School of Mechanical Engineering Seminar
Wednesday, March 7, 2018 at 14:00
Wolfson Building of Mechanical Engineering, Room 206
Co-evolving Rationalizable Strategies for a Multipayoff Cyber-security Game
Roi Chananel
MSc Student of Dr. Amiram Moshaiov
In many application areas, including engineering, biology and economy, conflicts between the involved players may occur. Such situations can often be modeled as non-cooperative Multi-Objective Games (ncMOGs) in a normal form. In such a multipayoff game, each player may have self-conflicting objectives. This study deals with finding the players' rationalizable strategies, assuming they are undecided about their self-objective preferences. Finding their rationalizable strategies, without a-priori decision on the objective preferences, allows the players to explore the performance tradeoffs before making a strategy selection decision.
To search for the rationalizable strategies, a generic co-evolutionary algorithm is suggested, which employs a Hall-of-Fame (HoF) for each of the players. The HoFs are historical memories of the evolutionary search that aim to support convergence. The proposed algorithm is studied using an attack-defense cyber-security ncMOG that was recently introduced. It is believed that the considered game is of a generic value to many defense related problems, e.g., adversarial robotics, in which the defender searches for optimal defense-resource allocation, under the uncertainties of the attacker actions.
The proposed algorithm is evaluated based on a reference set of solutions, which is obtained by sorting the entire decision spaces of the attacker and defender. The algorithm is also compared with a simpler version. The simpler algorithm employs elite populations, which are short memories of one generation, rather than long memories (HoFs). It is shown that both versions are able to find subsets of the rationalizable strategies for the considered game. Yet, as expected, the version with the HoFs is found to be superior. Finally, an analysis of the resulting subsets of the rationalizable strategies is carried out, which suggests that a modification is required when aiming at the entire set of the rationalizable strategies.
Evolution potential of a system
Gadi Tennenbaum
Msc student Prof. Yoram Reich
The question “whether to invest the resources in improving existing technology or to try developing a new technology” is occurring in any big technology company, small technology company, and even new entrepreneur's mind. This work is about decision making tools that will help their user to answer that question. The decision making tools presented in this work are methods for estimating the evolution potential of a system, and using it to decide. System evolution potential is a value that indicates how much the technology of a certain system could be developed until it reaches its development limit. Two methods will be presented; the difference between them is the assumptions that each method employs.
The methods are demonstrated on case studies to illustrate their potential. The two methods are presented with two different case studies – unmanned vehicle navigation system for the Subsystem method, and 3D metal printing for the Physical principles method.
The results suggest that:
- System evolution potential is a good decision-making tool because it can be estimated without dependency on the user’s knowledge about the system. However, the accuracy of the development limit forecast decreases when the user has less knowledge.
- In order to assess and verify the methods accuracy, experiments on different systems need to be done in the future.
