Towards a taxonomy-based configuration framework for securing IoT infrastructures
The Internet-of-Things (IoT) is an emerging topic for individuals and companies alike. However, maintaining security and privacy of the data produced by the IoT devices can be a challenging task, especially when Cloud-based services are involved. Moreover, every IoT infrastructure has likely different requirements on the protection level of the possibly same types of data. This talk shows some approaches for building a framework for configuring and administering the local IoT infrastructure with respect to those unique protection levels.
Method to test Internet of Things systems
Internet of Things (IoT) has been applied increasingly to several and heterogeneous areas in the last years. One of the main drawbacks of the IoT systems is the amount of information they have to handle. This information arrives as events that need to be analyzed and processed in real time in order to make correct decisions. Given that processing the data is crucial, testing the IoT systems that will work with that information is required and fundamental. In order to test IoT systems, it is necessary to generate a huge number of events with specific structures and values to test the functionalities required by these systems. As this task is very hard and very prone to error if done by hand, it is proposed a method to automatically generate events for testing IoT systems. Results after applying this method to different and real experiments show its viability.
Domain-Specific Language For Generating Administrative Process Applications
Many organizations are constantly reimplementing similar business process applications. In this case, we focus on a particular class of business process: the “administrative process”, which consists of managing a formal document through several states. This document is manipulated by different participants who can edit or view parts of it depending on their role in the organization and the current state of the process. State transitions usually happen due to human decisions, deadlines or a combination of both. The process usually concludes by reaching a “final” state (e.g. “accepted” or “rejected”).
The main problem is that reimplementing these kind of applications from scratch results in wasted time that could be dedicated to better understanding the process. Besides, developers can be not familiar with some of the best practices of the technology used and even after the process is correctly implemented, the framework that the implementation is based upon may become obsolete to the point of requiring a complete rewrite.
In this work, we propose to use a high-level domain-specific language called AdminDSL to describe administrative processes. This language is accompanied by a code generator targeting Python in the web framework Django. The main idea is to serve a functional application with a base code block which keeps being the same from one to another. These applications do not need to be final and they are designed for developers to work on it, extending it to the final result. Thus, this work ensures make development and maintenance faster and easier.
Threshold-based fall-detection in form of a wearable belt
The progress of medical care increases life expectancy and this leads to an ageing population. The probability of falls rises with growing age and several diseases. To guarantee prompt support a system is needed which is able to detect different kind of falls. In this ongoing research we developed a prototype of a fall-detection system in form of a belt that is based on threshold values. The proposed hardware architecture facilitates an accurate and redundant fall-detection because of the positioning of the sensor nodes. The detection is divided in two parts which comprises the rotational movement and the impact to the ground that signalizes the fall. With the data fusion of the gyroscope and accelerometer data we intend to establish rules to detect all kinds of falls. To elaborate the threshold-based approach a Complex Event Processing (CEP) algorithm is investigated. In addition we plan to improve the hardware architecture of our prototype to guarantee mobility and an affordable detection.
Towards Generating Fuzzy Rules via Fuzzy Formal Concept Analysis
Extracting knowledge from databases is a topic which interest has increased in a wide variety of areas like stock market, medicine or census data, to name a few. Fuzzy Formal Concept Analysis plays a crucial role in the characterization in the sets of objects related to different sets of attributes. A compact representation of this is provided by the rule base, composed by fully and partially true implications between attributes. This work shows the way to obtain the base of rules given a fuzzy formal context.
Overview of mitigation techniques for addressing premature convergence problems in evolutionary algorithms
Evolutionary Algorithms, especially having complex multi-modal or multi-objective search spaces, tend to suffer from premature convergence of the solutions in local optima. As a consequence, the efficiency of evolutionary computation for solving particular problems reduces drastically and might even culminate in a complete convergence of the individuals of the population. In this talk, different techniques addressing premature convergence are discussed, from simple restarts over specific operators to complex population models. Finally, a new lightweight approach for repairing converged populations is introduced.
A Distributed Server Design and Software Development Model for the Internet of Things in an Academic Environment
For the effective and efficient collaboration of researchers in the field of the Internet of Things automatic data handling, analysis and visualization is essential. A modular approach towards a data processing system design based on the publish/subscribe communication model is presented. Basic principles of network organization and core cloud technologies are considered as a part of the overall server configuration.
Design and testing of reliable systems
In this talk I will describe the main research line of my group: conformance testing. I will introduce the concept of conformance and present the most widely used conformance relation: ioco. Finally, I will mention our main contributions concerning probabilistic, timed, and distributed extensions of the original framework.
Paving the way for air quality personalized warnings
Air quality has taken a great relevance along the last years since it can seriously affect human beings health and life quality. Particularly, air quality may improve or worsen the effects of certain illnesses and even cause death inside specific risk groups. Therefore, air quality monitoring and warning is of great importance for a significant increasing number of citizens. However, currently we do not have the means to provide us with real time information about air quality in a comfortable way; even more existent approaches do not take into account the particular conditions of every person. In this speech we show our progress on a service-oriented and event-driven architecture to detect air quality changes and notify citizens in order to prevent higher risks for their health, therefore paving the way for air quality personalized warnings.
Jaideep Chawla
Using Esper Event Processing Language to implement Real-Time Human Activity Recognition
The presentation proposes an architecture to implement Human Activity Recognition (H.A.R.) on a real time basis. H.A.R. is the classification of activities performed by the user and has always been of interest to researchers for a wide spectrum of purposes ranging from medical to programming input devices. The first step in our research was to evaluate an approach to H.A.R. using a single wrist based device consisting of inertial 9 degrees of freedom sensors and a smartphone present in a user’s leg pocket. Data was collected in the smartphone and analyzed using machine learning techniques. The Machine learning approach has been described in figure 2. Four algorithms were evaluated for their ability to accurately classify the following movements performed by the user: walking, jogging and running and other movements typically performed while working out. The focus is on using devices in a manner that causes minimum inconvenience to the user. Preliminary experiments showed accuracies in excess of 80 % for the data collected for the activities from five users which is encouraging for an approach focused on maximizing user comfort. The next step is to explore an architecture that can support the implementation as described in figures 1 & 2 on a real time basis. For the implementation we look at Esper for complex event processing and event series analysis. The transition from a static implementation of H.A.R. to a dynamic implementation is done via replacing a script which is run manually with an event processing language (EPL) framework. EPL provides event stream analysis functions which syntactically look similar to structured query language but allow for processing of data over a window size of a certain seconds which facilitates real time implementation. Event pattern matching based on timestamp is applied on data stream being generated by the sensors present on the two devices (wrist module and smartphone) in order to aggregate the data into a single stream which can then be used as an input to a data mining model. Initial analysis of the approach seems promising and analysis of its merits and demerits will be performed once it’s implemented in entireness.