chapter01_introduction.tex.backup

\chapter{Introduction}
\label{ch:intro}

\section{Background and Motivation}
\label{sec:motivation}
We are currently witnessing an era of technological convergence that rivals some of the great technological upheavals of modern history\footnote{By ``technological convergence'', we mean that a set of technologies has undergone rapid advances simultaneously and thus have become available for technological uptake in combinatorial ways.}. The steam engine, the electric lamp, the transistor, the jetliner, the artificial satellite---it is in this same revered company that we can place the technological revolution we are now undergoing. According to authors Erik Brynjolfsson and Andrew McAfee, we are living in a ``second machine age'' (where the first machine age began with James Watt's steam engine), which they describe as ``an inflection point in the history of our economies and societies because of digitization.'' (Brynjolfsson and McAfee, 2014; p. 11). They define digitization as ``converting things into bits that can be stored on a computer and sent over a network'' (\emph{ibid.}, p. 10). The resulting digital information has remarkably different properties from the industrial products of the first machine age, a topic which Brynjolfsson and McAfee explore in detail in their book. They define ``digital technologies'' as ``those that have computer hardware, software, and networks at their core'' (\emph{ibid.}, p. 9). It is within this wider context of digital technologies and the second machine age that this thesis is best understood.

Digital technologies is a broad category; therefore, it is useful to narrow the focus to a few key technologies that are driving the development of the second machine age. This thesis focuses on four such technologies and looks at how they have impacted a particular application area, namely navigation. The chosen technologies include: (1) mobile telecommunication devices, (2) the Internet, (3) positioning technologies, and (4) a wide range of inexpensive yet highly capable sensors, namely microelectromechanical systems (MEMS). We note that these four technologies have converged over the course of a few decades, so that the changes are clearly evident within one human generation (i.e. 20-30 years). All of these technologies came to a technological crossroads in the late 20th century and early 21st century, so that a child born and raised in the 21st century will have vastly different technological possibilities, compared to one born and raised in the 20th century.

The first major manifestation of this technological convergence, especially with respect to consumer markets, is the so-called ``smartphone'', which incorporates or supports all four of the above-mentioned technologies. Looking at the history of mobile devices, it is difficult to say which mobile phone can be considered the first smartphone. In terms of marketing, the Ericsson R380, released in 2000, was the first mobile phone to be called a smartphone. In terms of the four technologies listed above, the Samsung SCH-S310, introduced in 2005, was probably the first to exhibit all four. The first iPhone was released in 2007, and the first Android phone was released in 2008.

About 64 million smartphones were sold globally in 2006 (Canalys, 2007), and by 2008 this number exceeded 139 million (Gartner, 2009). By 2012, there were already more than one billion smartphones in use worldwide (Strategy Analytics, 2013). This number is forecast to reach nearly 2.5 billion in 2015 (Korea Times, 2014). These devices allow their users to stay ``connected'' virtually everywhere they go, and consequently anyone can connect to these billion plus users from any networked device, including desktop computers and ``land-line'' phones---no matter where the user is located or travelling to. Ironically, in many technologically advanced societies, it is now considered a societal and/or behavioral challenge for one to go ``off the grid'' or ``disconnected'' for any extended period of time.

It is our view that the smartphone is only the first manifestation of this technological revolution. Many other so-called ``smart'' devices are soon to follow: ``smartwatches'' and the use  of various wearable sensors may soon become a mainstay consumer habit. In addition, the same technologies that have made smartphones possible and popular are quickly making their way into existing everyday devices, including cars, home appliances, and even toothbrushes. Furthermore, it is not just consumer markets that are being transformed but also many industrial markets, ranging from manufacturing to commercial shipping. It would be  na\"{\i}ve to speculate exactly how this revolution will play out in the coming decades, but is clear is that it is already changing the lifestyles, habits, and possibilities of people living in the early 21st century, especially those who can afford these (currently) ``high-end'' consumer devices.

Aside from being a convergence of new digital technologies, is there any unifying concept or principle that is underlying this technological revolution? Some would argue that it is the increased levels of \emph{mobility} that these technologies provide. Others have rallied under the banner of \emph{ubiquitous computing} or \emph{pervasive computing}, which describes the fact that computing devices can now be found nearly everywhere one looks. Certainly these are two important characteristics giving wind to this revolution, but we argue in this thesis towards another underlying principle that provides a common thread and deep insight into how our relationship to these computing devices is changing.

