1,721,158 research outputs found
Finding Resilient and Energy-saving Control Strategies in Smart Homes
No full textAbstractEvolutionary computing has demonstrated its effectiveness in supporting the development of robust and intelligent systems: when used in combination with formal and quantitative models, it becomes a primary tool in critical systems. Among the modern critical infrastructures, smart energy grids are getting a growing interest from many communities (academic, industrial and political) fostering the development of a robust energy distribution infrastructure. Energy grids are also an example of critical cyber physical social systems since their equilibrium can be perturbed not only by cyber and physical attacks but also by economical and social crises as well as changes in the consumption profiles. The paper illustrates a practical framework supporting the run-time evolution of the control logic inside the Smart Meter: the centre of modern Smart Homes. By combining the modeling and analysis capabilities of Fluid Stochastic Petri Nets and the flexibility of Genetic Programming, this approach can be used to adapt the control logic of the Smart Meters to the changes of the structure and functionalities of the Smart Home as well as of the operational environment. While the main objective of the evolution is to guarantee the energetic sustainability of the Smart Home, the fulfilment of the user's requirements about the energetic need of the home allows to preserve the identity of the Smart Meter during its evolution
A model-driven methodology to evaluate performability of metro systems
No full textMetro systems are required to continuously achieve acceptable levels of reliability, availability, maintainability, and performance (performability) in order to comply with the target values reported in operation and maintenance contracts. These requirements are regulated by several international standards that control the lifecycle defining both processes, documentation flows, and enabling techniques, aiming at controlling disturbances on service performed by the system. This chapter focuses on a complete modeldriven methodology with the aim to support the performability evaluation of a metro system during design and In-Service phases, as well as requirements assessment. In detail, the methodology allows the automatic generation of those formal models required for performability analysis, specialized according to the specific track layout and the defined operational strategies. The proposed methodology is perfectly coherent with the European Standard CENELEC EN 50126 and it allows the generation of all the technical reports needed in the related documentation
Automatic Resource Allocation for High Availability Cloud Services
No full textAbstractThis paper proposes an approach to support cloud brokers finding optimal configurations in the deployment of dependability and security sensitive cloud applications. The approach is based on model-driven principles and uses both UML and Bayesian Networks to capture, analyse and optimise cloud deployment configurations. While the paper is most focused on the initial allocation phase, the approach is extensible to the operational phases of the life-cycle. In such a way, a continuous improvement of cloud applications may be realised by monitoring, enforcing and re-negotiating cloud resources following detected anomalies and failures
Combining Genetic Algorithm and Fault Tree Analysis in Reliability/Cost Optimization for Critical Complex Systems
No full textOn managing security in smart e-health applications
No full textDistributed machine learning can give an adaptable but strong shared condition for the design of trusted AI applications; this is mainly due to lack of privacy of centralised remote learning mechanisms. This notwithstanding, also distributed approaches have been compromised by several attack models (mainly data poisoning): in such a situation, a malicious member of the learning party may inject bad data. As such applications are growing in criticality, learning models must face with security and protection just as with versatility issues. The aim of the paper is to improve these applications by providing extra security features for distributed and federated learning mechanisms: more in the details, the paper examines specific concerns such as the utilisation of blockchain, homomorphic cryptography and meta-modelling techniques to ensure protection as well as other non-functional properties
An Artificial Intelligence Approach for Automated Asset Management of Railway Systems
Full text linkAutomated diagnostic and predictive asset management capabilities are of paramount importance in the era of connected and automated cooperative mobility. A diagnostic vehicle can scan the rail network and process sensor measurements to prevent incoming disruptions and ensure smooth operation of automated transportation services. This requires the development of reliable algorithms that enable early warning and predictive asset management. An algorithm based on artificial intelligence techniques is presented here. The algorithm analyses diagnostic measures and relates them to observed faults on the rail network. In operation mode, the algorithm predicts maintenance needs based on current measurements
Effects of hidden layer sizing on CNN fine-tuning
No full textSome applications have the property of being resilient, meaning that they are robust to noise (e.g. due to error) in the data. This characteristic is very useful in situations where an approximate computation allows to perform the task in less time or to deploy the algorithm on embedded hardware. Deep learning is one of the fields that can benefit from approximate computing to reduce the high number of involved parameters thanks to its impressive generalization ability. A common approach is to prune some neurons and perform an iterative re-training with the aim of both reducing the required memory and to speed-up the inference stage. In this work we propose to face CNN size reduction from a different perspective: instead of reducing the network weights or look for an approximated network very close to the Pareto frontier, we investigate whether it is possible to remove some neurons only from the fully connected layers before the network training without substantially affecting the network performance. As a case study, we will focus on “fine-tuning”, a branch of transfer learning that has shown its effectiveness especially in domains lacking effective expert-designed features. To further compact the network, we apply weight quantization to the convolutional kernels. Results show that it is possible to tailor some layers to reduce the network size, both in terms of the number of parameters to learn and required memory, without statistically affecting the performance and without the need for any additional training. Finally, we investigate to what extent the sizing operation affects the network robustness against adversarial perturbations, a set of approaches aimed at misleading deep neural networks
- …
