Dara Curran

Increasing Population Diversity Through Cultural Learning

A number of learning models are commonly employed in the simulation of social behavior. These include population learning, lifetime learning and cultural learning. Population learning allows popula tions as a whole to evolve over time, typically through a Darwinian model of natural selection. Lifetime learning allows individuals to acquire knowledge during their lifetimes and cultural learning allows indi viduals to pass this knowledge to their peers or subsequent generations. This work examines the effects of cultural learning on both the fitness and the diversity of a population of neural network agents. A population employing population learning alone and one employing both population and cultural learning are assigned three benchmark tasks: the 5-bit parity problem, the game of tic-tac-toe and the game of connect-four. Each agent contains a genome which encodes a neural network controller used by the agent to perceive and react to environmental stimuli. Results show that the addition of cultural learning promotes improved fitness and significantly increases both genotypic (the genetic make up of individuals) and phenotypic (the behavior of individuals) diversity in the population.

The Effects of Cultural Learning in Populations of Neural Networks

Population learning can be described as the iterative Darwinian process of fitness-based selection and genetic transfer of information leading to populations of higher fitness and is often simulated using genetic algorithms. Cultural learning describes the process of information transfer between individuals in a population through non-genetic means. Cultural learning has been simulated by combining genetic algorithms and neural networks using a teacher-pupil scenario where highly fit individuals are selected as teachers and instruct the next generation. By examining the innate fitness of a population (i.e., the fitness of the population measured before any cultural learning takes place), it is possible to examine the effects of cultural learning on the populations genetic makeup. Our model explores the effect of cultural learning on a population and employs three benchmark sequential decision tasks as the evolutionary task for the population: connect-four, tic-tac-toe, and blackjack. Experiments are conducted with populations employing population learning alone and populations combining population and cultural learning. The article presents results showing the gradual transfer of knowledge from genes to the cultural process, illustrated by the simultaneous decrease in the populations innate fitness and the increase of its acquired fitness measured after learning takes place.

Norms and cultural learning in the N-player prisoner’s dilemma

Social dilemmas are characterised by a choice between actions which are individually rational but collectively sub-optimal and actions which are better for the collective but leave individuals open to exploitation. Evolutionary game theory has been adopted to model the evolution of successive generations of agents playing a social dilemma game. In evolutionary simulations of N-player social dilemmas, cooperation rarely emerges. This paper investigates cultural evolution (via norms that are recorded as artefacts) as a means to increase the fitness of the society by allowing individual strategies to base their actions, not just on their genetic material, but also to take into consideration (by learning) evidence recorded as artefacts. In the first set of experiments, these norms are propagated vertically and we show that allowing cultural learning for a set of strategies in a small population can result in a stable and cooperative society. In the second set of preliminary experiments, agents are organised spatially according to a random graph and norms are spread horizontally.

Incremental evolution of local search heuristics

In evolutionary computation, incremental evolution refers to the process of employing an evolutionary environment that becomes increasingly complex over time. We present an implementation of this approach to develop randomised local search heuristics for constraint satisfaction problems, combining research on incremental evolution with local search heuristics evolution. A population of local search heuristics is evolved using a genetic programming framework on a simple problem for a short period and is then allowed to evolve on a more complex problem. Experiments compare the performance of this population with that of a randomly initialised population evolving directly on the more complex problem. The results obtained show that incremental evolution can represent a significant improvement in terms of optimisation speed, solution quality and solution structure.

Evolving cultural learning parameters in an NK fitness landscape

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. n nOur 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.

An evolutionary neural network approach to intrinsic plagiarism detection

Intrinsic Plagiarism Detection attempts to identify portions of a document which have been plagiarised without the use of reference collections. This is typically achieved by developing a classifier using support vector machines or hand-crafted neural networks. This paper presents an evolutionary neural network approach to the development of an intrinsic plagiarism detection classifier which is capable of evolving both the weights and structure of a neural network. The neural network is empirically tested on a corpus of documents and is shown to perform well.

The effects of lifetime learning on the diversity and fitness of populations

Evolutionary learning refers to the process whereby a population of organisms evolves, or learns, by genetic means through a Darwinian process of iterated selection and reproduction of fit individuals. Hinton and Nowlan employed a genetic algorithm to study the effects of lifetime learning on the performance of genetic evolution [1]. Each agent in the model possesses a genome, comprised of a string of characters which can be one of 1, 0 or ?. Each agent is allowed a number of rounds of lifetime learning where for each ? in the genotype they ‘guess’ its value, assigning it either a 1 or a 0. Experimental results showed that, once learning was applied, the population converged on the problem solution, showing that individual learning is capable of guiding genetic evolution.

Spatial N-player Dilemmas in Changing Environments

In recent work it has been shown that, given the presence of a community structure in a population of agents, cooperation can be a robust outcome for agents participating in an N-player social dilemma. Through the use of simple imitative learning, cooperation can spread and be the dominant robust behaviour. In this paper, we show that such cooperation can exist in the presence of noise and that persistent small levels of noise can allow the population to adapt suitably to dramatic changes in the environment.

An analysis of the effects of lifetime learning on population fitness and diversity in an NK fitness landscape

This paper examines the effects of lifetime learning on the diversity and fitness of a population. Our experiments measure the phenotypic diversity of populations evolving by purely genetic means (population learning) and of others employing both population learning and lifetime learning. The results obtained show, as in previous work, that the addition of lifetime learning results in higher levels of fitness than population learning alone. More significantly, results from the diversity measure show that lifetime learning is capable of sustaining higher levels of diversity than population learning alone.

Measuring diversity in populations employing cultural learning in dynamic environments

This paper examines the effect of cultural learning on a population of neural networks. We compare the genotypic and phenotypic diversity of populations employing only population learning and of populations using both population and cultural learning in two types of dynamic environment: one where a single change occurs and one where changes are more frequent. We show that cultural learning is capable of achieving higher fitness levels and maintains a higher level of genotypic and phenotypic diversity.