Bernard Manderick

Evolvable Hardware and Its Applications to Pattern Recognition and Fault-Tolerant Systems

This paper describes Evolvable Hardware (EHW) and its applications to pattern recognition and fault-torelant systems. EHW can change its own hardware structure to adapt to the environment whenever environmental changes (including hardware malfunction) occur. EHW is implemented on a PLD(Programmable Logic Device)-like device whose architecture can be altered by re-programming the architecture bits. Through genetic algorithms, EHW finds the architecture bits which adapt best to the environment, and changes its hardware structure accordingly.

Learning causal bayesian networks from observations and experiments: a decision theoretic approach

We discuss a decision theoretic approach to learn causal Bayesian networks from observational data and experiments. We use the information of observational data to learn a completed partially directed acyclic graph using a structure learning technique and try to discover the directions of the remaining edges by means of experiment. We will show that our approach allows to learn a causal Bayesian network optimally with relation to a number of decision criteria. Our method allows the possibility to assign costs to each experiment and each measurement. We introduce an algorithm that allows to actively add results of experiments so that arcs can be directed during learning. A numerical example is given as demonstration of the techniques.

Identifiability of causal effects in a multi-agent causal model

This paper is a first step to extending Judea Pearls work on identification of causal effects to a multi-agent context. We introduce multi-agent causal models consisting of a collection of agents each having access to a non-disjoint subset of the variables constituting the domain. Every agent has a causal model, determined by nonexperimental data and an acyclic causal diagram over its variables. The algorithm under investigation in this paper, tests whether the assumptions made in a causal model are sufficient to calculate the effect of an intervention (i.e. whether the effect of an intervention is identifiable). It is a distributed algorithm with a minimum amount of inter-agent communication concerning solely shared variables and where the details of each local causal model are kept confidential.

Building a Genetic Programming Framework: The Added-Value of Design Patterns

A large body of public domain software exists which addresses standard implementations of the Genetic Programming paradigm. Nevertheless researchers are frequently confronted with the lack of flexibility and reusability of the tools when for instance one wants to alter the genotypes representation or the overall behavior of the evolutionary process. This paper addresses the construction of a object-oriented Genetic Programming framework using on design patterns to increase its flexibility and reusability.

Extended replicator dynamics as a key to reinforcement learning in multi-agent systems

Modeling learning agents in the context of Multi-agent Systems requires an adequate understanding of their dynamic behaviour. Evolutionary Game Theory provides a dynamics which describes how strategies evolve over time. Borgers et al. [1] and Tuyls et al. [11] have shown how classical Reinforcement Learning (RL) techniques such as Cross-learning and Q-learning relate to the Replicator Dynamics (RD). This provides a better understanding of the learning process. In this paper, we introduce an extension of the Replicator Dynamics from Evolutionary Game Theory. Based on this new dynamics, a Reinforcement Learning algorithm is developed that attains a stable Nash equilibrium for all types of games. Such an algorithm is lacking for the moment. This kind of dynamics opens an interesting perspective for introducing new Reinforcement Learning algorithms in multi-state games and Multi-Agent Systems.

BioLMiner and the BioCreative II.5 challenge

This paper proposes a prototype text mining system, BioLMiner (Biological Literature Miner). BioLMiner can automatically extract useful information from biological literature, like gene mentions, normalized gene mentions, interaction articles, protein-protein interaction pairs, etc. Figure ​Figure11 shows the overall system architecture of BioLMiner. In the future, we will automate all communication between the subsystems and plan to make BioLMiner available as open source software. Figure 1 Biological Literature Miner (BioLMiner) system architecture. The input data are the original articles from biological literature databases like MEDLINE [http://medline.cos.com/] or journals like FEBS letters [http://www.elsevier.com/locate/ febslet/]. The output data are the annotated articles together with the information extracted. Some existing gene and protein databases and biological resources are used as external background knowledge, like Entrez Gene [http://jura.wi.mit.edu/entrez_gene/], UniProt [http://www.uniprot.org], MINT [http://mint.bio.uniroma2.it], IntAct [http://www. ebi.ac.uk/intact] and BioThesaurus [http://pir.georgetown.edu/iprolink/biothesaurus] . The core components of BioLMiner are • the Gene Mention Recognizer (GMRer) • the Gene Normalizer (GNer) • the Interaction Article Classifier (IACer) • the Protein-Protein Interaction Pair Extractor (PPIEor) Two machine learning techniques are used to develop the four components, including Support Vector Machines (SVMs) [1] and Conditional Random Fields (CRFs) [2], to address classification and sequence labeling problems. For GMRer, a hybrid recognizer is developed based on one sequence labeling model using CRFs and two classification model using SVMs. For GNer, IACer and PPIEor, a binary classifier using SVMs is developed respectively. In order to achieve good performance, our main efforts focus on how to design methods to extract rich and informative features and to combine them effectively. These features fuse the information of the context in the article, domain specific knowledge, the analysis using natural language processing (NLP) tools or specific ones to the biological domain (Bio-NLP). A full description of BioLMiner can be found in [3,4]. BioLMiner participated in the interaction normalization task (INT) using GNer and interaction pair task (IPT) using PPIEor in the BioCreative II.5 challenge [5]. For the INT, the F β-1 measure was 0.289, which ranked second of the 10 participating teams for this task. For the IPT, the F β-1 measure was 0.252, which ranked first of the 9 participating teams for this task. The current state of the art performance is far from satisfactory, especially for the IPT. PPI pairs that appear in the figures or tables, span different sentences or interact with themselves cannot be handled well for the moment. More advanced techniques need to be exploited in the future, like anaphora resolution used for semantic analysis to detect the inter-sentence PPI pairs.

