José María Font

Evolutionary construction and adaptation of intelligent systems

This paper introduces evolutionary techniques for automatically constructing intelligent self-adapting systems, capable of modifying their inner structure in order to learn from experience and self-adapt to a changing environment. These evolutionary techniques comprise an evolutionary system that is engineered by grammar-guided genetic programming, enabling the development of sub-symbolic and symbolic intelligent systems: artificial neural networks and knowledge-based systems, respectively. A context-free-grammar based codification system for artificial neural networks and rules, an initialization method and a crossover operator have been designed to properly balance the exploration and exploitation capabilities of the proposed system. This speeds up the convergence process and avoids trapping in local optima. This system has been applied to a medical domain: the detection of knee injuries from the analysis of isokinetic time series. The results of the evolved symbolic and sub-symbolic intelligent systems have been statistically compared with each other as part of a quantitative and qualitative performance analysis.

Grammar-guided evolutionary automatic system for autonomously building biological oscillators

This paper presents a grammar-guided evolutionary automatic system (GGEAS) that is capable of autonomously building special-purpose problem-solving programs. GGEAS uses a grammar-guided genetic programming (GGGP) core that generates solutions to a given problem from scratch, evolving them via selection, crossover and replacement to obtain the near-optimal solution to that problem. The GGGP core solves the closure problem and avoids code bloat. This core only outputs valid solutions and is able to freely determine their size and architecture. GGEAS is supplemented by three external modules that can be configured for any application domain: context-free grammar (CFG) generator, semantic checker and fitness module. The context-free grammar (CFG) generator creates the context-free grammar used by the GGEAS core to formalize the problem constraints. The semantic checker ensures the validity of the solutions created. Finally, the fitness module directs the population evolution towards an optimal solution to the problem. In order to test the effectiveness and the scope of the system, GGEAS has been applied to generate oscillatory biological programs codified in the BlenX language. The results show that GGEAS is effective at creating biological oscillators in silico from scratch without any prior knowledge about the solution and under a range of environmental conditions.

Constrained Level Generation Through Grammar-Based Evolutionary Algorithms

This paper introduces an evolutionary method for generating levels for adventure games, combining speed, guaranteed solvability of levels and authorial control. For this purpose, a new graph-based two-phase level encoding scheme is developed. This method encodes the structure of the level as well as its contents into two abstraction layers: the higher level defines an abstract representation of the game level and the distribution of its content among different inter-connected game zones. The lower level describes the content of each game zone as a set of graphs containing rooms, doors, monsters, keys and treasure chests. Using this representation, game worlds are encoded as individuals in an evolutionary algorithm and evolved according to an evaluation function meant to approximate the entertainment provided by the game level. The algorithm is implemented into a design tool that can be used by game designers to specify several constraints of the worlds to be generated. This tool could be used to facilitate the design of game levels, for example to make professional-level content production possible for non-experts.

Grammar-guided evolutionary construction of bayesian networks

This paper proposes the EvoBANE system. EvoBANE automatically generates Bayesian networks for solving special-purpose problems. EvoBANE evolves a population of individuals that codify Bayesian networks until it finds near optimal individual that solves a given classification problem. EvoBANE has the flexibility to modify the constraints that condition the solution search space, self-adapting to the specifications of the problem to be solved. The system extends the GGEAS architecture. GGEAS is a general-purpose grammar-guided evolutionary automatic system, whose modular structure favors its application to the automatic construction of intelligent systems. EvoBANE has been applied to two classification benchmark datasets belonging to different application domains, and statistically compared with a genetic algorithm performing the same tasks. Results show that the proposed system performed better, as it manages different complexity constraints in order to find the simplest solution that best solves every problem.

Using Evolved Fuzzy Neural Networks for Injury Detection from Isokinetic Curves

In this paper we propose an evolutionary fuzzy neural networks system for extracting knowledge from a set of time series containing medical information. The series represent isokinetic curves obtained from a group of patients exercising the knee joint on an isokinetic dynamometer. The system has two parts: i) it analyses the time series input in order generate a simplified model of an isokinetic curve; ii) it applies a grammar-guided genetic program to obtain a knowledge base represented by a fuzzy neural network. Once the knowledge base has been generated, the system is able to perform knee injuries detection. The results suggest that evolved fuzzy neural networks perform better than non-evolutionary approaches and have a high accuracy rate during both the training and testing phases. Additionally, they are robust, as the system is able to self-adapt to changes in the problem without human intervention.

Rule Evolving System for Knee Lesion Prognosis from Medical Isokinetic Curves

This paper proposes a system for applying data mining to a set of time series with medical information. The series represent an isokinetic curve that is obtained from a group of patients performing a knee exercise on an isokinetic machine. This system has two steps: the first one is to analyze the input time series in order to generate a simplified model of an isokinetic curve; the second step applies a grammar-guided genetic program including an evolutionary gradient operator and an entropy-based fitness function to obtain a set of rules for a knowledge-based system. This system performs medical prognosis for knee injury detection. The results achieved have been statistically compared to another evolutionary approach that generates fuzzy rule-based systems.

Partition based real-valued encoding scheme for evolutionary algorithms

Encoding feasible solutions is one of the most important aspects to be taken into account in the field of evolutionary computation in order to solve search or optimization problems. This paper proposes a new encoding scheme for real-coded evolutionary algorithms. It is called partition based encoding scheme, and satisfies two restrictions. Firstly, each of the components of a decoded vector that conforms a candidate solution to a problem at hand belongs to a predefined interval. Secondly, the sum of the components of each of these decoded vectors is always equal to a predefined constant. The proposed encoding scheme inherently guarantees these constraints for all the individuals that are generated within the evolution process as a consequence of applying the genetic operators. Partition based encoding scheme is successfully applied to learning conditional probability tables for a given discrete Bayesian network topology, where each row of the tables must exactly add up to one, and the components of each row belong to the interval [0,1] as they are probability values. The results given by the proposed encoding system for this learning problem is compared to a deterministic algorithm and another evolutionary approach. Better results are shown in terms of accuracy with respect to the former one, and accuracy and convergence speed with respect to the later one.

Study of the Evolution of the Underlying Social Network of Discussion and Generation of New Ideas Applying an Instructional and Learning Model

This paper presents empirical results that show how an instructional and learning model might influence the underlying social network of discussion and generation of new ideas and, therefore, knowledge building. This study has been conducted on higher education students taking the third-year program development models course unit, as part of an accredited degree in informatics engineering. A moderately constructivist model with a blended learning approach was implemented in this course unit over the last few years. This combination has improved the academic outcomes achieved by students. In order to analyse what caused this favourable effect, we have analysed the evolution of the underlying social network of generation and discussion of new ideas among students throughout the course unit. We found that some key relationships in this underlying social network change, which suggests that there is a positive impact on knowledge building, learning and, ultimately, student educational achievement.