One common development, of course, is the increasing ability of computing devices to fulfil various user desires, e.g. to download large amounts of data at high speeds, to capture or render various high-quality multimedia content, to store and edit content in various ways, etc. What is not advancing or expanding---at least, not at any considerable rate---is the patience or attention span of the users themselves. Therefore, users are expecting (consciously or not) that their devices will ``do more'' with essentially the same total quantity and quality of human input. Fortunately, however, these devices are rapidly advancing in their ability to know what their users want or need without the user having to explicitly formulate and express these desires to the computer. This is the goal under which this thesis is motivated and focused---to improve our understanding of how computing devices can better understand us and our needs.

The primary method by which this thesis aims to achieve this goal is through \emph{machine learning}. According to Tom Mitchell and co-authors, ``machine learning research seeks to develop computer systems that automatically improve their performance through experience'' (Mitchell et al., 1990). This is our favourite definition of machine learning among the many found in the literature, but we note that achieving such a system is incredibly difficult. Most methods that go by the name of ``machine learning'' fail to meet the definition in terms of \emph{automatically} improving performance. Nonetheless, the discipline of machine learning has grown in recent decades, and the set of techniques going by the name of machine learning are indeed very powerful. In many ways, machine learning has become the preferred framework for building up systems that understand users' needs. Some observers may note that such systems exhibit---or at least attempt to exhibit---\emph{artificial intelligence}.

Artificial intelligence has been an elusive goal of computer science researchers ever since the term was coined in 1955\footnote{Although McCarthy is usually credited with coining the term artificial intelligence, we note that its first usage in the literature was a paper co-authored by McCarthy, Minsky, Rochester, and Shannon (McCarthy et al., 1955). Therefore, it is not entirely clear who first came up with this term.}. Although computers have not yet replicated human intelligence in a general sense, there are many tasks of increasing complexity that computers can already perform equally well or even better than the most gifted, well-trained humans. As detailed in (Brynjolfsson and McAfee, 2014), computers have been programmed to beat even the best human players of the game-show \emph{Jeopardy!}, to write corporate earnings previews for \emph{Forbes.com} that are indistinguishable from ones written by humans, and to diagnose breast cancer from images of tissue as good as or even better than pathologists could otherwise do\footnote{To be precise, what Brynjolfsson and McAfee describe is a system, known as C-Path, that helped to diagnose breast cancer and also identified new features of breast cancer tissue that were shown to be good features for predicting survival.}. Such examples demonstrate the increasing practicality of artificial intelligence, but what about understanding users' needs? Is it possible for a computer or computing system (including various sensors) to know what its user needs or wnats before he or she makes any keystroke or swipes any touchscreen? Such a system would be considered by many to exhibit a high level of artificial intelligence.

\section{Research objectives and Scope}
\label{sec:objectives}

The goal stated above is ambitious and open-ended. It is our view that we are not even close to unleashing the full potential of computing devices to understand their users. In many ways, smartphones and other so-called \emph{smart devices} are not yet ``smart''. They have the ``braun'' and not the brains, in the sense that they are powerful and capable but deficient in understanding the user's needs. This thesis aims to improve the state-of-the-art in a computer's ability to understand situations or contexts that humans find themselves in. Mobile computing researchers have adopted the term \emph{context awareness} to refer to this ability. In other domains, such as aviation, maritime, and military domains, the term used is situational awareness (or situation awareness)\footnote{For consistency, in this thesis we primarily use the term context awareness, although it can be considered synonymous with the term situation(al) awareness.}. In particular, this thesis will focus on how machine learning can be utilized for building context or situation awareness, in order to solve problems in navigation. Thus, we have limited the scope of the research to a reasonably-sized domain. That being said, improvements in the state-of-the-art in context awareness have wide-ranging applications, and it is our hope that the few example applications given in this thesis are seen as merely examples and not as end goals in themselves.

In this thesis, we focus on three tasks related to context awareness that are relevant to the field of navigation: (1) to recognize the mode of motion that a smartphone user is undergoing outdoors and (2) to recognize the activity of a smartphone user in an indoor office environment, and (3) to determine the optimal path of a ship travelling through ice-covered waters. These tasks are very different from one another, especially the third task with respect to the first two, demonstrating the breadth of problems encompassed by the topic of context awareness. They were chosen, in part, to show how machine learning can be a powerful tool to tackle a wide range of different problems. They also demonstrate wildly different aspects of ``understanding users' needs'' for different types of users.

The first task is important for navigation because a navigation system can adapt and improve its performance based on the motion mode in which it is used, but it would be easier if the user did not have to manually change the modes of the navigation system when he or she transitions, e.g. from walking to driving. In other words, a context-aware navigation system would automatically that a pedestrian user needs a pedestrian navigation system and a driving user needs a car navigation system; it would adapt itself automatically according to these different needs.

Similarly, the second task provides possible enhancements for a navigation or position tracking system that must work also indoors. For example, if the system detects that a user is working in a static position (e.g. seated at a desk), then it can apply a positioning filter that assumes little or no changes in user position (and perhaps go into a low power consumption mode), but when it detects that the user has stood up, it can change the filter to one that assumes greater possibilities for movement. If the system later detects that the user has done some routine activity, e.g. fetched a fresh cup of coffee, it can apply a post-processing filter to refine the position tracking history, perhaps removing outliers or some other desired refinement.

The third task is a rather classic problem in maritime navigation, but surprisingly this function has been and continues to be performed in a manual way (i.e. the ship captain or navigator manually choosing the route based on ice charts, local observations, and experience). It is also becoming increasingly important to find efficient paths through ice-covered waters due to the opening up of northern sea routes, as well as increased wintertime maritime transport in general (e.g. in the Baltic Sea). In terms of understanding the users needs, this capabiity means that if maritime conditions change such that the captain or navigator needs to alter its route, based on changing ice conditions or other factors, an ``ice-aware'' navigation system could automatically inform the ship's crew that a new route is recommended and even suggest the optimal route to the crew.

A plethora of other examples of the utility of context awareness could be given, even within the strict confines of navigation, but due to limitations in time, this thesis will only investigate the above three examples, which have been researched and published in separate publications and republished here for completeness.

\section{Main Contributions}
\label{sec:contributions}

This research provides a fruitful and previously underexamined link between machine learning and context awareness and exploited this link to demonstrate possible applications in the field of navigation. The author has developed and described in this thesis a conceptual framework for the multi-step processing of raw sensor data into contextual information and also provided a framework for describing contextual information in terms of seven key questions. This theoretical or conceptual development, covered in Chapter~\ref{ch:context_awareness}, benefits the research community by making abstract and ambigious concepts such as ``context'' and ``context awareness'' more concrete and clearly defined. During the timeframe which this thesis covers, the author has co-authored a textbook titled \emph{Geospatial Computing in Mobile Devices}, where the mentioned frameworks were described in detail. Two of the chapters from this textbook, where the author of this thesis was the primary author, are included in this thesis, dealing with the topics ``context awareness'' [P1] and ``contextual reasoning'' [P2].

In addition, this thesis contains three scientific publications, previously published in scientific journals or scientific conference proceedings. The contributions of the author to these works include:

1. A systematic evaluation of various machine learning algorithms applied to the problem of detecting ''mobility contexts``, including consideration of the computational cost of the resulting classifiers, due to the intended use in mobile devices [P3].

2. Contributing to the development of a combined probabilistic Location-Motion-Context (LoMoCo) model used to detect human behavior (i.e. activities) in an indoor office environment [P4]. This work includes development of a ubiquitous positioning system, an effort to which the author contributed to over a several year period and documented extensively in several other publications.

3. The development and preliminary evaluation of a novel algorithm for optimizing the routes of ships travelling in ice-covered waters, a so-called ''ice-aware`` route optimization system [P5]. The continued development of this algorithm is being carried out in several ongoing research projects

\section{Thesis Outline}
\label{sec:outline}
 
The remainder of this thesis is organized as follows. Chapter 2 provides a theoretical and historical overview of the topic of context awareness. Chapter 3 provides an overview of machine learning. Chapter 4 summarizes and provides an overview of the included publications. Finally, Chapter 5 offers some conclusions that can be drawn from the author's overall work to date in context awareness and provides some suggestions for future areas of research and development.