Högskolan i Skövde

Smart Cities have emerged as a global concept that argues for the effective exploitation of digital technologies to drive sustainable innovation and well-being for citizens. Despite the large investments being placed on Smart City infrastructure, however, there is still very scarce attention on the new learning approaches that will be needed for cultivating Digital Smart Citizenship competences, namely the competences which will be needed by the citizens and workforce of such cities for exploiting the digital technologies in creative and innovative ways for driving financial and societal sustainability. In this context, this paper introduces cyberphysical learning as an overarching model of cultivating Digital Smart Citizenship competences by exploiting the potential of Internet of Things technologies and social media, in order to create authentic blended and augmented learning experiences.

The technology and techniques for bio-sensors are rapidly evolving. Accordingly, there is significant business interest to identify upcoming technologies and new targets for the near future. Text information from internet reflects much of the recent information and public interests that help to understand the trend of a certain field. Thus, we utilize Dirichlet process topic modeling on different media sources containing short text (e.g., blogs, news) which is able to self-adapt the learned topic space to the data. We share the observations from the domain experts on the results derived from topic modeling on wearable biosensors from multiple media sources over more than eight years. We analyze the topics on wearable devices, forecast and market analysis, and bio-sensing techniques found from our method. 

The steel-making process in a Basic Oxygen Furnace (BOF) must meet a combination of target values such as the final melt temperature and upper limits of the carbon and phosphorus content of the final melt with minimum material loss. An optimal blow end time (cut-off point), where these targets are met, often relies on the experience and skill of the operators who control the process, using both collected sensor readings and an implicit understanding of how the process develops. If the precision of hitting the optimal cut-off point can be improved, this immediately increases productivity as well as material and energy efficiency, thus decreasing environmental impact and cost. We examine the usage of standard machine learning models to predict the end-point targets using a full production dataset. Various causes of prediction uncertainty are explored and isolated using a combination of raw data and engineered features. In this study, we reach robust temperature, carbon, and phosphorus prediction hit rates of 88, 92, and 89 pct, respectively, using a large production dataset. © 2020, The Author(s).

The problem of using machine learning (ML) to predict the process endpoint for a Basic Oxygen Furnace (BOF) process used for steelmaking has been largely studied. However, current research often lacks both the usage of a rich dataset and does not address revealing influential factors that explain the process. The process is complex and difficult to control and has a multi-objective target endpoint with a proper range of heat temperature combined with sufficiently low levels of carbon and phosphorus. Reaching this endpoint requires skilled process operators, who are manually controlling the heat throughout the process by using both implicit and explicit control variables in their decisions. Trained ML models can reach good BOF target prediction results, but it is still a challenge to extract the influential factors that are significant to the ML prediction accuracy. Thus, it becomes a challenge to explain and validate an ML prediction model that claims to capture the process well. This paper makes use of a complex and full production dataset to evaluate and compare different approaches for understanding how the data can determine the process target prediction. One approach is based on the collected process data and the other on the ML approach trained on that data to find the influential factors. These complementary approaches aim to explain the BOF process to reveal actionable information on how to improve process control.

With the evolution of modern Critical Infrastructures (CI), more Cyber-Physical systems are integrated into the traditional CIs. This makes the CIs a multidimensional complex system, which is characterized by integrating cyber-physical systems into CI sectors (e.g., transportation, energy or food & agriculture). This integration creates complex interdependencies and dynamics among the system and its components. We suggest using a model with a multi-dimensional operational specification to allow detection of operational threats. Embedded (and distributed) information systems are critical parts of the CI where disruption can lead to serious consequences. Embedded information system protection is therefore crucial. As there are many different stakeholders of a CI, comprehensive protection must be viewed as a cross-sector activity to identify and monitor the critical elements, evaluate and determine the threat, and eliminate potential vulnerabilities in the CI. A systematic approach to threat modeling is necessary to support the CI threat and vulnerability assessment. We suggest a Threat Graph Model (TGM) to systematically model the complex CIs. Such modeling is expected to help the understanding of the nature of a threat and its impact on throughout the system. In order to handle threat cascading, the model must capture local vulnerabilities as well as how a threat might propagate to other components. The model can be used for improving the resilience of the CI by encouraging a design that enhances the system's ability to predict threats and mitigate their damages. This paper surveys and investigates the various threats and current approaches to threat modeling of CI. We suggest integrating both a vulnerability model and an attack model, and we incorporate the interdependencies within CI cross CI sectors. Finally, we present a multi-dimensional threat modeling approach for critical infrastructure protection.

