Gradient boosting represents an effective approach for constructing ensembles. We demonstrate how genetic programming can take advantage of the method for a wide range of classification tasks. The resulting Gradient Boosted Programming approach assumes two phases. Phase 1 develops a diverse set of base learners (programs). Phase 2 applies a gradient boosting approach specific to the program representation. The resulting ensemble is additively constructed and a class probability distribution is learnt for each program. An extensive benchmarking study is conducted across 21 classification datasets that include requirements for operation under class imbalance, tens of classes, and feature identification. The proposed approach is significantly better under the 11 low cardinality classification tasks and generally identifies simpler models than other ensemble methods such as Random Forests and XGBoost.

Streaming classification tasks with shifting imbalanced data distributions imply that the process creating the data are non-stationary. Such a streaming context is particularly challenging because model building, at any point in time, can only be performed relative to a small (incomplete) sample of the data. We demonstrate that both crossover and ensemble learning are particularly important for competitive performance under two challenging benchmarks. One benchmark represents a Botnet detection task and the second reformulates the Forest cover type classification task for shift (as opposed to drift). Comparison with the contemporary Learn++.NSE streaming classifier indicates specific advantages and disadvantages as viewed from the perspective of null-bias and prequential performance metrics.

Botnet detection in Internet of Things (IoT) environments remain challenging due to evolving attack behaviors and limited generalizability of traditional detection methods. This paper introduces a lightweight, two-stage detection framework that combines a Random Forest classifier with a Botnet Validation Layer. The proposed system leverages a Drift-Aligned Learning Curve (DALC) to adapt to data drifts by incorporating Domain Name System (DNS) based analysis for botnet validation particularly in Command & Control (C2) activities. We utilize a feature set comprising of 20 flow level and 4 DNS level features. Evaluated on diverse datasets, the proposed system achieves 99.7% F1-Score during testing while demonstrating strong generalization.

The navigation of complex labyrinths under partially observable visual state is typically addressed using complex recurrent, convolutional learning architectures (i.e. deep reinforcement learning). Conversely, in this work, we show that navigation can be achieved through the emergent evolution of a simple Braitentberg-style vehicle. We demonstrate that the interaction between agent and labyrinth is sufficient to learn a complex navigation behaviour from simple heuristics. To do so, the approach of tangled program graphs is assumed in which programs cooperatively coevolve to develop a modular indexing scheme that employs < 2.5% of state space. We attribute this simplicity to several biases implicit in the representation, such as: (1) the use of pixel indexing as opposed to deploying a convolutional kernel or image processing operators, and; (2) extensive support for modularity in which behaviours are always decomposed into contexts and corresponding actions.

In this research, we investigate the impact of four prevalent types of attacks, namely Portscan, Slowloris, Synflood, and Vulnerability Scan, on nine distinct Internet of Things (IoT) devices. These attacks are very common on the IoT eco-systems because they often serve as precursors to more sophisticated attack vectors. By analyzing attack vector traffic characteristics and IoT device responses, we aim to shed light on IoT eco-system vulnerabilities. To achieve this, we utilize and evaluate two feature sets extracted from the network traffic metadata using a flow analyzer, avoiding deep packet inspection. The goal of this research is to evaluate the impact of traffic flow metadata for identifying attacks on IoT devices. We further analyze the two flow feature sets in terms of generalizability of machine learning based attack detection from one IoT network to another. Results show that while generalizability is possible, it also depends on several factors including the characteristics of Iot traffic.

While applying genetic programming to reinforcement learning tasks, little attempt is made to incorporate local state specific reward information. This reduces the sample efficiency of the approach when tasks are rich in rewards. In this work, we provide a proof of concept for using local rewards under grid world tasks to incrementally parameterize teams of programs. A planning step is then introduced to propagate Q-values through the team of programs. No interaction with the task is necessary to perform the planning step. Benchmarking is performed on$n\times n;n\in\{5,10,20,50\}$grid world navigation tasks to illustrate the utility of the approach.

Analysis of incoming packets in deployed systems is one of the main methods used for detection of anomalous behaviour. Techniques utilizing supervised learning subject to the need of retraining if the observed behaviour in the system changes over time. Unsupervised techniques mitigate this problem but are not always capable of real-time analysis. Real-time unsupervised techniques bring to the table both the adaptability to dynamic behaviour as well as the ability to detect and alert about anomalies in real-time. A recent state-of-the-art technique, MIDAS, shows real-time capabilities while being unsupervised, but recent works have showed that it still had some shortcomings regarding its performance over more specific datasets. An alternative method has been proposed, namely MIMC, that builds on the foundation set by MIDAS. In this paper it is shown that, for the datasets of interest, there is always a way to setup MIMC that yields a higher performance than MIDAS. Furthermore, a method for determining parameters for the technique is also presented, and it is shown that it improves the yielded performance even further in a majority of cases.

This paper proposes and explores new attribute correlations and combined effort of multiple instances of microcluster-based anomaly detection on port scans, distributed denial of service and botnet attacks. To this end, the proposed system for micro-clustering based anomaly detection is compared against the state-of-the-art technique on three different network datasets, namely CTU-IoT, CTU-13 and UNSW-NB15. Evaluations not only show the effectiveness and high performance of the proposed system on all three datasets but also demonstrate the generalizability of the newly proposed attribute correlations and combination strategies.

The collection of log messages regarding the operation of deployed services and application is an integral component to the forensic analysis for the identification and understanding of security incidents. Approaches for parsing and abstraction of such logs, despite widespread use and study, do not directly account for the individualities of the domain of information security. This, in return, limits their applicability on the field. In this work, we analyze the state-of-the-art log parsing and abstraction algorithms from the perspective of information security. First, we reproduce/replicate previous analysis of such algorithms from the literature. Then, we evaluate their ability for parsing and abstraction of log files for forensic analysis purposes. Our study demonstrates that while the state-of-the-art techniques are accurate in log parsing, improvements are necessary in terms of achieving a holistic view to aid in forensic analysis for the identification and understanding of security incidents.

The goal of the CIFAR-10 benchmark is recast from the perspective of discovering light-weight as well as accurate solutions. Specifically, the image data, on which CIFAR-10 is based, requires multiple practical issues to be addressed that are not often considered collectively when applying genetic programming to classification problems. Issues of particular interest include cardinality, multi-class classification and diversity maintenance. We demonstrate that diversity maintenance and cardinality can be approached simultaneously by adopting a data subset to compose pools of exemplars for lexicase selection. The issues of multi-class classification and solution simplicity are addressed by adopting the tangled program graph (TPG) approach to emergent modularity. In addition, the mutation operator is modified to ensure that class labels do not `die out' during evolution. The resulting benchmarking study demonstrates solutions that are significantly more accurate than AutoML while providing comparable accuracies with solutions from unsupervised feature discovery, i.e. 70% accuracy. However, unlike the latter TPG solutions are several orders of magnitude simpler.

