Search

The game of Go has attracted much attention from the artificial intelligence community. A key feature of Go is that humans begin to learn on a small board, and then incrementally learn advanced strategies on larger boards. While some machine learning methods can also scale the board, they generally only focus on a subset of the board at one time. Neuroevolution algorithms particularly struggle with scalable Go because they are often directly encoded (i.e. a single gene maps to a single connection in the network). Thus this paper applies an indirect encoding to the problem of scalable Go that can evolve a solution to 5×5 Go and then extrapolate that solution to 7×7 Go and continue evolution. The scalable method is demonstrated to learn faster and ultimately discover better strategies than the same method trained on 7×7 Go directly from the start. [ABSTRACT FROM AUTHOR]

While at present Evolutionary Robotics (ER) is generally not studied in mainstream robotics, the main idea in this article is that ER has the opportunity to gain relevance by taking seriously its natural inspiration. The chasm that separates the behavior of robots today from the robustness and fluidity of organisms in nature is most naturally addressed by an approach that indeed respects the process through which such organisms originated. Yet the challenge is to identify the elusive missing ingredient that would allow ER to realize its full potential. [ABSTRACT FROM AUTHOR]

While traditional approaches to machine learning are sensitive to high-dimensional state and action spaces, this paper demonstrates how an indirectly encoded neurocontroller for a simulated octopus arm leverages regularities and domain geometry to capture underlying motion principles and sidestep the superficial trap of dimensionality. In particular, controllers are evolved for arms with 8, 10, 12, 14, and 16 segments in equivalent time. Furthermore, when transferred without further training, solutions evolved on smaller arms retain the fundamental motion model because they simply extend the general kinematic concepts discovered at the original size. Thus this work demonstrates that dimensionality can be a false measure of domain complexity and that indirect encoding makes it possible to shift the focus to the underlying conceptual challenge. [ABSTRACT FROM AUTHOR]

Recently, a small number of papers have appeared in which the authors implement stochastic search algorithms, such as evolutionary computation, to generate game content, such as levels, rules and weapons. We propose a taxonomy of such approaches, centring on what sort of content is generated, how the content is represented, and how the quality of the content is evaluated. The relation between search-based and other types of procedural content generation is described, as are some of the main research challenges in this new field. The paper ends with some successful examples of this approach. [ABSTRACT FROM AUTHOR]

Biological brains can adapt and learn from past experience. In neuroevolution, i.e. evolving artificial neural networks (ANNs), one way that agents controlled by ANNs can evolve the ability to adapt is by encoding local learning rules. However, a significant problem with most such approaches is that local learning rules for every connection in the network must be discovered separately. This paper aims to show that learning rules can be effectively indirectly encoded by extending the Hypercube-based NeuroEvolution of Augmenting Topologies (HyperNEAT) method. Adaptive HyperNEAT is introduced to allow not only patterns of weights across the connectivity of an ANN to be generated by a function of its geometry, but also patterns of arbitrary learning rules. Several such adaptive models with different levels of generality are explored and compared. The long-term promise of the new approach is to evolve large-scale adaptive ANNs, which is a major goal for neuroevolution. [ABSTRACT FROM AUTHOR]

A relatively rare application of artificial intelligence at the nexus of art and music is dance. The impulse shared by all humans to express ourselves through dance represents a unique opportunity to artificially capture human creative expression. In particular, the spontaneity and relative ease of moving to the music without any overall plan suggests a natural connection between temporal patterns and motor control. To explore this potential, this paper presents a model called Dance Evolution, which allows the user to train virtual humans to dance to MIDI songs or raw audio, that is, the dancers can dance to any song heard on the radio, including the latest pop music. The dancers are controlled by artificial neural networks (ANNs) that ˵hear″ MIDI sequences or raw audio processed through a discrete Fourier transform-based technique. ANNs learn to dance in new ways through an interactive evolutionary process driven by the user. The main result is that when motion is expressed as a function of sound the effect is a plausible approximation of the natural human tendency to move to music. [ABSTRACT FROM AUTHOR]