Andrés Espinal

Design of Spiking Central Pattern Generators for Multiple Locomotion Gaits in Hexapod Robots by Christiansen Grammar Evolution.

This paper presents a method to design Spiking Central Pattern Generators (SCPGs) to achieve locomotion at different frequencies on legged robots. It is validated through embedding its designs into a Field-Programmable Gate Array (FPGA) and implemented on a real hexapod robot. The SCPGs are automatically designed by means of a Christiansen Grammar Evolution (CGE)-based methodology. The CGE performs a solution for the configuration (synaptic weights and connections) for each neuron in the SCPG. This is carried out through the indirect representation of candidate solutions that evolve to replicate a specific spike train according to a locomotion pattern (gait) by measuring the similarity between the spike trains and the SPIKE distance to lead the search to a correct configuration. By using this evolutionary approach, several SCPG design specifications can be explicitly added into the SPIKE distance-based fitness function, such as looking for Spiking Neural Networks (SNNs) with minimal connectivity or a Central Pattern Generator (CPG) able to generate different locomotion gaits only by changing the initial input stimuli. The SCPG designs have been successfully implemented on a Spartan 6 FPGA board and a real time validation on a 12 Degrees Of Freedom (DOFs) hexapod robot is presented.

Comparison of PSO and DE for Training Neural Networks

The use of computational resources required for Feed-Forward Artificial Neural Network (FFANN) training phase by means of classical techniques such as the back propagation learning rule can be prohibitive in some applications. A good training phase is needed for a high performance of a neural network. In searching for alternative methods for training phase of FFANN, some metaheuristic techniques have been used to do this task. This paper compares the performance of Particle Swarm Optimization (PSO) and Differential Evolution (DE) as training methods for FFANN under several well-known pattern recognition instances.

A FPGA-Based Neuromorphic Locomotion System for Multi-Legged Robots

The paper develops a neuromorphic system on a Spartan 6 field programmable gate array (FPGA) board to generate locomotion patterns (gaits) for three different legged robots (biped, quadruped, and hexapod). The neuromorphic system consists of a reconfigurable FPGA-based architecture for a 3G artificial neural network (spiking neural network), which acts as a Central Pattern Generator (CPG). The locomotion patterns, are then generated by the CPG through a general neural architecture, which parameters are offline estimated by means of grammatical evolution and Victor-Purpura distance-based fitness function. The neuromorphic system is fully validated on real biped, quadruped, and hexapod robots.

A SpiNNaker Application: Design, Implementation and Validation of SCPGs

In this paper, we present the numerical results of the implementation of a Spiking Central Pattern Generator (SCPG) on a SpiNNaker board. The SCPG is a network of current-based leaky integrate-and-fire (LIF) neurons, which generates periodic spike trains that correspond to different locomotion gaits (i.e. walk, trot, run). To generate such patterns, the SCPG has been configured with different topologies, and its parameters have been experimentally estimated. To validate our designs, we have implemented them on the SpiNNaker board using PyNN and we have embedded it on a hexapod robot. The system includes a Dynamic Vision Sensor system able to command a pattern to the robot depending on the frequency of the events fired. The more activity the DVS produces, the faster that the pattern that is commanded will be.

Comparing Metaheuristic Algorithms on the Training Process of Spiking Neural Networks

Spiking Neural Networks are considered as the third generation of Artificial Neural Networks. In these networks, spiking neurons receive/send the information by timing of events (spikes) instead by the spike rate; as their predecessors do. Spikeprop algorithm, based on gradient descent, was developed as learning rule for training SNNs to solve pattern recognition problems; however this algorithm trends to be trapped in local minima and has several limitations. For dealing with the supervised learning on Spiking Neural Networks without the drawbacks of Spikeprop, several metaheuristics such as: Evolutionary Strategy, Particle Swarm Optimization, have been used to tune the neural parameters. This work compares the performance and the impact of some metaheuristics used for training spiking neural networks.

