Abstract:
In recent years it has been shown that first order recurrent neural networks trained by gradient-descent can learn not only regular but also simple context-free and context-sensitive languages. However, the success rate was generally low and severe instability issues were encountered. ... Comparative experiments with and without incremental learning indicated that incremental learning can accelerate and facilitate training. Furthermore, incrementally trained networks generally resulted in monotonic trajectories in hidden unit activation space, while the trajectories of non-incrementally trained networks were oscillating. The non-incrementally trained networks were more likely to generalise.

Abstract:
Loanword formation seems to provide a good test bed for the growing field of computational phonology, since it occurs in a more tightly controlled environment than other language processing tasks. We show how feedforward neural networks and decision trees can be trained to predict the phonological structure of English loanwords in Japanese, and compare the performance of the two paradigms. In each case the system produces a phonemic representation of the Japanese form, after receiving as input the phonological feature matrix of the current and surrounding phonemes. The performance is improved with the inclusion of information about the stress pattern, orthography of reduced vowels and location of word boundaries.

Abstract:
Recent work by Siegelmann has shown that the computational power of neural networks matches that of Turing Machines. Proofs are based on a fractal encoding of states to simulate the memory and operations of stacks.
In the present work, it is shown that similar stack-like dynamics can be learned in recurrent neural networks from simple sequence prediction tasks. Two main types of network solutions are found and described qualitatively as dynamical systems: damped oscillation and entangled spiraling around fixed points. The potential and limitations of each solution type are established in terms of generalization on two different context-free languages. Both solution types constitute novel stack implementations - generally in line with Siegelmann's theoretical work - which supply insights into how embedded structures of languages can be handled in analog hardware.

Abstract:
Following Tesauro's work on TD-Gammon, we used a 4000 parameter feed-forward neural network to develop a competitive backgammon evaluation function. Play proceeds by a roll of the dice, application of the network to all legal moves, and choosing the move with the highest evaluation. However, no back-propagation, reinforcement learning methods were employed. Instead we apply simple hill-climbing in a relative fitness environment. We start with an initial champion of all zero weights and proceed simply by playing the current champion network against a slightly mutated challenger and changing weights if the challenger wins. Surprisingly, this worked rather well. We investigate how the peculiar dynamics of this domain enabled a previously discarded weak method to succeed, by preventing suboptimal equilibria in a `meta-game' of self-learning.

Abstract:
There is growing evidence to suggest that the success of a co-evolutionary learning system may depend critically on the nature of the environment in which the learner is placed, and on certain attributes of the task domain, rather than the details of the particular learning algorithm employed. We discuss how a learning system can be modeled as a meta-level game between abstract entities which we call performer, infiltrator and evaluator. Learning can sometimes fail due to collusive suboptimal equilibria in this meta-game of learning. But some domains have special attributes which seem to prevent such collusions and thereby facilitate co-evolutionary advancement. A better understanding of these issues may help to improve the design of co-evolutionary learning systems in the future.

Abstract:
Pollack (1991) demonstrated that second-order recurrent neural networks can act as dynamical recognizers for formal languages when trained on positive and negative examples, and observed both phase transitions in learning and IFS-like fractal state sets. Follow-on work focused mainly on the extraction and minimization of a finite state automaton (FSA) from the trained network. However, such networks are capable of inducing languages which are not regular, and therefore not equivalent to any FSA. Indeed, it may be simpler for a small network to fit its training data by inducing such a non-regular language. But when is the network's language not regular? In this paper, using a low dimensional network capable of learning all the Tomita data sets, we present an empirical method for testing whether the language induced by the network is regular or not. We also provide a detailed &epsilon-machine analysis of trained networks for both regular and non-regular languages.

Abstract:
A framework for the study of path integrals on adelic spaces is developed, and it is shown that a family of path space measures on the localizations of an algebraic number field may, under certain conditions, be combined to form a global path space measure on its adele ring. An operator on the field of p-adic numbers analogous to the harmonic oscillator operator is then analyzed, and used to construct an Ornstein-Uhlenbeck process on the adelic ring of the rationals.

Abstract:
In this paper, we present a real-time algorithm that plans mostly optimal trajectories for multiple robots in a dynamic environment. This approach combines the use of a Delaunay triangulation to discretise the environment, and a novel cubic spline representation for a robot trajectory that meets the kinematic and dynamic constraints of the robot. We show that for complex environments the shortest-distance path is not always the shortest-time path due to these constraints. The algorithm has been implemented on real robots, and we present experimental results in cluttered environments.

Abstract:
We present a new paradigm extending the Iterated Prisoner's Dilemma to multiple players. Our model is unique in granting players information about past interactions between all pairs of players - allowing for much more sophisticated social behaviour. We provide an overview of preliminary results and discuss the implications in terms of the evolutionary dynamics of strategies.

Abstract:
This paper outlines our project to construct an architecture for implementing characters with believable personality in interactive software fiction. There is a need for a detailed model of personality to drive the actions of simulated character agents in areas such as simulation and the computer game industry. While there are a number of related projects in similar areas, this project's emphasis is on creating a versatile and efficient model, suitable for use on standard hardware such as a home personal computer. The paper also outlines the prototypes being developed and a brief description of the project's future goals.