Tangled program graphs (TPG) support emergent modularity by first identifying subsets of programs that can usefully coexist (a team/ graph node) and then identifying the circumstance under which to reference other teams (arc adaptation). Variation operators manipulate the content of teams and arcs. This introduces cycles into the TPG structures. Previously, this effect was eradicated at run time by marking nodes while evaluating TPG individuals. In this work, a new marking heuristic is introduced, that of arc (learner) marking. This means that nodes can be revisited, but not the same arcs. We investigate the impact of this through 18 titles from the Arcade Learning Environment. The performance and complexity of the policies appear to be similar, but with specific tasks (game titles) resulting in preferences for one scheme over another.

Modularity represents a recurring theme in the attempt to scale evolution to the design of complex systems. However, modularity rarely forms the central theme of an artificial approach to evolution. In this work, we report on progress with the recently proposed Tangled Program Graph (TPG) framework in which programs are modules. The combination of the TPG representation and its variation operators enable both teams of programs and graphs of teams of programs to appear in an emergent process. The original development of TPG was limited to tasks with, for the most part, complete information. This work details two recent approaches for scaling TPG to tasks that are dominated by partially observable sources of information using different formulations of indexed memory. One formulation emphasizes the incremental construction of memory, again as an emergent process, resulting in a distributed view of state. The second formulation assumes a single global instance of memory and develops it as a communication medium, thus a single global view of state. The resulting empirical evaluation demonstrates that TPG equipped with memory is able to solve multi-task recursive time-series forecasting problems and visual navigation tasks expressed in two levels of a commercial first-person shooter environment.

Malicious insider attacks represent one of the most damaging threats to networked systems of companies and government agencies. There is a unique set of challenges that come with insider threat detection in terms of hugely unbalanced data, limited ground truth, as well as behaviour drifts and shifts. This work proposes and evaluates a machine learning based system for user-centered insider threat detection. Using machine learning, analysis of data is performed on multiple levels of granularity under realistic conditions for identifying not only malicious behaviours, but also malicious insiders. Detailed analysis of popular insider threat scenarios with different performance measures are presented to facilitate the realistic estimation of system performance. Evaluation results show that the machine learning based detection system can learn from limited ground truth and detect new malicious insiders in unseen data with a high accuracy. Specifically, up to 85% of malicious insiders are detected at only 0.78% false positive rate. The system is also able to quickly detect the malicious behaviours, as low as 14 minutes after the first malicious action. Comprehensive result reporting allows the system to provide valuable insights to analysts in investigating insider threat cases.

The capacity of genetic programming (GP) to evolve a 'hero' character in the Dota 2 video game is investigated. A reinforcement learning context is assumed in which the only input is a 320-dimensional state vector and performance is expressed in terms of kills and net worth. Minimal assumptions are made to initialize the GP game playing agents - evolution from a tabula rasa starting point - implying that: 1) the instruction set is not task specific; 2) end of game performance feedback reflects quantitive properties a player experiences; 3) no attempt is made to impart game specific knowledge into GP, such as heuristics for improving navigation, minimizing partial observability, improving team work or prioritizing the protection of specific strategically important structures. In short, GP has to actively develop its own strategies for all aspects of the game. We are able to demonstrate competitive play with the built in game opponents assuming 1-on-1 competitions using the 'Shadow Fiend' hero. The single most important contributing factor to this result is the provision of external memory to GP. Without this, the resulting Dota 2 bots are not able to identify strategies that match those of the built-in game bot.

Insider threat detection represents a challenging problem to companies and organizations where malicious actions are performed by authorized users. This is a highly skewed data problem, where the huge class imbalance makes the adaptation of learning algorithms to the real world context very difficult. In this work, applications of genetic programming (GP) and stream active learning are evaluated for insider threat detection. Linear GP with lexicase/multi-objective selection is employed to address the problem under a stationary data assumption. Moreover, streaming GP is employed to address the problem under a non-stationary data assumption. Experiments conducted on a publicly available corporate data set show the capability of the approaches in dealing with extreme class imbalance, stream learning and adaptation to the real world context.

We propose a Genetic Programming (GP) framework to address high-dimensional Multi-Task Reinforcement Learning (MTRL) through emergent modularity. A bottom-up process is assumed in which multiple programs self-organize into collective decision-making entities, or teams, which then further develop into multi-team policy graphs, or Tangled Program Graphs (TPG). The framework learns to play three Atari video games simultaneously, producing a single control policy that matches or exceeds leading results from (game-specific) deep reinforcement learning in each game. More importantly, unlike the representation assumed for deep learning, TPG policies start simple and adaptively complexify through interaction with the task environment, resulting in agents that are exceedingly simple, operating in real-time without specialized hardware support such as GPUs.

Non-stationary network traffic, together with stealth occurrences of malicious behaviors, make analyzing network traffic challenging. In this research, a machine learning framework is used to incrementally learn the network behavior and adapt to the changes in the traffic. This framework works under two main constraints: 1) label budget, 2) class imbalance; which makes it suitable for real-world network scenarios. Evaluations are performed on a public dataset with multiple Botnet scenarios under 0.5% and 5% label budgets; only around 2.2% of traffic is Botnet. Our results demonstrate the significance of the proposed Stream Genetic Programming solution and a general robustness to factors such as long latencies between instances of the same Botnet.

Algorithms that learn through environmental interaction and delayed rewards, or reinforcement learning (RL), increasingly face the challenge of scaling to dynamic, high-dimensional, and partially observable environments. Significant attention is being paid to frameworks from deep learning, which scale to high-dimensional data by decomposing the task through multilayered neural networks. While effective, the representation is complex and computationally demanding. In this work, we propose a framework based on genetic programming which adaptively complexifies policies through interaction with the task. We make a direct comparison with several deep reinforcement learning frameworks in the challenging Atari video game environment as well as more traditional reinforcement learning frameworks based on a priori engineered features. Results indicate that the proposed approach matches the quality of deep learning while being a minimum of three orders of magnitude simpler with respect to model complexity. This results in real-time operation of the champion RL agent without recourse to specialized hardware support. Moreover, the approach is capable of evolving solutions to multiple game titles simultaneously with no additional computational cost. In this case, agent behaviours for an individual game as well as single agents capable of playing all games emerge from the same evolutionary run.

