This thesis addresses orchestration of servers amongst multiple co-hosted internet services such as e-Banking, e-Auction and e-Retail in hosting centres. The hosting paradigm entails levying fees for hosting third party internet services on servers at guaranteed levels of service performance. The orchestration of server ensemble in hosting centres is considered in the context of maximising the hosting centre's revenue over a lengthy time horizon. The inspiration for the server orchestration approach proposed in this thesis is drawn from nature and generally classed as swarm intelligence, specifically, sophisticated collective behaviour of social insects borne out of primitive interactions amongst members of the group to solve problems beyond the capability of individual members. Consequently, the approach is self-organising, adaptive and robust. A new scheme for server ensemble orchestration is introduced in this thesis. This scheme exploits the many similarities between server orchestration in an internet hosting centre and forager allocation in a honeybee (Apis mellifera) colony. The scheme mimics the way a honeybee colony distributes foragers amongst flower patches to maximise nectar influx, to orchestrate servers amongst hosted internet services to maximise revenue. The scheme is extended by further exploiting inherent feedback loops within the colony to introduce self-tuning and energy-aware server ensemble orchestration. In order to evaluate the new server ensemble orchestration scheme, a collection of server ensemble orchestration methods is developed, including a classical technique that relies on past history to make time varying orchestration decisions and two theoretical techniques that omnisciently make optimal time varying orchestration decisions or an optimal static orchestration decision based on complete knowledge of the future. The efficacy of the new biomimetic scheme is assessed in terms of adaptiveness and versatility. The performance study uses representative classes of internet traffic stream behaviour, service user's behaviour, demand intensity, multiple services co-hosting as well as differentiated hosting fee schedule. The biomimetic orchestration scheme is compared with the classical and the theoretical optimal orchestration techniques in terms of revenue stream. This study reveals that the new server ensemble orchestration approach is adaptive in a widely varying external internet environments. The study also highlights the versatility of the biomimetic approach over the classical technique. The self-tuning scheme improves on the original performance. The energy-aware scheme is able to conserve significant energy with minimal revenue performance degradation. The simulation results also indicate that the new scheme is competitive or better than classical and static methods.

The demand for automated decision making, learning and inference in uncertain, risk sensitive and dynamically changing situations presents a challenge: to design computational approaches that promise to be widely deployable and flexible to adapt on the one hand, while offering reliable guarantees on safety on the other. The tension between these desiderata has created a gap that, in spite of intensive research and contributions made from a wide range of communities, remains to be filled. This represents an intriguing challenge that provided motivation for much of the work presented in this thesis. With these desiderata in mind, this thesis makes a number of contributions towards the development of algorithms for automated decision-making and inference under uncertainty. To facilitate inference over unobserved effects of actions, we develop machine learning approaches that are suitable for the construction of models over dynamical laws that provide uncertainty bounds around their predictions. As an example application for conservative decision-making, we apply our learning and inference methods to control in uncertain dynamical systems. Owing to the uncertainty bounds, we can derive performance guarantees of the resulting learning-based controllers. Furthermore, our simulations demonstrate that the resulting decision-making algorithms are effective in learning and controlling under uncertain dynamics and can outperform alternative methods. Another set of contributions is made in multi-agent decision-making which we cast in the general framework of optimisation with interaction constraints. The constraints necessitate coordination, for which we develop several methods. As a particularly challenging application domain, our exposition focusses on collision avoidance. Here we consider coordination both in discrete-time and continuous-time dynamical systems. In the continuous-time case, inference is required to ensure that decisions are made that avoid collisions with adjustably high certainty even when computation is inevitably finite. In both discrete-time and finite-time settings, we introduce conservative decision-making. That is, even with finite computation, a coordination outcome is guaranteed to satisfy collision-avoidance constraints with adjustably high confidence relative to the current uncertain model. Our methods are illustrated in simulations in the context of collision avoidance in graphs, multi-commodity flow problems, distributed stochastic model-predictive control, as well as in collision-prediction and avoidance in stochastic differential systems. Finally, we provide an example of how to combine some of our different methods into a multi-agent predictive controller that coordinates learning agents with uncertain beliefs over their dynamics. Utilising the guarantees established for our learning algorithms, the resulting mechanism can provide collision avoidance guarantees relative to the a posteriori epistemic beliefs over the agents' dynamics.

The growth in the interdependence and complexity of socio-technical systems requires the development of tools and techniques to aid in the formulation of better policies. The efforts of this research focus towards developing methodologies and support tools for better policy design and formulation. In this thesis, a new framework and a systematic approach for the formulation of policies are proposed. Focus has been directed to the interactions between policy measures, inspired by concepts in process design and network analysis. Furthermore, we have developed an agent-based approach to create a virtual environment for the exploration and analysis of different configurations of policy measures in order to build policy packages and test the effects of changes and uncertainties while formulating policies. By developing systematic approaches for the formulation and analysis of policies it is possible to analyse different configuration alternatives in greater depth, examine more alternatives and decrease the time required for the overall analysis. Moreover, it is possible to provide real-time assessment and feedback to the domain experts on the effect of changes in the configurations. These efforts ultimately help in forming more effective policies with synergistic and reinforcing attributes while avoiding internal contradictions. This research constitutes the first step towards the development of a general family of computer-based systems that support the design of policies. The results from this research also demonstrate the usefulness of computational approaches in addressing the complexity inherent in the formulation of policies. As a proof of concept, the proposed framework and methodologies have been applied to the formulation of policies that deal with transportation issues and emission reduction, but can be extended to other domains.