Research Area: Embedded Intelligence

How intelligence embedded in the devices can improve the usability and functionality of IoT services and products.

Embedded Intelligence is one of three research areas IOTAP focuses on. Key issues in this research area include:

  • intelligence of the individual devices
  • interaction between entities
  • management of data

Intelligence of the individual devices

Since large volumes of data will be generated in the IoT, it seems reasonable to embed mechanisms for data processing and decision-making in the devices (or close to the devices, e.g., in a gateway) to enable decentralized processing of information. In some applications there are real-time requirements that may be difficult to handle by remote processing at a server or in a cloud. Also, the robustness of the system can be increased if processing is distributed, as there often is no critical single point of failure. Thus, the scalability can be enhanced in several ways if intelligence is embedded in the individual devices. The research can be characterized as “modern” Artificial Intelligence (AI) where the intelligence is embodied in an “intelligent agent” (Russell and Norvig 2010).

IoTaP focuses on the following crucial issues concerning the intelligence of the individual devices:

  • What are the appropriate models for context-awareness in IoT applications? A device should be able to identify in which situation the device and/or its user currently is, and use this to bring added value. For example, context-awareness can be used to adjust the level of autonomy so that the degree of automation depends on the current situation and user preference or ability, or to adapt the interaction surface.
  • How can the behaviour of the devices be improved through learning from experience? This could be achieved by adapting the behaviour based on direct feedback provided by the user. Another approach is to improve decision-making by learning from sensory input in a more autonomous manner. Moreover, devices may learn from each other, e.g., by sharing their experiences. In some situations, it is important that the algorithms comprising the intelligence of the devices are efficient, e.g., when a large amount of data needs to be processed and the response time is crucial. Also, the power supply limitations of many devices place further requirements on efficient processing and communication.

Interaction between entities

There are a number of interesting research questions concerning the interaction between entities. Agreement technologies (Ossowski 2013) refer to computer systems in which entities, often called (software) agents, negotiate with one another, typically on behalf of humans, in order to come to mutually acceptable agreements. Semantic web and ontologies are two important concepts to realise such systems (Maedche and Staab 2001). Some other key concepts used to specify and verify such systems are semantic alignment, negotiation, virtual organisations, norms, and obligations. Agents can be seen as encapsulating functionality (possibly implemented as web services), and can be orchestrated to create new, higher-level functionality, which could be deployed and executed in a distributed fashion. Such orchestrations are often referred to as multi-agent systems (Wooldridge 2009).

Important research questions that IoTaP addresses in relation to interaction between entities are:

  • How can relevant devices and services automatically be discovered by other devices and services? When more and more devices are connected to the Internet, the problem of finding the relevant one for a particular situation becomes more difficult, especially in new environments. To reduce the cognitive load of the user, service discovery of this kind may be carried out by software agents that do not need continuous user intervention.
  • How can we make the devices, services, and users understand each other in an IoT context? The units need to be semantically interoperable, for example by using ontologies and other semantic web technologies. This is important, for instance, to implement service discovery functionality, as well as for making use of data that have been made publicly available. Also, in order to achieve a task, an agent (a device or a service) often needs to collaborate, or at least coordinate, with other agents. An interaction protocol defines a set of rules that guides the interaction between agents for supporting structured collaboration and negotiation in order to achieve a common or individual goal.
  • What are appropriate architectural designs for heterogeneous distributed IoT systems? What is the best distribution of functionality between smart devices and the supporting infrastructure? How can we make complex open IoT system of systems more robust? For instance, reliability may be increased through redundancies that can cover up for the loss of entities, or norm systems that guide the behaviour of the entities can be introduced. Moreover, risk analysis can be made beforehand, as well as validation through simulation. Another challenge is how to identify and avoid potential undesired effects of embedded intelligence with respect to resource utilization and user behaviour. For instance, bad system design could cause unnecessary system resource utilization peaks or undesirable influence of user behaviour.

Management of data

As the number of connected devices grows at a fast rate, the amount of data they generate will explode, in particular different types of sensor data. Regarding the management of data, we will address the following questions:

  • What are the best ways of utilising the large amounts of data that becomes available? Different methods for achieving this will be studied. For instance, we will investigate data mining methods for extracting relevant data and discovering patterns, data fusion methods for combining different data sources, and visualization techniques to make the data more comprehensible for human users.
  • How can individual user privacy be sustained in a super-connected and data-intense surrounding? In principle, it is desired that as much as possible of the data generated is made available to other services and devices. However, the ways in which user-related data can be collected, analysed, and utilised for various purposes – benign and commercial, as well as selfish and malicious – in a digital ecosystem are uncountable. It is thus important to study the need for and the design of privacy-enhancing mechanisms in order to ensure the privacy of the users. In this respect, other aspects directly linked with human interactions, such as the management of data ownership, the gathering of consent from users, and the robustness of data security methods are also important to study.

Russell, S., Norvig, P. (2010) Artificial Intelligence: A Modern Approach, 3rd Edition, Prentice Hall.

Ossowski, S., editor (2013) Agreement Technologies, Law, Governance & Technology Series, Vol. 8, Springer.

Maedche, A., & Staab, S. (2001) Ontology Learning for the Semantic Web. IEEE Intelligent Systems, Vol. 16(2):72-79.

Wooldridge, M. (2009) An Introduction to MultiAgent Systems, 2nd Edition, John Wiley & Sons.

Comments are closed.