For an operator of wireless telecommunication networks to make timely interventions in the network before minor faults escalate into issues that can lead to substandard system performance, good situation awareness is of high importance. Due to the increasing complexity of such networks, as well as the explosion of traffic load, it has become necessary to aid human operators to reach a good level of situation awareness through the use of exploratory data analysis and information fusion techniques. However, to understand the results of such techniques is often cognitively challenging and time consuming. In this paper, we present how telecommunication operators can be aided in their data analysis and sense-making processes through the usage and visualization of topic modeling results. We present how topic modeling can be used to extract knowledge from base station counter readings and make design suggestions for how to visualize the analysis results to a telecommunication operator.

Fast-moving trends, both in society and in highly competitive business areas, call for effective methods for automatic analysis. The availability of fast-moving sources in the form of short texts, such as social media and blogs, allows aggregation from a vast number of text sources, for an up to date view of trends and business insights. Topic modeling is established as an approach for analysis of large amounts of texts, but the scarcity of statistical information in short texts is considered to be a major problem for obtaining reliable topics from traditional models such as LDA. A range of different specialized topic models have been proposed, but a majority of these approaches rely on rather strong parametric assumptions, such as setting a fixed number of topics. In contrast, recent advances in the field of Bayesian non-parametrics suggest the Dirichlet process as a method that, given certain hyper-parameters, can self-adapt to the number of topics of the data at hand. We perform an empirical evaluation of the Dirichlet process multinomial (unigram) mixture model against several parametric topic models, initialized with different number of topics. The resulting models are evaluated, using both direct and indirect measures that have been found to correlate well with human topic rankings. We show that the Dirichlet Process Multinomial Mixture model is a viable option for short text topic modeling since it on average performs better, or nearly as good, compared to the parametric alternatives, while reducing parameter setting requirements and thereby eliminates the need of expensive preprocessing. 

We use a simulation approach to evaluate large scale resource usage in a distributed real-time database. Scalability is often limited by that resource usage is higher than what is added to the system when a system is scaled up. Our approach of Virtual Full Replication (VFR) makes resource usage scalable, which allows large scale real-time databases. In this paper we simulate a large scale distributed real-time database with VFR, and we compare it to a fully replicated database (FR) for a selected set of system parameters used as independent variables. Both VFR and FR support local timeliness of transactions by ensuring local availability for data objects accessed by transactions. The difference is that VFR has a scalable resource usage due to lower bandwidth usage for data update replication. The simulation shows that a simulator has several advantages for studying large scale distributed real-time databases and for studying scalability in resource usage in such systems.

A fully replicated distributed real-time database provides high availability and predictable access times, independent of user location, since all the data is available at each node. However, full replication requires that all updates are replicated to every node, resulting in exponential growth of bandwidth and processing demands with the number of nodes and objects added. To eliminate this scalability problem, while retaining the advantages of full replication, this thesis explores Virtual Full Replication (ViFuR); a technique that gives database users a perception of using a fully replicated database while only replicating a subset of the data.

We use ViFuR in a distributed main memory real-time database where timely transaction execution is required. ViFuR enables scalability by replicating only data used at the local nodes. Also, ViFuR enables flexibility by adaptively replicating the currently used data, effectively providing logical availability of all data objects. Hence, ViFuR substantially reduces the problem of non-scalable resource usage of full replication, while allowing timely execution and access to arbitrary data objects.

