Hybrid Intelligent Systems in Ubiquitous Computing

Introduction

Now ubiquitous computing, ambient intelligence and smart cooperative objects are viewed as a major paradigms shift from conventional desktop application development. This view is enabled through the use of diverse hardware (sensors, user devices, computing infrastructure etc.) and software, anticipating user needs and acting on their behalf in a proactive manner (Weiser, 1991; Satyanarayanan, 2001). This diversity of hardware and software information increases the degree of heterogeneity.

In order to realize such ubiquitous computing environment, three technology areas are required:

• 1)

  Sensing technology where information on user and surrounding environment are perceived and collected,

• 2)

  Context aware computing (Schilit, Adams & Want, 1994; Baldauf & Dustdar, 2004) technology where such information are processed and properly presented to users as different services,

• 3)

  Wireless network technologies (Mahalik, 2007) where information are collected from sensors and distributed to customers – services and users.

One of most perspective technologies for sensing and perception is neural networks.

We may pick out following main features of ubiquitous computing systems (UCS):

• 1)

  distribution of obtaining and processing of sensor information,

• 2)

  variety of information needed processing,

• 3)

  necessity of learning during interaction with environment, in particular, in respect to existing of unexpected events and objects needed for including into processing,

• 4)

  key role of different kinds of human-machine interaction,

• 5)

  high requirements to security,

• 6)

  data processing in real time,

• 7)

  wide usage of embedded processing units.

There are following tasks for neural networks in development of ubiquitous computing systems:

• 1)

  perception, i.e. recognition of objects and changes in environment, in particular, invariant recognition of moving objects, e.g. recognition of gesture, position and emotions of human beings,

• 2)

  clustering and recognition of events and scenarios (sequence of events in time),

• 3)

  prediction of future events and situations,

• 4)

  indoor localization of mobile devices and continues mapping,

• 5)

  reactive behavior based managing of actions,

• 6)

  speech recognition.

From above we can formulate following requirements to neural networks for UCS:

• 1)

  Relatively fast processing of information in both learning and recalling,

• 2)

  Incremental learning, i.e. availability to perceive new information without loss of old knowledge,

• 3)

  Availability of easy extraction of structure from learnt neural network for building of symbolic knowledge for usage in machine-human interaction and planning.

On the other hand context awareness usually is implemented by symbolic based reasoning and knowledge-based techniques (Hung, Shehzad, Kiani, Riaz, Ngoc & Lee, 2004). Besides rules based approach is appropriate for human-machine interface for programming of behavior of smart objects (Tarik, Sarcar, Hasn, Huq, Gavrilov, Lee & Lee, 2008) and for any explanation for user.

The following tasks are more relevant to rule based and other symbolic techniques: