Page 67 - Trust in ICT 2017

Trust in ICT 1

Figure 16 – The agent-environment interaction [53]

The reinforcement learning agent and its environment interact over a sequence of discrete time steps. The
specification of their interface defines a particular task: the actions are the choices made by the agent; the
states are the basis for making the choices; and the rewards are the basis for evaluating the choices.
Everything inside the agent is completely known and controllable by the agent; everything outside is
incompletely controllable but may or may not be completely known. A policy is a stochastic rule by which the
agent selects actions as a function of states. The agent's objective is to maximize the amount of reward it
receives over time.

Another interesting application related to trust decision implementation is proposed in [54] based on well-
known Kalman Theory [55]. It has proposed an autonomic and lightweight computational trust model for
pervasive systems based on a Kalman filter. When a service delivery occurs, a number of attributes describing
the quality of the service are measured and compared against the promised values; these discrepancies are
used to train a Kalman filter to assess the trustworthiness of a service provider.

Basic example is presented to explain the techniques involved with Kalman theory to achieve decision making
capability. For instance, let’s suppose client device A is willing to assess the trustworthiness of server device
B before deciding whether to interact with (e.g. request a service from) B or not. It does so by means of a
basic Kalman filter that predicts B’s trustworthiness at time t + 1 based on t previous observations of B’s
behaviour (direct experiences).

After each observation, the filter updates its inner state, so to make a more accurate estimate the next time.
The Kalman filter is particularly appealing to IoT as it is extremely light-weight, both in terms of memory
requirements and computational load (the recursive Kalman equations can be efficiently computed, adding
a negligible overhead on the device). Moreover, even in its simplest formulation, the Kalman filter is able to
capture many facets of human trust: it makes a prediction based on an arbitrary long history of interactions;
it implicitly represents the concept of confidence in the trust prediction, as the more frequently A interacts
with B, the more quickly the filter stabilises and reduces the distance between prediction and actual state;
finally, it enables simple yet effective modelling of the subjective nature of trust by means of the
measurement and system errors. In particular to model cautiousness of behaviours and to model confidence.

7.5 Specify key functionalities and standard interfaces for autonomic decision making
An autonomic system must be able to configure itself according to high-level policies and objectives, thereby
improving its effectiveness. One of the most important goals of self-configuration is the ability of a system to
reconfigure itself online, seamlessly incorporating new components while existing ones adapt to these new
features. On the other hand an autonomic decision making system (self-optimization) must be capable of
monitoring and tuning itself according to performance analysis. Performance-based tuning strategies play a
key role in the autonomic trust computing systems definition and are strictly related to the decision making
process.

Furthermore, the decision make process directly related may properties of the Trustor and Trustee.
According to [56], these influencing properties can be categorized in to five items as below:
• Trustee's objective properties, such as a trustee's security and dependability. Particularly,
reputation is a public assessment of the trustee regarding its earlier behaviours and performance.

59