This paper demonstrates how code reuse allows genetic programming (GP) to discover strategies for difficult gaming scenarios while maintaining relatively low model complexity. Critical factors in the proposed approach are illustrated through an in-depth study in two challenging task domains: RoboCup soccer and Ms. Pac-Man. In RoboCup, we demonstrate how policies initially evolved for simple subtasks can be reused, with no additional training or transfer function, in order to improve learning in the complex half-field offense (HFO) task. We then show how the same approach to code reuse can be applied directly in Ms. Pac-Man. In the latter case, the use of task-agnostic diversity maintenance removes the need to explicitly identify suitable subtasks a priori. The resulting GP policies achieve state-of-the-art levels of play in HFO and surpass scores previously reported in the Ms. Pac-Man literature, while employing less domain knowledge during training. Moreover, the highly modular policies discovered by GP are shown to be significantly less complex than state-of-the-art solutions in both domains. Throughout this paper, we pay special attention to a pair of task-agnostic diversity maintenance techniques, and empirically demonstrate their importance to the development of strong policies.

The Atari 2600 video game console provides an environment for investigating the ability to build artificial agent behaviours for a variety of games using a common interface. Such a task has received attention for addressing issues such as: 1) operation directly from a high-dimensional game screen; and 2) partial observability of state. However, a general theme has been to assume a common machine learning algorithm, but completely retrain the model for each game title. Success in this respect implies that agent behaviours can be identified without hand crafting game specific attributes/actions. This work advances current state-of-the-art by evolving solutions to play multiple titles from the same run. We demonstrate that in evolving solutions to multiple game titles, agent behaviours for an individual game as well as single agents capable of playing all games emerge from the same evolutionary run. Moreover, the computational cost is no more than that used for building solutions for a single title. Finally, while generally matching the skill level of controllers from neuro-evolution/deep learning, the genetic programming solutions evolved here are several orders of magnitude simpler, resulting in real-time operation at a fraction of the cost.

Active learning algorithms attempt to interactively develop a subset of data from which fitness evaluation is performed. Moreover, the distribution of labeled content within the data subset may adapt over time as genetic programming (GP) individuals improve. The basic goal is therefore to identify the most meaningful subset of data to improve the current model. Under a streaming data context additional challenges exist relative to the non-streaming scenario: non-stationary processes, partial observability anytime operation. This means that it is not possible to guarantee that the content of the data subset even provides exemplars for each class that could appear in the stream (i.e., different classes appear/disappear at different parts of the stream). With this in mind, an investigation is performed into the impact of adopting different policies for controlling the development of data subset content. To do so, a generic framework is defined in terms of sampling and archiving policies. The resulting evaluation under several large multi-class datasets with class imbalance indicates that adopting random sampling with a biased archiving policy is sufficient for evolving GP classifiers that match or better the current state-of-the-art, particularly when detecting minor classes.

Scaling genetic programming to organize large complex combinations of programs remains an under investigated topic in general. This work revisits the issue by first demonstrating the respective contributions of coevolution and diversity maintenance. Competitive coevolution is employed to organize a task in such a way that the most informative training cases are retained. Cooperative coevolution helps discover modularity in the solutions discovered and, in this work, is fundamental to constructing complex structures of programs that still execute efficiently (the policy tree). The role of coevolution and diversity maintenance is first independently established under the task of discovering reinforcement learning policies for solving Rubik's Cubes scrambled with 5-twists. With this established, a combined approach is then adopted for building large organizations of code for representing policies that solve 5 to 8-twist combinations of the Cube. The resulting 'deep' policy tree organizes hundreds of programs to provide optimal solutions to tens of millions of test cube configurations.

This work reports on an approach to direct policy discovery (a form of reinforcement learning) using genetic programming (GP) for the 3 by 3 by 3 Rubik's Cube. Specifically, a synthesis of two approaches is proposed: 1) a previous group theoretic formulation is used to suggest a sequence of objectives for developing solutions to different stages of the overall task; and 2) a hierarchical formulation of GP policy search is utilized in which policies adapted for an earlier objective are explicitly transferred to aid the construction of policies for the next objective. The resulting hierarchical organization of policies explicitly demonstrates task decomposition and policy reuse. Algorithmically, the process makes use of a recursive call to a common approach for maintaining a diverse population of GP individuals and then learns how to reuse subsets of programs (policies) developed against the earlier objective. Other than the two objectives, we do not explicitly identify how to decompose the task or mark specific policies for reuse. Moreover, at the end of evolution we return a population solving 100% of 17,675,698 different initial Cubes for the two objectives currently in use.

The application of genetic programming (GP) to streaming data analysis appears, on the face of it, to be a less than obvious choice. If nothing else, the (perceived) computational cost of model building under GP would preclude its application to tasks with non-stationary properties. Conversely, there is a rich history of applying GP to various tasks associated with trading agent design for currency and stock markets. In this work, we investigate the utility of a coevolutionary framework originally proposed for trading agent design to the related streaming data task of predicting individual household electric power consumption. In addition, we address several benchmarking issues, such as effective preprocessing of stream data using a candlestick representation originally developed for financial market analysis, and quantification of performance using a novel `area under the curve' style metric for streaming data. The computational cost of evolving GP solutions is demonstrated to be suitable for real-time operation under this task and shown to provide classification performance competitive with current established methods for streaming data classification. Finally, we note that the individual household electric power consumption dataset is more flexible than the more widely used electricity utility prediction dataset, because it supports benchmarking at multiple temporal time scales.

Retracement is the tendency of markets to move between upper `resistance' and lower `support' price levels. Human traders frequently make use of visual tools to help identify these resistance and support levels so that they can by used in their trading decisions. These decision can be put into trading strategies composed of rules designed to mitigate losses after a trade is started, often called `stop loss' orders, or to take profit at a near optimal time, often called `take profit' orders. However, identifying such resistance and support levels is notoriously difficult given market volatility. Indeed, the levels need recalculating on a continuous basis, and only hold to an approximate degree. In this work we describe an approach for evolving buy-stay-sell currency trading rules using genetic programming. These rules are explicitly linked to technical indicators that incorporate features characterizing retracement. Benchmarking is then performed using the most recent three years of data from the EURUSD foreign exchange market with three different methods of identifying retracement based on moving average, pivot points and Fibonacci ratios. Investment strategies employing Fibonacci ratios and found to provide superior performance among the strategies examined.

Half-field Offense (HFO) is a sub-task of Robocup 2D Simulated Soccer. HFO is a challenging, multi-agent machine learning problem in which a team of offense players attempt to manoeuvre the ball past a defending team and around the goalie in order to score. The agent's sensors and actuators are noisy, making the problem highly stochastic and partially observable. These same real-world characteristics have made Keepaway soccer, which represents one sub-task of HFO, a popular testbed in the reinforcement learning and task-transfer literature in particular. We demonstrate how policies initially evolved for Keepaway can be reused within a symbiotic framework for coevolving policies in genetic programming (GP), with no additional training or transfer function, in order to improve learning in the HFO task. Moreover, the highly modular policies discovered by GP are shown to be significantly less complex than solutions based on traditional value-function optimization while achieving the same level of play in HFO.

