Catálogo de publicaciones - libros

Cultural learning allows individuals to acquire knowledge from others through non-genetic means. The effect of cultural learning on the evolution of artificial organisms has been the focus of much research. This paper examines the effects of cultural learning on the fitness and diversity of a population and, in addition, the effect of self-adaptive cultural learning parameters on the evolutionary process. The NK fitness landscape model is employed as the problem task and experiments employing populations endowed with both evolutionary and cultural learning are compared to those employing evolutionary learning alone.

Our experiments measure the fitness and diversity of both populations and also track the values of two self-adaptive cultural parameters. Results show that the addition of cultural learning has a beneficial effect on the population in terms of fitness and diversity maintenance. Furthermore, analysis of the self-adaptive parameter values shows the relative quality of the cultural process throughout the experiment and highlights the benefits of self-adaptation over fixed parameter values.

Deacon [1] considers that the reverse Baldwin effect can be one of the major forces in language evolution. The reverse Baldwin effect is essentially a redistributional process of genes as a result of environmental changes which mask and unmask selection pressures. Although Deacon indicates that in the case of language evolution, niche construction is deeply involved in masking and unmasking processes, neither specific explanations for the mechanism nor examples have been given. In this study we use evolutionary computation simulations to explore how niche constructing properties of language evolution can induce at least the masking effect, and hence lead to genetic degradation. The simulation demonstrates that the masking effect is indeed a part of the evolutionary process found in the normal Baldwin effect.

We have developed a new reinforcement learning technique called Bayesian-discrimination-function-based reinforcement learning (BRL). BRL is unique, in that it not only learns in the predefined state and action spaces, but also simultaneously changes their segmentation. BRL has proven to be more effective than other standard RL algorithms in dealing with multi-robot system (MRS) problems, where the learning environment is naturally dynamic. This paper introduces an extended form of BRL that improves its learning efficiency. Instead of generating a random action when a robot encounters an unknown situation, the extended BRL generates an action calculated by a linear interpolation among the rules with high similarity to the current sensory input. In both physical experiments and computer simulations, the extended BRL showed higher search efficiency than the standard BRL.

When bees and wasps leave the nest to forage, they perform orientation or learning flights. This behaviour includes a number of stereotyped flight manoeuvres mediating the active acquisition of visual information. If we assume that the bee is attempting to localise itself in the world with reference to stable visual landmarks, then we can model the orientation flight as a probabilistic Simultaneous Localisation And Mapping (SLAM) problem. Within this framework, one effect of stereotypical behaviour could be to make the agent’s own movements easier to predict. In turn, leading to better localisation and mapping performance. We describe a probabilistic framework for building quantitative models of orientation flights and investigate what benefits a more reliable movement model would have for an agent’s visual learning.

Both reactive and deliberative qualities are essential for a good action selection mechanism. We present a model that embodies a hybrid of two very different neural network architectures inside an animat: one that controls their high level deliberative behaviours, such as the selection of sub-goals, and one that provides reactive and navigational capabilities. Animats using this model are evolved in novel and dynamic environments, on complex tasks requiring deliberative behaviours: tasks that cannot be solved by reactive mechanisms alone and which would traditionally have their solutions formulated in terms of search-based planning. Significantly, no a priori information is given to the animats, making explicit forward search through state transitions impossible. The complexity of the problem means that animats must first learn to solve sub-goals without receiving any reward. Animats are shown increasingly complex versions of the task, with the results demonstrating, for the first time, incremental neuro-evolutionary learning on such tasks.

Using a simple simulation model of evolution and learning, this paper provides an evolutionary argument why Lamarckian inheritance - the direct transfer of lifetime learning from parent to offspring - may be so rare in nature. Lamarckian inheritance allows quicker genetic adaptation to new environmental conditions than non-lamarckian inheritance. While this may be an advantage in the short term, it may be detrimental in the long term, since the population may be less well prepared for future environmental changes than in the absence of Lamarckianism.

Artificial agents controlled by dynamic recurrent node networks with fixed weights are evolved to search for food and associate it with one of two different temperatures depending on experience. The task requires either instrumental or classical conditioned responses to be learned. The paper extends previous work in this area by requiring that a situated agent be capable of re-learning during its lifetime. We analyse the best-evolved agent’s behaviour and explain in some depth how it arises from the dynamics of the coupled agent-environment system.

In the context of situated and embodied cognition, we evaluate an information-theoretic approach to the construction of the of an artificial agent. We make the assumption that the construction of such a is an emergent property of the coupling between the agent and its environment where the goal of the agent is to maximize its control abilities. An information-theoretic approach of the perception-action loop allows us to evaluate the capacity of the agent to inject information into its environment and to later recapture this information in its own sensors. We define a construction mechanism based on an automaton that generates internal states relevant to the agent in terms of perception-action loop. Optimizing this automaton leads to internal representations that can be a basis for the construction of the of the agent. We illustrate the properties of the proposed mechanism in a simple example where an agent is acting in a box world. Simulation results show that this construction mechanism leads to a representation that captures important properties of the environment.

This paper describes a study in the use of digital evolution to produce cooperative communication behavior in a population of digital organisms. The results demonstrate that digital evolution can produce organisms capable of distributed problem solving through interactions between members of the population and their environment. Specifically, the organisms cooperate to distribute among the population the largest value sensed from the environment. These digital organisms have no “built-in” ability to perform this task; each population begins with a single organism that has only the ability to self-replicate. Over thousands of generations, random mutations and natural selection produce an instruction sequence that realizes this behavior, despite continuous turnover in the population.

In this paper we describe a model in which artificial evolution is employed to design neural mechanisms that control the motion of two autonomous robots required to communicate through sound to perform a common task. The results of this work are a “proof-of-concept”: they demonstrate that evolution can exploit a very simple sound communication system, to design the mechanisms that allow the robots cooperate by employing acoustic interactions. The analysis of the evolved strategies uncover the basic properties of the communication protocol.