In the thesis we pursue ViFuR by exploring the use of database segmentation. We give a scheme (ViFuR-S) for static segmentation of the database prior to execution, where access patterns are known a priori. We also give an adaptive scheme (ViFuR-A) that changes segmentation during execution to meet the evolving needs of database users. Further, we apply an extended approach of adaptive segmentation (ViFuR-ASN) in a wireless sensor network - a typical dynamic large-scale and resource-constrained environment. We use up to several hundreds of nodes and thousands of objects per node, and apply a typical periodic transaction workload with operation modes where the used data set changes dynamically. We show that when replacing full replication with ViFuR, resource usage scales linearly with the required number of concurrent replicas, rather than exponentially with the system size.

Distributed real-time systems increase in size an complexity, and the nodes in such systems become difficult to implement and test. In particular, communication for synchronization of shared information in groups of nodes becomes complex to manage. Several authors have proposed to using a distributed database as a communication subsystem, to off-load database applications from explicit communication. This lets the task for information dissemination be done by the replication mechanisms of the database. With increasingly larger systems, however, there is a need for managing the scalability for such database approach. Furthermore, timeliness for database clients requires predictable resource usage, and scalability requires bounded resource usage in the database system. Thus, predictable resource management is an essential function for realizing timeliness in a large scale setting.

We discuss scalability problems and methods for distributed real-time databases in the context of the DeeDS database prototype. Here, all transactions can be executed timely at the local node due to main memory residence, full replication and detached replication of updates. Full replication contributes to timeliness and availability, but has a high cost in excessive usage of bandwidth, storage, and processing, in sending all updates to all nodes regardless of updates will be used there or not. In particular, unbounded resource usage is an obstacle for building large scale distributed databases. For many application scenarios it can be assumed that most of the database is shared by only a limited number of nodes. Under this assumption it is reasonable to believe that the degree of replication can be bounded, so that a bound also can be set on resource usage.

The thesis proposal identifies and elaborates research problems for bounding resource usage in large scale distributed real-time databases. One objective is to bound resource usage by taking advantages of pre-specified data needs, but also by detecting unspecified data needs and adapting resource management accordingly. We elaborate and evaluate the concept of virtual full replication, which provides an image of a fully replicated database to database clients. It makes data objects available where needed, while fulfilling timeliness and consistency requirements on the data.

In the first part of our work, virtual full replication makes data available where needed by taking advantages of pre-specified data accesses to the distributed database. For hard real-time systems, the required data accesses are usually known since such systems need to be well specified to guarantee timeliness. However, there are many applications where a specification of data accesses can not be done before execution. The second part of our work extends virtual full replication to be used with such applications. By detecting new and changed data accesses during execution and adapt database replication, virtual full replication can continuously provide the image of full replication while preserving scalability.

One of the objective of the thesis work is to quantify scalability in the database context, so that actual benefits and achievements can be evaluated. Further, we find out the conditions for setting bounds on resource usage for scalability, under both static and dynamic data requirements.

Bandwidth reservation for real-time networks offers an approach for real-time networking in a switched Ethernet or IP setting. In a switched network, switches prevent from indeterministic back-off times at collisions and here bandwidth reservation limits send rates to protects from overallocation of link bandwidth. The work in this paper aims at avoiding problems with a centralized bandwidth broker for resource reservation for bandwidth in a throttled real-time network. A centralized broker (such as a ’GlobeThrottle’) is a single point of failure in a distributed system and is also a hot-spot resource, which all nodes of the system use for registering new real-time channels. To avoid both these problems we propose a distributed algorithm to be used instead of a central bandwidth broker. Also, the GlobeThrottle approach uses TCP/IP communication for channel allocation, which gives indeterministic time channel allocation. However, this problem is not addressed in this paper.

In the proposed solution, Real-time Communication Through a Distributed Resource Reservation Approach (STRUTS), real-time channels are throttled and the throttle level for each sending node is agreed between all nodes. The agreement must be atomic to avoid transitional bandwidth over-usage due to temporary inconsistencies between nodes (’mutual inconsistencies’) in throttling level information for different nodes. In this paper we use two-phase-commit (2PC) for atomic node agreements, where changes in channels information will be visible at the same time instant at all nodes. Thus, the throttling based on the agreed channel information will guarantee that the maximal bandwidth is not exceeded. Using a distributed agreement avoids the problems of a single point of failure and hot-spot behavior, and is thus scalable to some extent. However, 2PC commitment incurs other problems with scalability since it requires that the network is not partitioned and also that nodes are locked during the agreement process, which prevents other nodes from allocating channels for the same node concurrently. When using 2PC for agreement it is not possible to have deterministic channel allocation time when there are no guarantees for the maximum locking time on the channel data. We have through experimental results verified that by using STRUTS, we avoid overallocation of links.