Keepaway soccer is a challenging robot control task that has been widely used as a benchmark for evaluating multi-agent learning systems. The majority of research in this domain has been from the perspective of reinforcement learning (function approximation) and neuroevolution. One of the challenges under multi-agent tasks such as keepaway is to formulate effective mechanisms for diversity maintenance. Indeed the best results to date on this task utilize some form of neuroevolution with genotypic diversity. In this work, a symbiotic framework for evolving teams of programs is utilized with both genotypic and behavioural forms of diversity maintenance considered. Specific contributions of this work include a simple scheme for characterizing genotypic diversity under teams of programs and its comparison to behavioural formulations for diversity under the keepaway soccer task. Unlike previous research concerning diversity maintenance in genetic programming (GP), we are explicitly interested in solutions taking the form of teams of programs.

A framework for coevolving genetic programming teams with Pareto archiving is benchmarked under two representative tasks for non-stationary streaming environments. The specific interest lies in determining the relative contribution of diversity and aging heuristics to the maintenance of the Pareto archive. Pareto archiving, in turn, is responsible for targeting data (and therefore champion individuals) as appropriate for retention beyond the limiting scope of the sliding window interface to the data stream. Fitness sharing alone is considered most effective under a non-stationary stream characterized by continuous (incremental) changes. Fitness sharing with an aging heuristic acts as the preferred heuristic when the stream is characterized by non-stationary stepwise changes.

Most research into frameworks for evolving trading agents emphasize aspects associated with the evolution of technical indicators and decision trees / rules. One of the factors that drives the development of such frameworks is the non-stationary, streaming nature of the task. However, it is the heuristics used to interface the evolutionary framework to the streaming data which potentially have most impact on the quality of the resulting trading agents. We demonstrate that including a validation partition has a significant impact on determining the overall success of the trading agents. Moreover, rather than conduct evolution on a continuous basis, only retraining when changes in trading quality are detected also yields significant advantages. Neither of these heuristics are widely recognized by research in evolving trading agent frameworks, although both are relatively easy to add to current frameworks. Benchmarking over a 3 year period of the EURUSD foreign exchange supports these findings.

Run time libraries (RTL) in genetic programming (GP) represent a scenario in which individuals evolved under an earlier independent evolutionary run can be potentially incorporated into a following GP run. To date, schemes for exploiting the RTL metaphor have emphasized syntactic over behavioural approaches. Thus, instructions are added to the later run such that the previous code can be explicitly indexed. In this work we demonstrate how the RTL concept is naturally supported by adopting a symbiotic framework for coevolution. Under the Pinball reinforcement learning task, we demonstrate how the initial RTL can be coevolved within a simpler formulation of the task and then used as the basis for providing solutions to a more difficult target task under the same domain. The resulting solutions are stronger than an RTL as coevolved against the target task alone or symbiosis as evolved without support for RTL.

New emerging high-dimensional data sets have made traditional clustering algorithms increasingly inefficient. More sophisticated approaches are required to cope with the increasing dimensionality and cardinality of such data sets. Feature selection methods are proposed as a solution to deal with this problem, however they fail for data sets where the attribute support for different clusters is not the same. For this category of data sets subspace clustering algorithms have been introduced over the past decade. We approach this problem from the perspective of Genetic Algorithms by adopting a hierarchical data structure deployed in three stages. 1) a traditional clustering algorithm is applied independently to each attribute of the data set, thus defining a grid of potential 1-d cluster centroids. 2) representing multi-dimensional cluster centroids by indexing 1-d cluster centroids. 3) converting the problem of finding the best combination of cluster centroids into that of discrete optimization and applying a multi-objective evolutionary algorithm, which uses group fitness evaluation to give a fitness to a group of clusters, as defined by process 2. Synthetic data sets with different characteristics are generated as the ground truth to evaluate the resulting algorithm for Evolutionary Subspace Clustering (ESC) as well as benchmark against alternative subspace and full-space clustering algorithms. ESC returns competitive accuracy and while typically utilizing less attributes and scaling as attribute count increases.

Classification as applied to streaming data implies that only a small number of new training instances appear at each generation and are never explicitly reintroduced by the stream. Pareto competitive coevolution provides a potential framework for archiving useful training instances between generations under an archive of finite size. Such a coevolutionary framework is defined for the online evolution of classifiers under genetic programming. Benchmarking is performed under multi-class data sets with class imbalance and training partitions with between 1,000's to 100,000's of instances. The impact of enforcing different constraints for accessing the stream are investigated. The role of online adaptation is explicitly documented and tests made on the relative impact of label error on the quality of streaming classifier results.

Adopting a symbiotic model of evolution separates context for deploying an action from the action itself. Such a separation provides a mechanism for task decomposition in temporal sequence learning. Moreover, previously learned policies are taken to be synonymous with meta actions (actions that are themselves policies). Should solutions to the task not be forthcoming in an initial round of evolution, then solutions from the earlier round represent the 'meta' actions for a new round of evolution. This provides the basis for evolving policy trees. A benchmarking study is performed using the Acrobot handstand task. Solutions to date from reinforcement learning have not been able to approach the performance of those established 14 years ago using an A* search and a priori knowledge regarding the Acrobot energy equations. The proposed symbiotic approach is able to match and, for the first time, better these results. Moreover, unlike previous work, solutions are tested under a broad range of Acrobot initial conditions, with hierarchical solutions providing significantly better generalization performance.

Evolutionary methods for addressing the temporal sequence learning problem generally fall into policy search as opposed to value function optimization approaches. Various re cent results have made the claim that the policy search approach is at best inefficient at solving episodic 'goal seeking' tasks i.e., tasks under which the reward is limited to describing properties associated with a successful outcome have no qualification for degrees of failure. This work demonstrates that such a conclusion is due to a lack of diversity in the training scenarios. We therefore return to the Acrobot 'height' task domain originally used to demonstrate complete failure in evolutionary policy search. This time a very simple stochastic sampling heuristic for defining a population of training configurations is introduced. Benchmarking two recent evolutionary policy search algorithms - Neural Evolution of Augmented Topologies (NEAT) and Symbiotic Bid-Based (SBB) Genetic Programming - under this condition demonstrates solutions as effective as those returned by advanced value function methods. Moreover this is achieved while remaining within the evaluation limit imposed by the original study.

An important part of network management requires the accurate identification and classification of network traffic for decisions regarding bandwidth management, quality of service, and security. This work explores the use of a Multi-Objective Genetic Algorithm (MOGA) for both, feature selection and cluster count optimization, for an unsupervised machine learning technique, K-Means, applied to encrypted traffic identification. Specifically, a hierarchical K-Means algorithm is employed, comparing its performance to the MOGA with a non-hierarchical (flat) K-Means algorithm. The latter has already been benchmarked against common unsupervised techniques found in the literature, where results have favored the proposed MOGA. The purpose of this paper is to explore the gains, if any, obtained by increasing cluster purity in the proposed model by means of a second layer of clusters. In this work, SSH is chosen as an example of an encrypted application. However, nothing prevents the proposed model to work with other types of encrypted traffic, such as SSL or Skype. Results show that with the hierarchical MOGA, significant gains are observed in terms of the classification performance of the system.

