Information Technology Reference
In-Depth Information
as possible in the development process, i.e., starting from the requirements and
risk analysis, as well as the architectural and detailed design. An explicit rep-
resentation of security needs and security mechanisms is the stepping stone to
systematically support the evolution of security in Eternal Systems.
4.5 Adaptation and Awareness through Machine Learning
The simultaneous explosion of information, the integration of technology, and the
continuous evolution from software-intensive systems to ultra-large-scale systems
require new and innovative approaches for building, running, and managing soft-
ware systems. Self-adaptation - systems that are able to adjust their behaviour
in response to their perception of the environment and the system itself has
become an important research topic.
Adaptation can be seen as an intelligent function that can automatically select
different functionalities, e.g., by composing different software components. Such
function can hardly be based on predefined handcrafted rules since predicting the
future working conditions caused by changes in the environment or in the user
requirements is too complex. As previously mentioned ML owns two important
properties; it can:
-
learn its function models using millions of variables, accurately describing
the system and environment conditions; and
-
use a probabilistic characterization of the environment to produce the most
effective evolution choice. Such management of uncertainty also produces
evolution functions that are robust to unexpected conditions.
However, the use of ML requires the modelling of system and environment
conditions in terms of input objects for the target learning algorithm. As previ-
ously mentioned structural kernels can help the definition of such objects. The
most comprehensive examples of the use of kernels can be found in automatic
extraction tools that harvest the unstructured data sources that abound on the
web.
5 Future Research
5.1 Diversity Awareness
In the early phases of software development, such as requirements, the antic-
ipated diversity of the set of systems to be developed has to be discovered
and specified by suitable modelling approaches. Current practice captures vari-
ability at the requirements level mainly by domain models, feature models or
by decision-oriented modelling concepts. This so called problem space variabil-
ity is well understood and can be rigorously analysed by mathematical means.
However, it is mostly disconnected from solution space variability that has to
be formulated in terms of the reusable development artefacts. This highlights
the need for new variability modelling techniques that can be integrated into a
Search WWH ::
Custom Search