We implement Virtual full replication for a distributed real-time database by segmenting the database on multiple data properties. Virtual full replication provides an image to the application of full replication in a partially replicated database, by replicating data to meet the actual data needs of the users of the data. This is useful since fully replicated real-time databases, that allow updates at all nodes, do not scale well as updates must be replicated to every other node for replica consistency, also to nodes where only a small share of the database will ever be used. We propose an algorithm that segments the database on multiple data properties without causing a combinatorial problem. We show, by analysis and an implementation, that scalability for such a system can be improved due to scalable resource usage, while application semantics of full replication is unchanged.

 It is important for the life time of a wireless sensor network (WSN) to reduce the amount of data transferred through the network. As a typical approach, sensor data is filtered before propagating updates, to a node at the edge of a network, where it can be fused. Information Fusion inside the network can reduce the amount of data propagated, by fusing data before and in propagation, without losing the information value in it. We explore infrastructures for distributed fusion, with fusion nodes located at strategic nodes inside the network, as an approach of structured distributed fusion for WSNs. We propose an infrastructure for a white-board approach that uses a distributed real-time database with virtual full replication. With such an approach, both raw and fused data are logically available at all nodes and physically available where used, such that only used data will be propagated and use resources. The actual resource usage will be relative to the actual demand for data, rather than to the amount of data published at the white-board. We present an exploration of such an infrastructure, and points out future key research questions for such a white-board approach.

We propose virtual full replication by adaptive segmentation (ViFuR-A), and evaluate its ability to maintain scalability in a replicated real-time database. With full replication and eventual consistency, transaction timeliness becomes independent of network delays for all transactions. However, full replication does not scale well, since all updates must be replicated to all nodes, also when data is needed only at a subset of the nodes. With virtual full replication that adapts to actual data needs, resource usage can be bounded and the database can be made scalable. We propose a scheme for adaptive segmentation that detects new data needs and adapts replication. The scheme includes an architecture, a scalable protocol and a replicated directory service that together maintains scalability. We show that adaptive segmentation bounds the required storage at a significantly lower level compared to static segmentation, for a typical workload where the data needs change repeatedly. Adaptation time can be kept constant for the workload when there are sufficient resources. Also, the storage is constant with an increasing amount of nodes and linear with an increasing rate of change to data needs.

Sensor networks have limited resources and often support large-scale applications that need scalable propagation of sensor data to users. We propose a white-board style of communication in sensor networks using a distributed real-time database supporting Virtual Full Replication with Adaptive Segmentation. This allows mobile client nodes to access, transparently and efficiently, any sensor data at any node in the network. We present a two-tiered wireless architecture, and an adaptation protocol, for scalable and adaptive white-board communication in large-scale sensor networks. Sensor value readings at nodes of the sensor tier are published at nodes of the database tier as database updates to objects in a distributed real-time database. The search space of client nodes for sensor data is thus limited to the number of database nodes. With this scheme, we can show scalable resource usage and short adaptation times for several hundreds of database nodes and up to 50 moving clients.

We propose to use a distributed real-time database with Virtual Full Replication by Adaptive Segmentation, for whiteboard communication in a sensor network with mobile sink nodes. Sensor networks are large scale applications with limited resources, so they need scalable propagation of sensor data, both to the users inside the network and to the network edges. Virtual full replication enables scalable and adaptive propagation of sensor data, by bounding resource usage to the current data needs. We use a two-tiered wireless sensor network, where each sensor value is published in the distributed database at gateways of the upper tier. Mobile users search for sensor data only at the gateways, which limits the search space.