In this work, we explore the state space of a network application protocol by employing genetic programming techniques. To this end, we target Simple Mail Transfer Protocol (SMTP), which is a well-known and open protocol on the Internet. In order to achieve our goal, we aim to evolve the payload such that solution individuals result in an email being sent successfully through the targeted server. The proposed system implements an archive paradigm where, upon completion of the evolutionary process, a collection (archive) of solutions are presented. Specifically, they can all achieve the goal, but each does so in a unique manner. This collection allows us to examine the state space of the application protocol, giving us the ability to verify that these variations are either intended by the protocol, or should be addressed for security reasons.

Intuitively population based algorithms such as genetic programming provide a natural environment for supporting solutions that learn to decompose the overall task between multiple individuals, or a team. This work presents a framework for evolving teams without recourse to prespecifying the number of cooperating individuals. To do so, each individual evolves a mapping to a distribution of outcomes that, following clustering, establishes the parameterization of a (Gaussian) local membership function. This gives individuals the opportunity to represent subsets of tasks, where the overall task is that of classification under the supervised learning domain. Thus, rather than each team member representing an entire class, individuals are free to identify unique subsets of the overall classification task. The framework is supported by techniques from evolutionary multiobjective optimization (EMO) and Pareto competitive coevolution. EMO establishes the basis for encouraging individuals to provide accurate yet nonoverlaping behaviors; whereas competitive coevolution provides the mechanism for scaling to potentially large unbalanced datasets. Benchmarking is performed against recent examples of nonlinear SVM classifiers over 12 UCI datasets with between 150 and 200,000 training instances. Solutions from the proposed coevolutionary multiobjective GP framework appear to provide a good balance between classification performance and model complexity, especially as the dataset instance count increases.

This work explores the use of clustering objectives in a Multi-Objective Genetic Algorithm (MOGA) for both, feature selection and cluster count optimization, under the application of flow based encrypted traffic identification. We first explore whether it is possible to achieve the performance of a gold standard model (i.e., classification objectives), using a MOGA based on clustering objectives. Then, we explore the performance gain (if it exists) of applying a logarithmic transformation to the data prior to running the MOGA. Results show that MOGA trained with clustering objectives can closely reproduce the behavior of a gold standard model, not only in terms of the selected features, but also in terms of the achieved detection rate and false positives rate, above 90% and less than 1% respectively. On the other hand, no gain was observed by applying logarithmic transformation to the data.

The ultimate goal of subspace clustering algorithms is to identify both the subset of attributes supporting a cluster and the location of the cluster in the subspace. In this work a generic evolutionary approach to bottom-up subspace clustering is proposed consisting of three steps. The first applies a non-evolutionary clustering algorithm attribute-wise to establish the lattice from which subspace clusters will be designed. In the second step a multi-objective Genetic Algorithm (MOGA) is used to evolve good candidate subspace clusters (CSC) through a combinatorial search w.r.t. the attribute-wise lattice from step 1. The third step then searches in the space of CSC from the population of the the first MOGA to find the best combination of subspace clusters, again under a MOGA formulation. Important properties of the approach are that a standard clustering algorithm is deployed in step one to build the initial lattice of attribute-wise clusters. This helps to decouple the computational expense of clustering using Evolutionary Computation, with the MOGA applied in steps 2 and 3 building clusters through a combinatorial search relative to the original lattice parameters. Benchmarking on data sets with tens to hundreds of attributes illustrates the feasibility of the approach.

Models of Genetic Programming (GP) frequently reflect a neo-Darwinian view to evolution in which inheritance is based on a process of gradual refinement and the resulting solutions take the form of single monolithic programs. Conversely, introducing an explicitly symbiotic model of inheritance makes a divide-and-conquer metaphor for problem decomposition central to evolution. Benchmarking gradualist versus symbiotic models of evolution under a common evolutionary framework illustrates that not only does symbiosis result in more accurate solutions, but the solutions are also much simpler in terms of instruction and attribute count over a wide range of classification problem domains.

Mimicry attacks have been the focus of detector research where the objective of the attacker is to generate multiple attacks satisfying the same generic exploit goals for a given vulnerability. In this work, multi-objective Genetic programming is used to establish a “black-box” approach to mimicry attack generation. No knowledge is made of internal data structures of the target anomaly detector, only the anomaly rate reported by the detector. Such a “black box” methodology enables a vulnerability testing approach where both open-source and commodity anomaly detection systems can be tested. The approach successfully identifies exploits when benchmarked over four detectors and four applications.

This work focuses on testing anomaly detectors from the perspective of a Multi-objective Evolutionary Exploit Generator (EEG). Such a framework provides users of anomaly detection systems two capabilities. Firstly, no knowledge of protected data structures need to be assumed (i.e. the detector is a black-box), where the time, knowledge and availability of tools to perform such an analysis might not be generally available. Secondly, the evolved exploits are then able to demonstrate weaknesses in the ensuing detector parameterization. Therefore, the system administrator can identify the suitable parameters for the effective operation of the detector. EEG is employed against two second generation anomaly detectors, namely pH and pH with schema mask, on four UNIX applications in order to perform a vulnerability assessment and make a comparison between the two detectors.

In this work, we investigate the ability of genetic programming techniques to evolve valid network packets, including all relevant header values, towards a specific goal. We see this as a first step in building a fuzzing system that can learn to adapt for vulnerability analysis. By developing a system that learns the packets that are required to be transmitted towards targets, using feedback from an external network source, we make a step towards having a system that can intelligently explore the capabilities of a given security system. In order to validate our system's capabilities we evolve a variety of port scan patterns while running the packets through an IDS, with the goal to minimizes the alarms raised during the scanning process. Results show that the system not only successfully evolves valid TCP packets, but also remains stealthy in its activity.

Benchmarking of a team based model of Genetic Programming demonstrates that the naturally embedded style of feature selection is usefully extended by the teaming metaphor to provide solutions in terms of exceptionally low attribute counts. To take this concept to its logical conclusion the teaming model must be able to build teams with a non-overlapping behavioral trait, from a single population. The Symbiotic Bid-Based (SBB) algorithm is demonstrated to fit this purpose under an evaluation utilizing data sets with 650 to 5,000 attributes. The resulting solutions are one to two orders simpler than solutions identified under the alternative embedded paradigms of C4.5 and MaxEnt.