Abstract:
We report simulations of Cohen, Dunbar and McClelland's (CDM) model of the Stroop effect showing how damace to the network can replicate empirical data on older individuals and those with dementia of the Alzheimer's type (DAT). The study is significant on three counts: firstly, the specific patterns of damage required are theoretically justifiable, leading to a deeper understanding of the role of connectionist models in understanding DAT; secondly, the simulations show that in specific neurodegenerative conditions, the CDM model can replicate behavioral effects, providing increased confidence in the model and countering recently published claims to the contrary; and thirdly, a new component of the model, the preliminary time constant, is identified as a major factor in the ability of our damaged model to successfully replicate empirical data, indicating a potential role for deficits in attention to task instructions.

Abstract:
As a test-bed for studying evolutionary and other machine learning techniques, we have developed a simulated hockey game called Shock in which players attempt to shoot a puck into their enemy's goal during a fixed time period. Multiple players may participate - one can be controlled by a human user, while the others are guided by artificial controllers. In previous work, we introduced the Shock environment and presented players that received global input (as if from an overhead camera) and were trained on a restricted task, using an evolutionary hill-climbing algorithm, with a staged learning approach. Here, we expand upon this work by developing players which instead receive input from local, Braitenberg-style sensors. These players are able to learn the task with fewer restrictions, using a simpler fitness measure based purely on whether or not a goal was scored. Moreover, they evolve to develop robust strategies for moving around the rink and scoring goals.

Abstract:
Robustness has long been recognised as a critical issue for co-evolutionary learning. It has been achieved in a number of cases, though usually in domains which involve some form of non-determinism. We examine a deterministic domain - a pseudo real-time two-player game called Tron - and evolve a neural network player using a simple hill-climbing algorithm. The results call into question the importance of determinism as a requirement for successful co-evolutionary learning, and provide a good opportunity to examine the relative importance of other factors.

Abstract:
A simulated hockey environment is introduced as a test bed for studying adaptive behavior and evolution of robot controllers. A near-frictionless playing surface is employed, partially mimicking zero gravity conditions. We show how a neural network using a simple evolutionary algorithm can develop nimble strategies for moving about the rink and scoring goals quickly and effectively.

Abstract:
We consider a new form of co-evolutionary learning in which human students and software tutors become partners in a cooperative learning process. While the students are learning from the tutors, the tutors will also be learning how to do their job more effectively through their interactions with the students. Earlier work on game-playing machine learners has brought to light important issues for co-evolutionary learning; in particular, that these interactions can be modeled as a meta-game between teachers and students, and that learning may fail due to suboptimal equilibria - for example, because of collusion between the individual players - in this meta-game of learning. We discuss some of the challenges that these issues may raise for the designers of intelligent software tutoring systems and describe a prototype Java applet that we have built in an effort to explore how these challenges may best be met.

Abstract:
A simple recurrent neural network is trained on a one-step look ahead prediction task for symbol sequences of the context-sensitive aⁿbⁿcⁿ language. Using an evolutionary hill climbing strategy for incremental learning the network learns to predict sequences of strings up to depth n = 12. Experiments and the algorithms used are described. The activation of the hidden units of the trained network is displayed in a 3-D graph and analysed.

Abstract:
We develop a new framework for studying the biases that recurrent networks bring to language processing tasks. A semantic concept represented by a point in Euclidean space is translated into a symbol sequence by an encoder network. This sequence is then fed into a decoder network which attempts to translate it back to the original concept. We show how a pair of recurrent networks acting as encoder and decoder can develop their own symbolic language that is serially transmitted between them, either forwards or backwards. The encoder and decoder bring different constraints to the task, and the conflicting nature of these constraints, reflected in the language that ultimately emerges, may provide important clues to the structure of human languages.

Abstract:
Loanword phonology seeks to model the process by which foreign words are `nativised' or incorporated into the phonological system of the borrowing language ... In this paper, we report the results obtained from a feed-forward neural network, trained on an 1100 word corpus of American English loanwords borrowed into Japanese...

Abstract:
Although recurrent neural networks can be trained to recognize formal languages from positive and negative examples, it is difficult for them to learn large and complex data sets due to the geometry of their state space, which tends to push activations into the corners and dilute the effect of differentials backpropagated through multiple levels of recurrence. We propose a new framework for dynamical recognizers in which the state space is the surface of an n-sphere and the maps are either orthogonal or quasi-orthogonal.

Abstract:
This study shows experimentally how first order recurrent neural networks learn to predict the mildly context-sensitive language MA₃ = {aⁿbⁿcⁿ; n >=1}. The training algorithm is evolutionary hill climbing with an incremental learning strategy. The experiments explore generalisation abilities of the networks emerging in the process of evolution. Generalization with respect to depth n of the language strings and regarding their ordering in the training sequence is evaluated. Experiments without an incremental learning strategy indicate that the latter accelerates or facilitates training. Interestingly networks trained incrementally and non-incrementally show a qualitatively different pattern of hidden unit activity. Training on the context-free aⁿbⁿ language and the context-sensitive aⁿbⁿcⁿ language is compared. The results show that even when the networks are identical and the data sets have the same size, learning of the context-free language is easier than learning of the context-sensitive language.

Abstract:
Modifications to Recursive Auto-Associative Memory are presented, which allow it to store deeper and more complex data structures than previously reported. These modifications include adding extra layers to the compressor and reconstructor networks, employing integer rather than real-valued representations, pre-conditioning the weights and pre-setting the representations to be compatible with them. The resulting system is tested on a data set of syntactic trees extracted from the Penn Treebank.