Substitution matrix based kernel functions for protein secondary structure prediction

Different approaches to using substitution matrices in kernel functions for protein secondary structure prediction (PSSP) with support vector machines are investigated. This work introduces a number of kernel functions that calculate inner products between amino acid sequences based on the entries of a substitution matrix (SM), i.e. a matrix that contains evolutionary information about the substitutability of the different amino acids that make up proteins. The starting point is always the same, i.e. a pseudo inner product (PI) between amino acid sequences making use of a SM. It is shown what conditions a SM should satisfy in order for the PI to make sense and subsequently it is shown how a substitution distance (SD) based on the PI can be defined. Next, different ways of using both the PI and the SD in kernel functions for support vector machine (SVM) learning are discussed. In a series of experiments the different kernel functions are compared with each other and with other kernel functions that do not make use of a SM. The results show that the information contained in a SM can have a positive influence on the PSSP results, provided that it is employed in the correct way.

Knowledge Gradient for Multi-objective Multi-armed Bandit Algorithms

We extend knowledge gradient (KG) policy for the multi-objective multi-armed bandit problems to efficiently explore the Pareto optimal arms. We consider two partial order relationships to order the mean vectors, i.e. Pareto and scalarized functions. Pareto KG finds the optimal arms using Pareto search, while the scalarizations-KG transform the multi-objectives arms into one-objective arm to find the optimal arms. To measure the performance of the proposed algorithms, we propose three regret measures. We compare the performance of knowledge gradient policy with UCB1 on a multi-objective multi-armed bandit problem, where KG outperforms UCB1.

An adaptive rule-based classifier for mining big biological data

The adaptive rule-based classifier is used to classify multi-class biological data.It applies random subspace and boosting approaches with ensemble of decision trees.Decision tree induction is used for evolving rules from the biological data.k-nearest-neighbor is used for removing ambiguity between the contradictory rules.The classifier is evaluated using 148 Exome data sets and 10 life sciences data sets. In this paper, we introduce a new adaptive rule-based classifier for multi-class classification of biological data, where several problems of classifying biological data are addressed: overfitting, noisy instances and class-imbalance data. It is well known that rules are interesting way for representing data in a human interpretable way. The proposed rule-based classifier combines the random subspace and boosting approaches with ensemble of decision trees to construct a set of classification rules without involving global optimisation. The classifier considers random subspace approach to avoid overfitting, boosting approach for classifying noisy instances and ensemble of decision trees to deal with class-imbalance problem. The classifier uses two popular classification techniques: decision tree and k-nearest-neighbor algorithms. Decision trees are used for evolving classification rules from the training data, while k-nearest-neighbor is used for analysing the misclassified instances and removing vagueness between the contradictory rules. It considers a series of k iterations to develop a set of classification rules from the training data and pays more attention to the misclassified instances in the next iteration by giving it a boosting flavour. This paper particularly focuses to come up with an optimal ensemble classifier that will help for improving the prediction accuracy of DNA variant identification and classification task. The performance of proposed classifier is tested with compared to well-approved existing machine learning and data mining algorithms on genomic data (148 Exome data sets) of Brugada syndrome and 10 real benchmark life sciences data sets from the UCI (University of California, Irvine) machine learning repository. The experimental results indicate that the proposed classifier has exemplary classification accuracy on different types of biological data. Overall, the proposed classifier offers good prediction accuracy to new DNA variants classification where noisy and misclassified variants are optimised to increase test performance.

An evolutionary game theoretic perspective on learning in multi-agent systems

In this paper we revise Reinforcement Learning and adaptiveness in Multi-Agent Systems from an Evolutionary Game Theoretic perspective. More precisely we show there is a triangular relation between the fields of Multi-Agent Systems, Reinforcement Learning and Evolutionary Game Theory. We illustrate how these new insights can contribute to a better understanding of learning in MAS and to new improved learning algorithms. All three fields are introduced in a self-contained manner. Each relation is discussed in detail with the necessary background information to understand it, along with major references to relevant work.