Subspace clustering algorithms in their most general form attempt to describe data with clusters that are not constrained to index a common set of attributes. Previous evolutionary approaches to this problem have assumed a weaker model in which clusters are built in a common subset. Moreover, a filter method is generally assumed in which a classical clustering algorithm is employed in the inner loop. Needless to say, this presents a considerable computational overhead. In this work we recognize the utility of assuming a `bottom-up' approach to subspace clustering. Specifically, we apply a classical clustering algorithm to each attribute to establish 1-d clusters that are then indexed by a MOGA to design a population of subspace clusters. The ensuing search is entirely in terms of a combinatorial optimization problem, thus computationally very efficient. A final single objective GA is then applied to search the set of subspace clusters identified under the MOGA for the most suitable combination.

Bid-based Genetic Programming (GP) provides an elegant mechanism for facilitating cooperative problem decomposition without an a priori specification of the number of team members. This is in contrast to existing teaming approaches where individuals learn a direct input-output map (e.g., from exemplars to class labels), allowing the approach to scale to problems with multiple outcomes (classes), while at the same time providing a mechanism for choosing an outcome from those suggested by team members. This paper proposes a symbiotic relationship that continues to support the cooperative bid-based process for problem decomposition while making the credit assignment process much clearer. Specifically, team membership is defined by a team population indexing combinations of GP individuals in a separate team member population. A Pareto-based competitive coevolutionary component enables the approach to scale to large problems by evolving informative test points in a third population. The ensuing Symbiotic Bid-Based (SBB) model is evaluated on three large classification problems and compared to the XCS learning classifier system (LCS) formulation and to the support vector machine (SVM) implementation LIBSVM. On two of the three problems investigated the overall accuracy of the SBB classifiers was found to be competitive with the XCS and SVM results. At the same time, on all problems, the SBB classifiers were able to detect instances of all classes whereas the XCS and SVM models often ignored exemplars of minor classes. Moreover, this was achieved with a level of model complexity significantly lower than that identified by the SVM and XCS solutions.

A mimicry attack is a type of attack where the basic steps of a minimalist 'core' attack are used to design multiple attacks achieving the same objective from the same application. Research in mimicry attacks is valuable in determining and eliminating weaknesses of detectors. In this work, we provide a genetic programming based automated process for designing all components of a mimicry attack relative to the Stide detector under a vulnerable Traceroute application. Results indicate that the automatic process is able to generate mimicry attacks that reduce the alarm rate from ~65% of the original attack, to ~2.7%, effectively making the attack indistinguishable from normal behaviors

A bid-based approach for coevolving Genetic Programming classifiers is presented. The approach coevolves a population of learners thatdecompose the instance space by way of their aggregate bidding behaviour. To reduce computation overhead, a small, relevant, subsetof training exemplars is (competitively) coevolved alongside the learners. The approach solves multi-class problems using a single population and is evaluated on three large datasets. It is found tobe competitive, especially compared to classifier systems, whilesignificantly reducing the computation overhead associated withtraining.

The Probabilistic Adaptive Mapping Developmental Genetic Programming (PAM DGP) algorithm that cooperatively coevolves a population of adaptive mappings and associated genotypes is used to learn recursive solutions given a function set consisting of general (not implicitly recursive) machine-language instructions. PAM DGP using redundant encodings to model the evolution of the biological genetic code is found to more efficiently learn 2nd and 3rd order recursive Fibonacci functions than related developmental systems and traditional linear GP. PAM DGP using redundant encoding is also demonstrated to produce the semantically highest quality solutions for all three recursive functions considered (Factorial, 2nd and 3rd order Fibonacci). PAM DGP is then shown to have produced such solutions by evolving redundant mappings to select and emphasize appropriate subsets of the function set useful for producing the naturally recursive solutions.

One-class classification naturally only provides one class of exemplars on which to construct the classification model. In this work, multi-objective genetic programming (GP) allows the one-class learning problem to be decomposed by multiple GP classifiers, each attempting to identify only a subset of the target data to classify. In order for GP to identify appropriate subsets of the one-class data, artificial outclass data is generated in and around the provided inclass data. A local Gaussian wrapper is employed where this reinforces a novelty detection as opposed to a discrimination approach to classification. Furthermore, a hierarchical subset selection strategy is used to deal with the necessarily large number of generated outclass exemplars. The proposed approach is demonstrated on three one-class classification datasets and was found to be competitive with a one-class SVM classifier and a binary SVM classifier.

A novel approach to the classification of large and unbalanced multi-class data sets is presented where the widely acknowledged issues of scalability, solution transparency, and problem decomposition are addressed simultaneously within the context of the genetic programming (GP) paradigm. A cooperative coevolutionary training environment that employs multi-objective evaluation provides the basis for problem decomposition and reduced solution complexity, while scalability is achieved through a Pareto competitive coevolutionary framework, allowing the system to be applied to large data sets (tens or hundreds of thousands of exemplars) without recourse to hardware-specific speedups. Moreover, a key departure from the canonical GP approach to classification is utilized in which the output of GP is expressed in terms of a non-binary, local membership function (e.g. a Gaussian), where it is no longer necessary for an expression to represent an entire class. Decomposition is then achieved through reformulating the classification problem as one of cluster consistency, where an appropriate subset of the training patterns can be associated with each individual such that problems are solved by several specialist classifiers rather than by a single 'super' individual.

The growing Recurrent Self-Organizing Map (GRSOM) is embedded into a standard Self-Organizing Map (SOM) hierarchy. To do so, the KDD benchmark dataset from the International Knowledge Discovery and Data Mining Tools Competition is employed. This dataset consists of 500,000 training patterns and 41 features for each pattern. Unlike most of the previous methods, only 6 of the basic features are employed. The resulting model has a capability of detection (false positive) rate of 89.6% (5.66%), where this is as good as the data-mining approaches that uses all 41 features and twice as faster than a similar hierarchical SOM architecture.

The computational overhead of genetic programming (GP) may be directly addressed without recourse to hardware solutions using active learning algorithms based on the random or dynamic subset selection heuristics (RSS or DSS). This correspondence begins by presenting a family of hierarchical DSS algorithms: RSS-DSS, cascaded RSS-DSS, and the balanced block DSS algorithm, where the latter has not been previously introduced. Extensive benchmarking over four unbalanced real-world binary classification problems with 30000-500000 training exemplars demonstrates that both the cascade and balanced block algorithms are able to reduce the likelihood of degenerates while providing a significant improvement in classification accuracy relative to the original RSS-DSS algorithm. Moreover, comparison with GP trained without an active learning algorithm indicates that classification performance is not compromised, while training is completed in minutes as opposed to half a day.

In development genetic programming (DGP) approaches where the search space is divided into genotypes and phenotypes, a mapping (or "genetic code") is needed to connect the two spaces. This model has subsequently been extended so that mappings evolve, and recently an implementation was proposed that co-evolves a genotype population and a population of adaptive mappings. Here, the authors identify and investigate performance obstacles for this recent implementation. They then introduce a new probabilistic adaptive mapping DGP that avoids those performance problems and explores a greater search space of genotype-mapping combinations without significant computational expense. The algorithm is shown to be more robust and to outperform the comparison adaptive mapping algorithm on challenging settings of the chosen test benchmark.

