The EternalS Roadmap – Defining a Research Agenda for Eternal Systems - Trustworthy Eternal Systems via Evolving, Software Data and Knowledge - page 145

5.4 Security
Risk Management as described above is a key process in securing software sys-
tems. Currently, risk analysis is a manual and time consuming process and as
such creating methodologies and finding ways of automating this process is a
good candidate for research within the context of Eternal Systems. Some ex-
isting research work has been completed and some tools (CORAS, Proseco) are
available in this area. The SecureChange project, part of the Forever Yours group
of FET projects has completed some research in the area of systematic processes
for risk analysis.
Security and how it is applied to the various branches of Cloud Computing
will continue to be a very important area of research with immediate practical
applications.
5.5 Machine Learning
The key challenge for Machine Learning in the context of Eternal Systems is to
develop methods that allow systems to adapt and evolve as their environment
changes. For instance, we may encounter problems such as:
- Adaptation of legacy systems.
- Reconciliation of systems whose interfaces are evolving.
- Introduction of new components in an existing environment before and after
- Automatic risk analysis to deal with evolving security concerns.
Machine-learned systems may need to evolve as the distribution of the data
on which they operate evolves. This is particularly true for natural language
processing systems since new terms frequently enter the vocabulary. Component-
based systems need adaptation mechanisms as partly incompatible components
are introduced and some components become obsolete.
A further research area is the automatic application of user requirements by
means of natural language processing. The latter can automatically interpret the
modifications in user requirements and convert them in actions to make evolve
software system, e.g., by selecting new components in the system.
An important area where ML can be brought to bear on software engineering
and variability management is automated test case generation and execution.
This would help solve the major problem of testing in SPLE for product deriva-
tion and application engineering use cases
Further Machine Learning challenges in the domain of software engineering
of Eternal Systems include:
- Researching the correct criteria for the selection of state of the art machine
learning techniques such as Finite State Automata and Kernel Methods,
particularly Support Vector Machines and their application to real world
problem domains that today can only be addressed with simplified or inad-
equate models.