Lars Niklasson

Trajectory clustering for coastal surveillance

Achieving superior situation awareness is a key task for military, as well as civilian, decision makers. Today, automatic systems provide us with an excellent opportunity for assisting the human decision maker in achieving this awareness. Due to the potential of information overload one important aspect is to understand where to focus attention. Anomaly detection is concerned with finding deviations from normalcy and it is an increasingly important topic when providing decision support, since it can give hints towards where more analysis is needed. In this paper we explore trajectory clustering as a means for representing normal behavior in a coastal surveillance scenario. Trajectory clustering however suffers from some drawbacks in this type of setting and we therefore propose a new approach, spline-based clustering, with a potential for solving the task of representing the normal course of events.

Artificial Neural Networks in Medicine and Biology

Artificial neural networks (ANNs) are new mathematical techniques which can be used for modelling real neural networks, but also for data categorisation and inference tasks in any empirical science. This means that they have a twofold interest for the philosopher. First, ANN theory could help us to understand the nature of mental phenomena such as perceiving, thinking, remembering, inferring, knowing, wanting and acting. Second, because ANNs are such powerful instruments for data classification and inference, their use also leads us into the problems of induction and probability. Ever since David Hume expressed his famous doubts about induction, the principles of scientific inference have been a central concern for philosophers.

Time series segmentation using an adaptive resource allocating vector quantization network based on change detection

We present a novel architecture for unsupervised time series segmentation which is based on change detection rather than traditional error minimization. The architecture, which consists of a simple vector quantizer that dynamically allocates model vectors when needed, is able to split a multidimensional noisy time series generated from the sensors of a mobile robot into relevant segments using just a single presentation of the data. We compare the architecture with an existing system created by Nolfi and Tani (1999), which is based on traditional overall error minimization, and note that our system is able to detect stable and distinct signal regions which are not detected by their system.

Finding behavioural anomalies in public areas using video surveillance data

In this paper we propose an approach for detecting anomalies in data from visual surveillance sensors. The approach includes creating a structure for representing data, building ldquonormal modelsrdquo by filling the structure with data for the situation at hand, and finally detecting deviations in the data. The approach allows detections based on the incorporation of a priori knowledge about the situation and on data-driven analysis. The main advantages with the approach compared to earlier work is the low computational requirements, iterative update of normal models and a high explainability of found anomalies. The proposed approach is evaluated off-line using real-world data and the results support that the approach could be used to detect anomalies in real-time applications.

The Coaching Cycle: A Coaching-by-Gaming Approach in Serious Games

Military organizations have a long history of using simulations, role-play, and games for training. This also encompasses good practices concerning how instructors utilize games and gaming behavior. Unfortunately, the work of instructors is rarely described explicitly in research relating to serious gaming. Decision makers also tend to have overconfidence in the pedagogical power of games and simulations, particularly where the instructor is taken out of the gaming loop. The authors propose a framework, the coaching cycle, that focuses on the roles of instructors. The roles include instructors acting as game players. The fact that the instructors take a more active part in all training activities will further improve learning. The coaching cycle integrates theories of experiential learning (where action precedes theory) and deliberate practice (where the trainee’s skill is constantly challenged by a coach). Incorporating a coaching-by-gaming perspective complicates, but also strengthens, the player-centered design approach to game development in that we need to take into account two different types of players: trainees and instructor. Furthermore, the authors argue that the coaching cycle allows for a shift of focus to a more thorough debriefing, because it implies that learning of theoretical material before simulation/game playing is kept to a minimum. This shift will increase the transfer of knowledge.