Genetic Programming offers freedom in the definition of the cost function that is unparalleled among supervised learning algorithms. However, this freedom goes largely unexploited in previous work. Here, we revisit the design of fitness functions for genetic programming by explicitly considering the contribution of the wrapper and cost function. Within the context of supervised learning, as applied to classification problems, a clustering methodology is introduced using cost functions which encourage maximization of separation between in and out of class exemplars. Through a series of empirical investigations of the nature of these functions, we demonstrate that classifier performance is much more dependable than previously the case under the genetic programming paradigm.

In this work, we employed genetic programming to evolve a "white hat" attacker; that is to say, we evolve variants of an attack with the objective of providing better detectors. Assuming a generic buffer overflow exploit, we evolve variants of the generic attack, with the objective of evading detection by signature-based methods. To do so, we pay particular attention to the formulation of an appropriate fitness function and partnering instruction set. Moreover, by making use of the intron behavior inherent in the genetic programming paradigm, we are able to explicitly obfuscate the true intent of the code. All the resulting attacks defeat the widely used 'Snort' Intrusion Detection System.

The conversion and extension of the Incremental Pareto-Coevolution Archive algorithm (IPCA) into the domain of Genetic Programming classifier evolution is presented. In order to accomplish efficiency in regards to classifier evaluation on training data, the coevolutionary aspect of the IPCA algorithm is utilized to simultaneously evolve a subset of the training data that provides distinctions between candidate classifiers. The algorithm is compared in terms of classification "score" (equal weight to detection rate, and 1 - false positive rate), and run-time against a traditional GP classifier using the entirety of the training data for evaluation, and a GP classifier which performs Dynamic Subset Selection. The results indicate that the presented algorithm outperforms the subset selection algorithm in terms of classification score, and outperforms the traditional classifier while requiring roughly 1 430 of the wall-clock time.

A novel approach to classification is proposed in which a Pareto-based ranking of individuals is used to encourage multiple individuals to participate in the solution. To do so, the classification problem is re-expressed as a cluster consistency problem, thus allowing utilization of techniques from multi-objective optimization. Such a formulation enables classification problems to be automatically decomposed and solved by several specialist classifiers rather than by a single 'super' individual. In this paper, we demonstrate the proposed approach to two benchmark binary problems and recommend a natural extension to multi-class problems. Results indicate the general appropriateness of the approach.

The work presented in this paper is intended to test crucial system services against stack overflow vulnerabilities. The focus of the test is the user-accessible variables, that is to say, the inputs from the user as specified at the command line or in a configuration file. The tester is defined as a process for automatically generating a wide variety of user-accessible variables that result in malicious buffers (an exploit). In this work, the search for successful exploits is formulated as an optimization problem and solved using evolutionary computation. Moreover the resulting attacks are passed through the Snort misuse detection system to observe the detection (or not) of each exploit

The availability of modern imaging techniques such as confocal scanning laser tomography (CSLT) for capturing high-quality optic nerve images offer the potential for developing automatic and objective methods for supporting clinical decision-making in glaucoma. We present a hybrid approach that features the analysis of CSLT images using moment methods to derive abstract image defining features, and the use of these features to train classifiers for automatically distinguishing CSLT images of healthy and diseased optic nerves. As a first step, in this paper, we present investigations in feature subset selection methods for reducing the relatively large input space produced by the moment methods. Our results demonstrate that our methods discriminate between healthy and glaucomatous optic nerves based on shape information automatically derived from CSLT tomography images.

A cascaded model is introduced for mining large datasets using genetic programming without recourse to specialist hardware. Such an algorithm satisfies the seeming conflicting requirements of scalability and accuracy on large datasets by incrementally building GP classifiers through the use of a hierarchical dynamic subset selection algorithm. Models are built incrementally with each layer of the cascade receiving as input the original feature vector, plus the output from the previous layer(s). In order to encourage each layer to explicitly solve new aspects of the problem a combination of sum square error and niching is utilized. Thus, previous layers of the model are considered a niche, and the cost function is a shared error metric.

In this work the multi-class classification capabilities of genetic programming (GP) are explored in the context of a competitive co-evolutionary system, in which a population of GP classifiers is trained against an evolving population of trainers (exemplar selectors) with the goal of reducing GP training time for large multi-class classification problems. Moreover, the niche-enabling mechanisms established in the genetic algorithm (GA) literature, known as crowding and sharing, are implemented for the classifier population in order to provide multi-class solutions from a single population in the same trial. The results as presented in the paper indicate the appropriateness of the competitive co-evolutionary training approach under GP multi-class classification.

Turing complete Genetic Programming (GP) models introduce the concept of internal state, and therefore have the capacity for identifying interesting temporal properties. Surprisingly, there is little evidence of the application of such models to problems for prediction. An empirical evaluation is made of a simple recurrent linear GP model over standard prediction problems.

The tagging problem in natural language processing is to find a way to label every word in a text as a particular part of speech, e.g., proper noun. An effective way of solving this problem with high accuracy is the transformation-based or "Brill" tagger. In Brill's system, a number of transformation templates are specified a priori that are instantiated and ranked during a greedy search-based algorithm. This paper describes a variant of Brill's implementation that instead uses a genetic algorithm to generate the instantiated rules and provide an adaptive ranking. Based on tagging accuracy, the new system provides a better hybrid evolutionary computation solution to the part-of-speech (POS) problem than the previous attempt. Although not able to make up for the use of a priori knowledge utilized by Brill, the method appears to point the way for an improved solution to the tagging problem.

The mixtures of experts (MoE) model provides the basis for building modular neural network solutions. In this work we are interested in methods for decomposing the input before forwarding to the MoE architecture. By doing so we are able to define the number of experts from the data itself. Specific schemes are shown to be appropriate for regression and classification problems, where each appear to have different preferences.

The dynamic subset selection (DSS) active learning algorithm is generalized to include the case of unsupervised learning. To do so, training set partitioning, exemplar difficulty and age, and early stopping criteria are introduced into the self organizing feature map algorithm. The resulting model is able to build a hierarchical SOFM on a large (500,000 pattern) dataset in 3 hours. In comparison, the same architecture without active learning requires 33 hours to construct. No reduction in accuracy is recorded for the DSS SOFM model.

The hierarchical RSS-DSS algorithm is introduced for dynamically filtering large datasets based on the concepts of training pattern age and difficulty, while utilizing a data structure to facilitate the efficient use of memory hierarchies. Such a scheme provides the basis for training genetic programming (GP) on a data set of half a million patterns in 15 min. The method is generic, thus, not specific to a particular GP structure, computing platform, or application context. The method is demonstrated on the real-world KDD-99 intrusion detection data set, resulting in solutions competitive with those identified in the original KDD-99 competition, while only using a fraction of the original features. Parameters of the RSS-DSS algorithm are demonstrated to be effective over a wide range of values. An analysis of different cost functions indicates that hierarchical fitness functions provide the most effective solutions.