To ensure reliable network performance, anomaly detection is an important part of the telecommunication operators’ work. This includes that operators need to timely intervene with the network, should they encounter indications of network performance degradation. In this paper, we describe the results of an initial experiment for anomaly detection with regard to network performance, using topic modeling on base station run-time variable data collected from live Radio Access Networks (RANs). The results show that topic modeling clusters semantically related data in the same way as human experts would and that the anomalies in our test cases could be identified in latent Dirichlet allocation (LDA) topic models. Our experiment further reveals which information provided by the topic model is particularly usable to support human anomaly detection in this application domain.

To ensure that a wireless telecommunication system is reliably functioning at all times, root-causes of potential network failures need to be identified and remedied, ideally before a noticeable network performance degradation occurs. Network operators are today observing a multitude of key performance indicators (KPIs) and are notified of possible network problems through alarms issued by different parts of the network. However, the number of cascading alarms together with the number of observable KPIs are easily overwhelming the operator’s cognitive capacity. In this paper we show how exploratory data analysis and machine learning, in particular topic modelling, can assist the operator when monitoring network performance and identifying anomalous network behaviour as well as supporting the operator’s analysis of the anomaly and identification of its root-cause. 

Artificial intelligence (AI) is nowadays included into an increasing number of critical systems. Inclusion of AI in such systems may, however, pose a risk, since it is, still, infeasible to build AI systems that know how to function well in situations that differ greatly from what the AI has seen before. Therefore, it is crucial that future AI systems have the ability to not only function well in known domains, but also understand and show when they are uncertain when facing something unknown. In this paper, we evaluate four different methods that have been proposed to correctly quantifying uncertainty when the AI model is faced with new samples. We investigate the behaviour of these models when they are applied to samples far from what these models have seen before, and if they correctly attribute those samples with high uncertainty. We also examine if incorrectly classified samples are attributed with an higher uncertainty than correctly classified samples. The major finding from this simple experiment is, surprisingly, that the evaluated methods capture the uncertainty differently and the correlation between the quantified uncertainty of the models is low. This inconsistency is something that needs to be further understood and solved before AI can be used in critical applications in a trustworthy and safe manner. © 2020, Springer Nature Switzerland AG.

We present a flexible deep convolutional neural network method for the analysis of arbitrary sized graph structures representing molecules. This method, which makes use of the Lipinski RDKit module, an open-source cheminformatics software, enables the incorporation of any global molecular (such as molecular charge and molecular weight) and local (such as atom hybridization and bond orders) information. In this paper, we show that this method significantly outperforms another recently proposed method based on deep convolutional neural networks on several datasets that are studied. Several best practices for training deep convolutional neural networks on chemical datasets are also highlighted within the article, such as how to select the information to be included in the model, how to prevent overfitting and how unbalanced classes in the data can be handled.

In medicinal chemistry programs it is key to design and make compounds that are efficacious and safe. This is a long, complex, and difficult multiparameter optimization process, often including several properties with orthogonal trends. New methods for the automated design of compounds against profiles of multiple properties are thus of great value. Here we present a fragment-based reinforcement learning approach based on an actor-critic model, for the generation of novel molecules with optimal properties. The actor and the critic are both modeled with bidirectional long short-term memory (LSTM) networks. The AI method learns how to generate new compounds with desired properties by starting from an initial set of lead molecules and then improving these by replacing some of their fragments. A balanced binary tree based on the similarity of fragments is used in the generative process to bias the output toward structurally similar molecules. The method is demonstrated by a case study showing that 93% of the generated molecules are chemically valid and more than a third satisfy the targeted objectives, while there were none in the initial set.

We present a flexible deep convolutional neural network method for the analyse of arbitrary sized graph structures representing molecules. The method makes use of RDKit, an open-source cheminformatics software, allowing the incorporation of any global molecular (such as molecular charge) and local (such as atom type) information. We evaluate the method on the Side Effect Resource (SIDER) v4.1 dataset and show that it significantly outperforms another recently proposed method based on deep convolutional neural networks. We also reflect on how different types of information and input data affect the predictive power of our model. This reflection highlights several open problems that should be solved to further improve the use of deep learning within cheminformatics.

