Abstract Technological development and various information services becoming common has had the effect that data from everyday situations is available. Utilizing this technology and the data it produces in an efficient manner is called context-aware or ubiquitous computing. The research includes the specifications of each application, the requirements of the communication systems, issues of privacy, and human - computer interaction, for example. The environment should learn from the user's behaviour and communicate with the user. The communication should not be only reactive, but proactive as well. This thesis is divided into two parts, both representing methodology for enabling intelligence in our everyday surroundings. In part one, three different applications are defined for studying context-recognition and routine learning: a health monitoring system, a context-aware health club application, and automatic device configuration in an office space. The path for routine learning is straight forward and it is closely related to pattern recognition research. Sensory data is collected from users in various different situations, the signals are pre-processed, and the contexts recognized from this sensory data. Furthermore, routine learning is realized through association rules. The routine learning paradigm developed here can utilize already recognized contexts despite their meaning in the real world. The user makes the final decision on whether the routine is important or not, and has authority over every action of the system. The second part of the thesis is built on experiments on identifying a person walking on a pressure-sensitive floor. Resolving the characteristics of the special sensor producing the measurements, which lies under the normal flooring, is one of the tasks of this research. The identification is tested with Hidden Markov models and Learning Vector Quantization. The methodology developed in this thesis offers a step along the long road towards functional and calm intelligent environments.

The premise for AI systems like personal assistants to provide guidance and suggestions to an end-user is to understand, at any moment in time, the personal context that the user is in. The context – where the user is, what she is doing and with whom – allows the machine to represent the world in user’s terms. The context must be inferred from a stream of sensor readings generated by smart wearables such as smartphones and smartwatches, and the labels are acquired from the user directly. To perform robust context prediction in this real-world scenario, the machine must handle the egocentric nature of the context, adapt to the changing world and user, and maintain a bidirectional interaction with the user to ensure the user-machine alignment of world representations. To this end, the machine must learn incrementally on the input stream of sensor readings and user supervision. In this work, we: (i) introduce interactive classification in the wild and present knowledge drift (KD), a special form of concept drift, occurring due to world and user changes; (ii) develop simple and robust ML methods to tackle these scenarios; (iii) showcase the advantages of each of these methods in empirical evaluations on controlled synthetic and real-world data sets; (iv) design a flexible and modular architecture that combines the methods above to support context recognition in the wild; (v) present an evaluation with real users in a concrete social science use case.

Several applications have been proposed for wireless sensor networks, including habitat monitoring, structural health monitoring, pipeline monitoring, and precision agriculture. Among the desirable features of wireless sensor networks, one is the ease of deployment. Since the nodes are capable of self-organization, they can be placed easily in areas that are otherwise inaccessible to or impractical for other types of sensing systems. In fact, some have proposed the deployment of wireless sensor networks by dropping nodes from a plane, delivering them in an artillery shell, or launching them via a catapult from onboard a ship. There are also reports of actual aerial deployments, for example the one carried out using an unmanned aerial vehicle (UAV) at a Marine Corps combat centre in California -- the nodes were able to establish a time-synchronized, multi-hop communication network for tracking vehicles that passed along a dirt road. While this has a practical relevance for some civil applications (such as rescue operations), a more realistic deployment involves the careful planning and placement of sensors. Even then, nodes may not be placed optimally to ensure that the network is fully connected and high-quality data pertaining to the phenomena being monitored can be extracted from the network. This work aims to address the problem of random deployment through two complementary approaches: The first approach aims to address the problem of random deployment from a communication perspective. It begins by establishing a comprehensive mathematical model to quantify the energy cost of various concerns of a fully operational wireless sensor network. Based on the analytic model, an energy-efficient topology control protocol is developed. The protocol sets eligibility metric to establish and maintain a multi-hop communication path and to ensure that all nodes exhaust their energy in a uniform manner. The second approach focuses on addressing the problem of imperfect sensing from a signal processing perspective. It investigates the impact of deployment errors (calibration, placement, and orientation errors) on the quality of the sensed data and attempts to identify robust and error-agnostic features. If random placement is unavoidable and dense deployment cannot be supported, robust and error-agnostic features enable one to recognize interesting events from erroneous or imperfect data.

Several applications have been proposed for wireless sensor networks, including habitat monitoring, structural health monitoring, pipeline monitoring, and precision agriculture. Among the desirable features of wireless sensor networks, one is the ease of deployment. Since the nodes are capable of self-organization, they can be placed easily in areas that are otherwise inaccessible to or impractical for other types of sensing systems. In fact, some have proposed the deployment of wireless sensor networks by dropping nodes from a plane, delivering them in an artillery shell, or launching them via a catapult from onboard a ship. There are also reports of actual aerial deployments, for example the one carried out using an unmanned aerial vehicle (UAV) at a Marine Corps combat centre in California -- the nodes were able to establish a time-synchronized, multi-hop communication network for tracking vehicles that passed along a dirt road. While this has a practical relevance for some civil applications (such as rescue operations), a more realistic deployment involves the careful planning and placement of sensors. Even then, nodes may not be placed optimally to ensure that the network is fully connected and high-quality data pertaining to the phenomena being monitored can be extracted from the network. This work aims to address the problem of random deployment through two complementary approaches: The first approach aims to address the problem of random deployment from a communication perspective. It begins by establishing a comprehensive mathematical model to quantify the energy cost of various concerns of a fully operational wireless sensor network. Based on the analytic model, an energy-efficient topology control protocol is developed. The protocol sets eligibility metric to establish and maintain a multi-hop communication path and to ensure that all nodes exhaust their energy in a uniform manner. The second approach focuses on addressing the problem of imperfect sensing from a signal processing perspective. It investigates the impact of deployment errors (calibration, placement, and orientation errors) on the quality of the sensed data and attempts to identify robust and error-agnostic features. If random placement is unavoidable and dense deployment cannot be supported, robust and error-agnostic features enable one to recognize interesting events from erroneous or imperfect data.