The cascade architecture for incremental learning is demonstrated within the context of genetic programming. Such a scheme provides the basis for building steadily more complex models until a desired degree of accuracy is reached. The architecture is demonstrated for several data mining datasets. Efficient training on standard computing platforms is retained using the RSS-DSS algorithm for stochastically sampling datasets in proportion to exemplar 'difficulty' and 'age'. Finally, the ensuing empirical study provides the basis for recommending the utility of sum square cost functions in the datasets considered.

In areas, such as planning, state space searches are often conducted to find solutions. Usually, the heuristic is derived from knowledge of the domain. In many cases the knowledge of a domain is limited or the domain is so complex that an effective heuristic cannot be formulated. As an alternative, machine-learning techniques such as neural networks may be used to derive the heuristic. The game of Freecell was selected as a suitable benchmark domain, in which "knowledge based heuristics" and "neural heuristics" were employed to find solutions for randomly generated games. An amalgamation of the two, in which the neural network developed a heuristic from several knowledge based heuristics, was also used. Of the neural derived heuristics, the best-case architecture did not employ the "knowledge based heuristics". Moreover, neural heuristics were not able to improve upon those defined a priori.

Intrusion Detection Systems are typically deployed for real time operation, but are limited to identifying attacks once initiated. In this work we instead investigate the potential for predicting an attack before it occurs. To do so, a two-stage process is employed with a classification stage following that of a predictor. Predictors are based on the SOM and classifier on an SVM. Training and test is conducted using the 'TCP' connection features from the DARPA KDD competition data set. In spite of the simplicity of the model, the system is able to provide false positive and false negative rates of 23.8% and 7.1% respectively for one step-ahead prediction.

An approach to network intrusion detection is investigated, based purely on a hierarchy of Self-Organizing Feature Maps. Our principle interest is to establish just how far such an approach can be taken in practice. To do so, the KDD benchmark dataset from the International Knowledge Discovery and Data Mining Tools Competition is employed. This supplies a connection-based description of a factitious computer network in which each connection is described in terms of 41 features. Unlike previous approaches, only 6 of the most basic features are employed. The resulting system is capable of detection (false positive) rates of 89% (4.6%), where this is at least as good as the alternative data-mining approaches that require all 41 features.

An improved version of the self-organization by lateral inhibition model (SOLI) has been applied to two synthetic data sets as well as the breast cancer and liver data sets, two well-known benchmark data sets. The methodology developed combines the use of various validity indices with the developed combines with the use of various validity indices with the SOLI model to discover the proper cluster structure within the data sets. In addition, the results explain why the breast cancer data set trends to be clustered so accurately while the liver data set trends to be so difficult to cluster accurately.

Although adaptive and heuristic approaches perform well under idealized conditions to the packet network routing problem, such algorithms are also dependent on global information that is not available under real-world conditions. This work benchmarks routing under local information conditions using the AntNet algorithm and makes recommendations regarding future approaches.

The application of an object-oriented (OO) methodology to the design of a platform for heterogeneous digital libraries with multiagent technologies is demonstrated. Emphasis is placed on maximizing the autonomy of the query processing activity. The Fusion OO paradigm is specifically employed as the basis for the design process due to the significance attributed to the development of object interfaces under static and dynamic conditions. Finally, characteristics of the proposed Domain Index Server (DIS) platform are contrasted with those of an alternative platform (the University of Michigan Digital Library, UMDL) by way of the respective Fusion descriptions. This identifies a different emphasis on the interface design between objects in the two platforms: the DIS system uses more dynamic links while the UMDL system focuses on permanent and constant links. Simulation of the two platforms provides performance data that demonstrates the higher capacity of the DIS scheme.

Page-based linear genetic programming (GP) is proposed in which individuals are described in terms of a number of pages. Pages are expressed in terms of a fixed number of instructions, which is constant for all individuals in the population. Pairwise crossover results in the swapping of single pages, and thus, individuals are of a fixed number of instructions. Head-to-head comparison with Tree-structured GP and block-based linear GP indicates that the page-based approach evolves succinct solutions without penalizing generalization ability.

This work details and compares the application of configurable and reconfigurable computing techniques to the estimation of binary 3/spl times/3 masks of 4-, 6- and 8-connectivity and logical operations of AND, OR and NOT. Specifically it is shown that the promise of reconfigurable as opposed to configurable computing provides a very efficient implementation of the morphological operator kernel.

Optimal control requires the definition of a control policy from the behaviour of a plant, without the luxury of a desired reference trajectory. In the case of this work the direct method of optimal adaptive control is taken where feedback from the environment has no sign or directional information. Moreover, the case of continuous valued as opposed to binary valued control actions is required. The proposed architecture demonstrates extensive use of hierarchical partitioning of the problem in order to decompose the task into a composition of subtasks. The significance of variance terms in the design of RBFs is emphasized, and the entire network demonstrated on benchmark nonlinear control tasks. In each case the emphasis is towards the location of robust solutions without recourse to any a priori information.

Genetic programming arguably represents the most general form of evolutionary computation. However, such generality is not without significant computational overheads. Particularly, the cost of evaluating the fitness of individuals in any form of evolutionary computation represents the single most significant computational bottleneck. A less widely acknowledged computational overhead in GP involves the implementation of the crossover operator. To this end a page-based definition of individuals is used to restrict crossover to equal length code fragments. Moreover, by using a register-machine context, the significance of a priori internal register external output definitions is emphasized.

A platform for performing multi-agent searches in heterogeneous digital libraries is proposed. This differs significantly from previous approaches by completely removing the concept of a centralized search engine. Specifically, the organization of information held on domain index servers is constrained to conform to a virtual tree representation based on facets and global keyword concept schema particular to the set of information providers associated with the domain of interest (e.g. preparatory intranet). Simulation studies are used to compare this platform against a digital library platform presently in use, which employs the traditional central server scheme. Improvements in terms of query service time and robustness are demonstrated.

This paper proposes and demonstrates a framework for Sigma-Pi networks such that the combinatorial increase in product terms is avoided. This is achieved by only implementing a subset of the possible product terms (sub-net Sigma-Pi). Application of a dynamic weight pruning algorithm enables redundant weights to be removed and replaced during the learning process, hence permitting access to a larger weight space than employed at network initialization. More than one learning rate is applied to ensure that the inclusion of higher order descriptors does not result in over description of the training set (memorization). The application of such a framework is tested using a problem requiring significant generalization ability. Performance of the resulting sub-net Sigma-Pi network is compared to that returned by optimal multi-layer perceptrons and general Sigma-Pi solutions.