We present a new self-organizing algorithm for classification of a data that combines and extends the strengths of several common machine learning algorithms, such as algorithms in self-organizing neural networks, ensemble methods and deep neural networks. The increased expression power is combined with the explanation power of self-organizing networks. Our algorithm outperforms both deep neural networks and ensembles of deep neural networks. For our evaluation case, we use production monitoring data from a complex steel manufacturing process, where data is both high-dimensional and has many nonlinear interdependencies. In addition to the improved prediction score, the algorithm offers a new deep-learning based approach for how computational resources can be focused in data exploration, since the algorithm points out areas of the input space that are more challenging to learn.

Basic oxygen steel making is a complex chemical and physical industrial process that reduces a mix of pig iron and recycled scrap into low-carbon steel. Good understanding of the process and the ability to predict how it will evolve requires long operator experience, but this can be augmented with process target prediction systems. Such systems may use machine learning to learn a model of the process based on a long process history, and have an advantage in that they can make useof vastly more process parameters than operators can comprehend. While it has become less of a challenge to build such prediction systems using machine learning algorithms, actual production implementations are rare. The hidden reasoning of complex prediction model and lack of transparency prevents operator trust, even for models that show high accuracy predictions. To express model behaviour and thereby increasing transparency we develop a reinforcement learning (RL) based agent approach, which task is to generate short polynomials that can explain the model of the process from what it has learnt from process data. The RL agent is rewarded on how well it generates polynomials that can predict the process from a smaller subset of the process parameters. Agent training is done with the REINFORCE algorithm, which enables the sampling of multiple concurrently plausible polynomials. Having multiple polynomials, process developers can evaluate several alternative and plausible explanations, as observed in the historic process data. The presented approach gives both a trained generative model and a set of polynomials that can explain the process. The performance of the polynomials is as good as or better than more complex and less interpretable models. Further, the relative simplicity of the resulting polynomials allows good generalisation to fit new instances of data. The best of the resulting polynomials in our evaluation achieves a better R 2 score on the test set in comparison to the other machine learning models evaluated.

A Basic Oxygen Furnace (BOF) for steel making is a complex industrial process that is difficult to monitor due to the harsh environment, so the collected production data is very limited given the process complexity. Also, such production data has a low degree of variability. An accurate machine learning (ML) model for predicting production outcome requires both large and varied data, so utilising data from multiple BOFs will allow for more capable ML models, since both the amount and variability of data increases. Data collection setups for different BOFs are different, such that data sets are not compatible to directly join for ML training. Our approach is to let a neural network benefit from these collection differences in a joint training model. We present a neural network-based approach that simultaneously and jointly co-trains on several data sets. Our novelty is that the first network layer finds an internal representation of each individual BOF, while the other layers use this representation to concurrently learn a common BOF model. Our evaluation shows that the prediction accuracy of the common model increases compared to separate models trained on individual furnaces’ data sets. It is clear that multiple data sets can be utilised this way to increase model accuracy for better production prediction performance. For the industry, this means that the amount of available data for model training increases and thereby more capable ML models can be trained when having access to multiple data sets describing the same or similar manufacturing processes. 

The competitiveness in the manufacturing industry raises demands for using recent data analysis algorithms for manufacturing process development. Data-driven analysis enables extraction of novel knowledge from already existing sensors and data, which is necessary for advanced manufacturing process refinement involving aged machinery. Improved data analysis enables factories to stay competitive against newer factories, but without any hefty investment. In large manufacturing operations, the dependencies between data are highly complex and therefore very difficult to analyse manually. This paper applies a deep learning approach, using a recurrent neural network with long short term memory cells together with an attention mechanism to model the dependencies between the measured product shape, as measured before the most critical manufacturing operation, and the final product quality. Our approach predicts the ratio of flawed products already before the critical operation with an AUC-ROC score of 0.85, i.e., we can detect more than 80 % of all flawed products while having less than 25 % false positive predictions (false alarms). In contrast to previous deep learning approaches, our method shows how the recurrent neural network reasons about the input shape, using the attention mechanism to point out which parts of the product shape that have the highest influence on the predictions. Such information is crucial for both process developers, in order to understand and improve the process, and for process operators who can use the information to learn how to better trust the predictions and control the process.