Final Report Summary - NH-MCDM (New Horizons for Multi Criteria Decision Making)
In this project the consortium intended to bring about a major step change in the decision-making process for complex interdisciplinary problems. Such problems usually require the simultaneous consideration of multiple performance criteria. Multi-objective optimization (MOO), and particularly evolutionary multi-objective optimization (EMO), is now seen as having the potential to have a major impact in such problems, as evidenced by the rapidly growing number of research publications in this field and by the availability of a number of related software tools and users, both in academia and in industry. As part of this new direction, Multi Criteria Decision Making (MCDM) based on the results obtained through EMO/MOO is an important challenge. Although the popularity of MOO and EMO is growing, their development faces many challenges, among them how to improve the representations of solutions to MOPs, how to solve many-objective problems, and how to provide support for decision-making. Through research addressing the above challenges, the consortium aimed to build a multi-criteria decision-making framework for solving complex problems using state-of-the-art optimization software. This ambitious project aims at to providing a flexible, accessible, and structured approach for managing the design process in such a way that it is transparent, robust, and evidence-based. The consortium included leading researchers in the field of evolutionary multi-objective optimization from five countries: Israel, UK, Brazil, Mexico and Canada.
The consortium's objectives for this project included: (1) achieving an exchange of knowledge through conferences, seminars, and courses; (2) advancing the theoretical background of EMO, thus enhancing its ability to be utilized in industrial applications; and (3) enhancing future cooperation by opening new research directions.
To achieve the first objective of an exchange of knowledge, the consortium lead and has been involved in organizing and executing several related conferences and workshops such as EMO2013, EVOLVE2012 and the first Israeli Workshop on Multi-Criteria optimization and decision making. Moreover, during almost hundred months of mutually spent visits, the partners have conducted many seminars, short courses and brainstorming sessions that supported building and developing the collaboration’s foundations thus boosting the mutual research.
To achieve the second objective of advancing the theoretical background of EMO, the consortium has carried out extensive research studies related to EMO algorithms, which have resulted in novel applicable algorithms such as PICA, PSA, PSA-NSGA-II, INSPM and more. The PICA algorithm has already been shown to be competent with state-of-the-art algorithms. The PSA algorithm has opened the way for optimizing functions and for developing new approaches for optimizing topologies that is considered to be a very important yet difficult task for any optimizer. The INSPAM and related research have paved the way for a profound improvement in decision-makers' ability to make decisions regarding the solution of problems involving many-design parameters and many objectives.
The aspiration to engage industry through suggesting new approaches and paradigms has led to the development of several new research directions. One such example is the development of a novel methodology for optimizing products for both single objective and multi-objective problems. This new methodology enables the optimizing of products that adapt when subjected to changes in the environment, thus enhancing their robustness. The evolution of optimal, actively robust solutions, necessitated the development of new set-based algorithms. To that end, new measures have been developed, in turn opening the way for further research on complexity, parallelism, and benchmarking. This optimization approach, known as "active robust optimization," is attracting interest and is now further elaborated as part of an EPSRC and Jaguar & Landover grant.
Another application for EMO and MCDM has been the topic of extensive research being carried out on multi-objective games. Indeed, games of this type are usually avoided due to their high complexity. Yet not only has our research suggested evolutionary algorithms to solve both discrete and continuous multi-objective games, it has also revealed a new equilibrium for such games. This equilibrium is being further elaborated to introduce a tool to support the players of such games in making decisions on optimal strategies. The relative simplicity of using an evolutionary search when compared to elaborated mathematical approaches for solving multi objective games, opened the way for the introduction of a new salesman problem, namely the multi-objective salesman game. Solving such a game seems to have clear applications in decision making within businesses that compete on the same resources.
Other research activities have aimed at further extending the current state of the art on EMO. One such activity involved the embedment of a novel measure for solving topology optimization problems that include multi-modality, by utilizing topologies represented through finite elements. Another extension focuses on new measures embedded in an interactive search that hybridize radial neural nets with a model of decision-makers' preferences within EMO algorithms. In addition to developing unique paradigms, research within the consortium has targeted the construction of frameworks for supporting multiple decision-makers while considering solutions to complex design processes. These included the embedding of interactive approaches within hybrid algorithms that depend upon the amalgamation of neural nets, local gradient analysis, and robustness considerations.
