Project

Back to overview

Viability Evolution

English title Viability Evolution
Applicant Floreano Dario
Number 127143
Funding scheme Project funding (Div. I-III)
Research institution Laboratory of Intelligent Systems School of Engineering EPFL
Institution of higher education EPF Lausanne - EPFL
Main discipline Information Technology
Start/End 01.05.2010 - 31.08.2013
Approved amount 459'820.00
Show all

Keywords (9)

Artificial evolution; Evolutionary Robotics; Reverse Engineering; Evolutionary Computation; Multi objective optimization; Incremental Evolution; Open ended Evolution; Robot controllers; Reverse Engineering of Biological Networks

Lay Summary (English)

Lead
Lay summary
Evolution in nature has developed complex beings with varied capabilities that can adapt to changing environments. Evolutionary computation (EC) is the field of science that aims to develop engineering and problem-solving tools by modeling the evolutionary process in nature. Existing EC algorithms are mostly built around the principle of survival of the fittest in nature. This has limited their adaptive potential and placed a large emphasis on the fitness function used to model the problem objectives. In this project, a novel evolutionary framework will be developed that better models evolution in nature by incorporating viability based events. The core intuition is a shift of emphasis from an evolutionary process purely driven by selection of the fittest to one that is founded on the notion of viability of the evolving individuals. The immediate benefits of the proposed framework include the elimination of the need for a single fitness function, and the ability to seamlessly incorporate continuously changing environments and/or task constraints. In order to realize its full potential, the proposed framework will be coupled with the previously developed Analog Genetic Encoding (AGE). The proposed viability evolution framework will then be validated in two scenarios, namely, synthesis of robot control circuits under changing environmental conditions or task constraints, and incremental reverse engineering of large gene regulatory networks. This project is expected to contribute significantly to the fields of multi-objective optimization, and open-ended evolution. From a multi-objective optimization perspective, the proposed framework introduces a radically new method for specifying the multiple objectives using viability constraints. From an incremental evolution perspective, the proposed framework will allow the gradual addition and elimination of viability criteria without the need for redefining the fitness function. The formalization and implementation of dynamic viability will offer a novel, principled way for performing incremental evolution. The two validation scenarios are also expected to have significant impact in their respective fields. The proposed method will offer a novel way to incrementally evolve robotic systems that can adapt to changing environmental conditions, and/or task constraints. The proposed method may allow discovery of the interactions among gene networks of unprecedented complexity and may possibly link network structure with functional effects. Consequently, it could lead to significant advancements that allow the testing of gene pharmacology and personalized medicine.
Direct link to Lay Summary Last update: 21.02.2013

Responsible applicant and co-applicants

Employees

Publications

Publication
Differences in the concept of fitness between artificial evolution and natural selection
Lichocki Pawel, Keller Laurent, Floreano Dario (2012), Differences in the concept of fitness between artificial evolution and natural selection, in Artificial Life 13, 13th International Conference on the Simulation and Synthesis of Living Systems, East Lansing, Michigan, USA.
Historical contingency affects signaling strategies and competitive abilities in evolving populations of simulated robots
Wischmann Steffen, Floreano Dario, Keller Laurent (2012), Historical contingency affects signaling strategies and competitive abilities in evolving populations of simulated robots, in PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED STATES OF AMERICA, 109(3), 864-868.
Beyond Graphs: A New Synthesis
Mattiussi Claudio, Dürr Peter, Marbach Daniel, Floreano Dario (2011), Beyond Graphs: A New Synthesis, in Journal of Computational Science, 2, 165-177.
GeneNetWeaver: in silico benchmark generation and performance profiling of network inference methods
Schaffter Thomas, Marbach Daniel, Floreano Dario (2011), GeneNetWeaver: in silico benchmark generation and performance profiling of network inference methods, in BIOINFORMATICS, 27(16), 2263-2270.
Selection methods regulate evolution of cooperation in digital evolution
Lichocki Pawel, Floreano Dario, Keller Laurent, Selection methods regulate evolution of cooperation in digital evolution, in Journal of the Royal Society Interface.

Associated projects

Number Title Start Funding scheme
112060 Automatic Synthesis and Reverse-engineering of Analog Networks 01.09.2006 Project funding (Div. I-III)

Abstract

This project is part of an ongoing research effort to develop an artificial evolutionary method that approximates the creativity, open-endedness, and complexity of evolving biological systems. In the first part (Swiss NSF project nr. 620-58049, ending in September 2009) of this long-term goal, I focused on the development of a genetic representation suitable for describing and evolving a large class of systems of biological and technological relevance, which could be described as analog networks, that is, collections of devices interconnected by weighted links. Analog networks are typically complex systems since they are characterized by properties such as the presence of dynamics at different time-scales and nonlinear feedback loops, which make them difficult to analyze and synthesize with existing scientific and engineering tools.In this second part, I will focus on the development, implementation, and evaluation of a novel evolutionary framework suitable for both the synthesis of complex networks under changing environment or task requirements and the incremental reverse engineering of biological networks that are too complex to be resolved directly. The new framework will be based on the introduction of the concept of viability in artificial evolution, which will incorporate, where necessary, the more familiar notion of selective reproduction of the fittest. The core intuition on which this project is based consists in a shift of emphasis from an evolutionary process purely driven by selection of the fittest to an evolutionary process that is founded on the notion of viability of the evolving individuals. This apparently simple modification is aimed at drastically changing and simplifying the concept of fitness function in artificial evolution, at naturally incorporating continuously changing environment or task constraints, and at being conducive to an incremental and open-ended evolutionary process.In order to capitalize on the possibility of incremental and open-ended evolution offered by the novel viability framework, it will be necessary to resort to genetic representations that can change in size and complexity, display neutrality, and possibly re-use previously evolved solutions. Therefore, the new evolutionary framework will be coupled with the Analog Genetic Encoding (AGE) developed in the previous project (Swiss NSF project 620-58049, ending in September 2009). AGE is a method for the description and evolution of the topology and links of any analog network. AGE, which captures several representational and dynamic aspects of the biological genome at a computationally tractable level of abstraction, was not only successfully applied to the evolutionary synthesis of electronic and control circuits, but was also awarded the first prize at the international competition on reverse engineering of biological gene regulatory networks (Marbach et al., 2009). While AGE was conceived from the very beginning for future investigations of open-ended evolution, so far it has been validated only with simple evolutionary algorithms based on selective reproduction of the fittest according to pre-designed fitness functions. Consequently, we tackled only stationary problems whose solutions could be numerically assessed and ranked, and we reverse-engineered only biological networks that were relatively small (but still sufficiently challenging for other state-of-the-art algorithms). In this proposal, I envision only some modifications of the AGE method for a better coupling with the novel concept of viability in evolution.I expect that the coupling of the novel viability framework with the previously developed genetic representation will allow a) the incremental synthesis of complex networks under changing environmental or task constraints; and b) the incremental reverse engineering of large biological networks possibly composed of sub-networks. The resulting evolutionary system will be validated in two scenarios: the synthesis of robot control circuits under changing environmental or task constraints and the incremental reverse engineering of large gene regulatory networks. While this novel form of Viability Evolution will capitalize on the potentials offered by AGE, it will be fully compatible with any other genetic representation ranging from fixed-length bit strings all the way to developmental and variable-length representations. Indeed, the first set of thorough experiments aimed at comparatively assessing the novel evolutionary framework will be conducted with more traditional genetic representations than AGE to prevent the combination of multiple factors in the determination of the results. AGE will be introduced only in the second stage of the project when the essential components of the novel evolutionary framework will be functional and tested.
-