Generating Bin Packing Heuristic Through Grammatical Evolution Based on Bee Swarm Optimization

In the recent years, Grammatical Evolution (GE) has been used as a representation of Genetic Programming (GP). GE can use a diversity of search strategies including Swarm Intelligence (SI). Bee Swarm Optimization (BSO) is part of SI and it tries to solve the main problems of the Particle Swarm Optimization (PSO): the premature convergence and the poor diversity. In this paper we propose using BSO as part of GE as strategies to generate heuristics that solve the Bin Packing Problem (BPP). A comparison between BSO, PSO, and BPP heuristics is performed through the nonparametric Friedman test. The main contribution of this paper is to propose a way to implement different algorithms as search strategy in GE. In this paper, it is proposed that the BSO obtains better results than the ones obtained by PSO, also there is a grammar proposed to generate online and offline heuristics to improve the heuristics generated by other grammars and humans.

Evolvability Metric Estimation by a Parallel Perceptron for On-Line Selection Hyper-Heuristics

Online hyper-heuristic selection is a novel and powerful approach to solving complex problems. This approach dynamically selects, based on the state of a given solution, the most promising operator (from a pool of operators) to continue the search process. The dynamic selection is usually based on the analysis of the latest applications of a given operator during actual execution, estimating the potential success of the operator at the current solution state. The estimation can be made by evolvability metrics. Calculating an evolvability metric is computationally expensive since it requires the generation and evaluation of a neighborhood of solutions. This paper aims to estimate the potential success of an operator for a given solution state by using a pre-trained neural network; known as a parallel perceptron. The proposal accelerates the online selection process, allowing us to achieve better performance than hyper-heuristic models, which directly use evolvability functions.

Evolutionary Indirect Design of Feed-Forward Spiking Neural Networks

The present paper proposes the automatic design of Feed-Forward Spiking Neural Networks by representing several inherent aspects of the neural architecture in a proposed Context-Free Grammar; which is evolved through an Evolutionary Strategy. In the indirect design, the power of the design and the capabilities of the designed neural network are strongly related with the complexity of the grammars. The neural networks designed with the proposed grammar are tested with two well-known benchmark datasets of pattern recognition. Finally, neural networks derived from the proposed grammar are compared with other generated by similar grammars which were designed for the same purposed, the neural network design.

Developing Architectures of Spiking Neural Networks by Using Grammatical Evolution Based on Evolutionary Strategy

The Artificial Neural Networks (ANNs) have been used for solving problems in many theoretical and practical areas. Advances on the field of ANNs have derived in Spiking Neural Networks (SNNs); which are considered as the third generation of ANNs. SNNs receive/send the information by timing of events (spikes) instead by the spike rate; as their predecessors do. Although SNNs are capable to solve some functions with fewer neurons than networks of previous generations, there aren’t rules to set the architecture of any kind of ANN for solving a specific task; usually the architecture is set empirically based on the designer’s experience and the neural network’s performance over the problem. Recently, metaheuristic algorithms are being implemented to optimize some aspect on ANNs such as weight, connections and even the architecture. This work proposes a generic framework for automatic construction of Fully-Connected Feed-Forward Spiking Neural Networks through an indirect representation by means of Grammatical Evolution (GE) based on Evolutionary Strategy (ES) algorithm. Two well-known benchmarks datasets of pattern recognition were used for testing the proposal of this paper.

Symbolic Regression by Means of Grammatical Evolution with Estimation Distribution Algorithms as Search Engine

Grammatical Evolution (GE) is a Grammar-based form of Genetic Programming (GP) and it has been used to evolve programs or rules. The GE uses a population of linear genotypic strings and it is transformed by mapping process, those string are evolved using a search engine like the Genetic Algorithm (GA), Differential Evolution (DE), Particle Swarm Optimization (PSO), among others. One of the big trouble of these algorithms is the parameter tuning. In this paper is proposed an Estimation Distribution Algorithm (EDA) as search engine using the Symbolic Regression as a benchmark, due to the few parameters used by the EDA. The results were compared against the obtained by DE as search engine using the Friedman nonparametric test.