To meet the third objective of opening new research directions, the consortium carried out novel and exciting research studies. One such stimulating study has explored the notion that robots' robustness to uncertain scenarios and malfunctions may be enhanced through what has been termed as mechanical cognitivization. This enhanced robustness is attained by training the robots by means of mechanically restricted modes (e.g. training them to perform a task while not using all their possible capabilities). This training method is commonly adopted by humans while training for a particular sports activity (e.g. when a climber or a swimmer is prohibited from using one of his hands). An initial proof-of-concept highlights the potential of this research direction. To this end we are working on developing EMO algorithms that will enhance the evolution of robots that would benefit the most from such training.
The consortium's objectives for this project included: (1) achieving an exchange of knowledge through conferences, seminars, and courses; (2) advancing the theoretical background of EMO, thus enhancing its ability to be utilized in industrial applications; and (3) enhancing future cooperation by opening new research directions.
To achieve the first objective of an exchange of knowledge, the consortium lead and has been involved in organizing and executing several related conferences and workshops such as EMO2013, EVOLVE2012 and the first Israeli Workshop on Multi-Criteria optimization and decision making. Moreover, during almost hundred months of mutually spent visits, the partners have conducted many seminars, short courses and brainstorming sessions that supported building and developing the collaboration’s foundations thus boosting the mutual research.
To achieve the second objective of advancing the theoretical background of EMO, the consortium has carried out extensive research studies related to EMO algorithms, which have resulted in novel applicable algorithms such as PICA, PSA, PSA-NSGA-II, INSPM and more. The PICA algorithm has already been shown to be competent with state-of-the-art algorithms. The PSA algorithm has opened the way for optimizing functions and for developing new approaches for optimizing topologies that is considered to be a very important yet difficult task for any optimizer. The INSPAM and related research have paved the way for a profound improvement in decision-makers' ability to make decisions regarding the solution of problems involving many-design parameters and many objectives.
The aspiration to engage industry through suggesting new approaches and paradigms has led to the development of several new research directions. One such example is the development of a novel methodology for optimizing products for both single objective and multi-objective problems. This new methodology enables the optimizing of products that adapt when subjected to changes in the environment, thus enhancing their robustness. The evolution of optimal, actively robust solutions, necessitated the development of new set-based algorithms. To that end, new measures have been developed, in turn opening the way for further research on complexity, parallelism, and benchmarking. This optimization approach, known as "active robust optimization," is attracting interest and is now further elaborated as part of an EPSRC and Jaguar & Landover grant.
Another application for EMO and MCDM has been the topic of extensive research being carried out on multi-objective games. Indeed, games of this type are usually avoided due to their high complexity. Yet not only has our research suggested evolutionary algorithms to solve both discrete and continuous multi-objective games, it has also revealed a new equilibrium for such games. This equilibrium is being further elaborated to introduce a tool to support the players of such games in making decisions on optimal strategies. The relative simplicity of using an evolutionary search when compared to elaborated mathematical approaches for solving multi objective games, opened the way for the introduction of a new salesman problem, namely the multi-objective salesman game. Solving such a game seems to have clear applications in decision making within businesses that compete on the same resources.
Other research activities have aimed at further extending the current state of the art on EMO. One such activity involved the embedment of a novel measure for solving topology optimization problems that include multi-modality, by utilizing topologies represented through finite elements. Another extension focuses on new measures embedded in an interactive search that hybridize radial neural nets with a model of decision-makers' preferences within EMO algorithms. In addition to developing unique paradigms, research within the consortium has targeted the construction of frameworks for supporting multiple decision-makers while considering solutions to complex design processes. These included the embedding of interactive approaches within hybrid algorithms that depend upon the amalgamation of neural nets, local gradient analysis, and robustness considerations.
To meet the third objective of opening new research directions, the consortium carried out novel and exciting research studies. One such stimulating study has explored the notion that robots' robustness to uncertain scenarios and malfunctions may be enhanced through what has been termed as mechanical cognitivization. This enhanced robustness is attained by training the robots by means of mechanically restricted modes (e.g. training them to perform a task while not using all their possible capabilities). This training method is commonly adopted by humans while training for a particular sports activity (e.g. when a climber or a swimmer is prohibited from using one of his hands). An initial proof-of-concept highlights the potential of this research direction. To this end we are working on developing EMO algorithms that will enhance the evolution of robots that would benefit the most from such training.