Difference between revisions of "Robustness"

Revision as of 14:56, 23 June 2009

Definition of Robustness

• Robustness
  • Is there a common definition (which is valid for all research areas)?
  • Should there be a definition of robustness?

The adjacent discussion delivered the following: Robustness is a term which has to be specified for every single case. The reason is simply that a system can only be robust against certain disturbances but not against all. Hence, a common definition does not exist.

Robustness as a Design Problem

Even though the term robustness is also quite loose we realized that it can be often routed back to design problems. Usually, only the emergent behavior can be easily observed and by this, it is not trivial how the system should be changed on its component behavior level to help it to increase its robustness.

If there is little knowledge about the local rules how can the we draw conclusions and update the system? [Elaborate this sentence more, pls.]

Internal Robustness

[Please explain the term] There are several ways for designing SOS based on trial-and-error methods, by deriving local rules from model system with extra knowledge or using some intelligent search. However without proper validation and testing there could be a possibility where the emerged behavior shows unanticipated phenomenas.

System Borders

As a final thought, we think it is important to define the current borders of the system before we start dealing with its robustness. As a simple example could be a "robust" algorithm which obviously cannot withstand a blackout. In what way is it robust then?

Robustness in control theory

The research area of control theory seems to us as a classical field where the definitions cannot be brought inline with the theory in this field. For instance, we have a certain system which we would like to control. Robustness in this case means that we would like to adapt to certain disturbances. In order to notice these disturbances, there has to be a kind of feedback loop which allows for reacting on unintended changes in the output of the system. These reactions on unintended changes are carried out by a so-called controller which will feed inputs to the system which has to be controlled. However, this controller can be seen as a centralized unit handling all disturbances. It is thus not clear, if such an approach can be still regarded as a self-organizing system. More in detail, it will either depend on the point of view, from which someone is looking at the system, or on the definition used to characterize a self-organizing system (see Definition_of_Self-Organizing_Systems). For instance, using definition (3), the described system would still fit into the context of a self-organizing system, whereas it would not, if definition (1) is used as reference.

Research Questions

• When properties are emergent, how can you make sure that the

emergent situation is following your intentions? Robustness from within ?