much work has been done to enhance them with aggregate integrity. In this work, we propose a scalable and energy efficient bihomomorphic method to preserve end-to-end confidentiality and aggregate integrity against outsider attacks. A distributed validation scheme is used to prevent blind rejection caused by outsiders. To provide better resilience, a key refreshment method is used to prevent analysis attacks.