Evolving decision trees using oracle guides

Some data mining problems require predictive models to be not only accurate but also comprehensible. Comprehensibility enables human inspection and understanding of the model, making it possible to trace why individual predictions are made. Since most high-accuracy techniques produce opaque models, accuracy is, in practice, regularly sacrificed for comprehensibility. One frequently studied technique, often able to reduce this accuracy vs. comprehensibility tradeoff, is rule extraction, i.e., the activity where another, transparent, model is generated from the opaque. In this paper, it is argued that techniques producing transparent models, either directly from the dataset, or from an opaque model, could benefit from using an oracle guide. In the experiments, genetic programming is used to evolve decision trees, and a neural network ensemble is used as the oracle guide. More specifically, the datasets used by the genetic programming when evolving the decision trees, consist of several different combinations of the original training data and “oracle data”, i.e., training or test data instances, together with corresponding predictions from the oracle. In total, seven different ways of combining regular training data with oracle data were evaluated, and the results, obtained on 26 UCI datasets, clearly show that the use of an oracle guide improved the performance. As a matter of fact, trees evolved using training data only had the worst test set accuracy of all setups evaluated. Furthermore, statistical tests show that two setups, both using the oracle guide, produced significantly more accurate trees, compared to the setup using training data only.

Towards template-based situation recognition

The process of tracking and identifying developing situations is an ability of importance within the surveillance domain. We refer to this as situation recognition and believe that it can enhance situation awareness for decision makers. Situation recognition requires that many subproblems are solved. For instance, we need to establish which situations are interesting, how to represent these situations, and which inferable events and states that can be used for representing them. We also need to know how to track and identify situations and how to determine the correlation between present information about situations with knowledge. For some of these subproblems, data-driven approaches are suitable, whilst knowledge-driven approaches are more suitable for others. In this paper we discuss our current research efforts and goals concerning template-based situation recognition. We provide a categorization of approaches for situation recognition together with a formalization of the template-based situation recognition problem. We also discuss this formalization in the light of a pick-pocket scenario. Finally, we discuss future directions for our research on situation recognition. We conclude that situation recognition is an important problem to look into for enhancing the overall situation awareness of decision makers.

The Importance of Diversity in Neural Network Ensembles - An Empirical Investigation

When designing ensembles, it is almost an axiom that the base classifiers must be diverse in order for the ensemble to generalize well. Unfortunately, there is no clear definition of the key term diversity, leading to several diversity measures and many, more or less ad hoc, methods for diversity creation in ensembles. In addition, no specific diversity measure has shown to have a high correlation with test set accuracy. The purpose of this paper is to empirically evaluate ten different diversity measures, using neural network ensembles and 11 publicly available data sets. The main result is that all diversity measures evaluated, in this study too, show low or very low correlation with test set accuracy. Having said that, two measures; double fault and difficulty show slightly higher correlations compared to the other measures. The study furthermore shows that the correlation between accuracy measured on training or validation data and test set accuracy also is rather low. These results challenge ensemble design techniques where diversity is explicitly maximized or where ensemble accuracy on a hold-out set is used for optimization.

Genetic programming - a tool for flexible rule extraction

Although data mining is performed to support decision making, many of the most powerful techniques, like neural networks and ensembles, produce opaque models. This lack of interpretability is an obvious disadvantage, since decision makers normally require some sort of explanation before taking action. To achieve comprehensibility, accuracy is often sacrificed by the use of simpler, transparent models, such as decision trees. Another alternative is rule extraction; i.e. to transform the opaque model into a comprehensible model, keeping acceptable accuracy. We have previously suggested a rule extraction algorithm named G-REX, which is based on genetic programming. One key property of G-REX, due to the use of genetic programming, is the possibility to use different representation languages. In this study we apply G-REX to estimation tasks. More specifically, three representation languages are evaluated using eight publicly available data sets. The quality of the extracted rules is compared to two standard techniques producing comprehensible models; multiple linear regression and the decision tree algorithm C&RT. The results show that G-REX outperforms the standard techniques, but that the choice of representation language is important.

Information fusion supporting team situation awareness for future fighting aircraft

In the military aviation domain, the decision maker, i.e. the pilot, often has to process huge amounts of information in order to make correct decisions. This is further aggravated by factors such as time-pressure, high workload and the presence of uncertain information. A support system that aids the pilot to achieve his/her goals has long been considered vital for performance progress in military aviation. Research programs within the domain have studied such support systems, though focus has not been on team collaboration. Based on identified challenges of assessing team situation awareness we suggest an approach to future military aviation support systems based on information fusion. In contrast to most previous work in this area, focus is on supporting team situation awareness, including team threat evaluation. To deal with these challenges, we propose the development of a situational adapting system, which presents information and recommendations based on the current situation.