José Santos Reyes

Discrete-time backpropagation for training synaptic delay-based artificial neural networks

The aim of this paper is to endow a well-known structure for processing time-dependent information, synaptic delay-based ANNs, with a reliable and easy to implement algorithm suitable for training temporal decision processes. In fact, we extend the backpropagation algorithm to discrete-time feedforward networks that include adaptable internal time delays in the synapses. The structure of the network is similar to the one presented by [1], that is, in addition to the weights modeling the transmission capabilities of the synaptic connections, we model their length by means of a parameter that indicates the delay a discrete-event suffers when going from Zthe origin neuron to the target neuron through a synaptic connection. Like the weights, these delays are also trainable, and a training algorithm can be derived that is almost as simple as the backpropagation algorithm, and which is really an extension of it. We present examples of the application of these networks and algorithm to the prediction of time series and to the recognition of patterns in electrocardiographic signals. In the first case, we employ the temporal reasoning characteristics of these networks for the prediction of future values in a benchmark example of a time series: the one governed by the Mackey-Glass chaotic equation. In the second case, we provide a real life example. The problem consists in identifying different types of beats through two levels of temporal processing, one relating the morphological features which make up the beat in time and another one that relates the positions of beats in time, that is, considers rhythm characteristics of the ECG signal. In order to do this, the network receives the signal sequentially, no windowing, segmentation, or thresholding are applied.

Training Higher Order Gaussian Synapses

In this article we present an algorithm that permits training networks that include gaussian type higher order synapses. This algorithm is an extension of the classical backpropagation algorithm. Higher order synapses permit carrying out tasks using simpler networks than traditionally employed. The key to this simplicity is in the structure of the synapses: a gaussian with three trainable parameters. The fact that it is a function and consequently presents a variable output depending on its inputs and that it possesses more than one trainable parameter that allows it to implement non linear processing functions on its inputs, endows the networks with a large capacity for learning and generalization. We present two examples where these capacities are shown. The first one is a target tracking module for a the visual system of a real robot and the second one is an image classification system working on real images.

Cancer Stem Cell Modeling Using a Cellular Automaton

We used a cellular automaton model for cancer growth simulation at cellular level, based on the presence of different cancer hallmarks acquired by the cells. The rules of the cellular automaton determine cell mitotic and apoptotic behaviors, which are based on the acquisition of the hallmarks in the cells by means of mutations. The simulation tool allows the study of the emergent behavior of tumor growth. This work focuses on the simulation of the behavior of cancer stem cells to inspect their capability of regeneration of tumor growth in different scenarios.

Progressive Construction of Compound Behavior Controllers for Autonomous Robots Using Temporal Information

In this work we present a methodology for the progressive construction of compound behavior controllers for real autonomous robots. Some of these behaviors require temporal processing which is achieved through the inclusion of temporal delays in the synapses of the artificial neural networks used for their implementation. Starting from a set of simple behaviors implemented by means of evolved monolithic controllers, the evolution strategy employed obtains behaviors in higher levels either choosing the necessary low level behaviors from the previously selected set or through the coevolution of part of the low level behaviors and the higher level one. Emphasis is placed on making the behaviors robust and capable of performing in a real robot.

NETTOOL: A Hybrid Connectionist-Symbolic Development Environment

In this work we present an environment, NETTOOL, for the development of hybrid connectionist-symbolic systems, in which the connectionist representation is based on the same knowledge representation model as that of symbolic systems. The hybridation between the knowledge elements is local, the connectionist training algorithm is also localized in each element of the network so that the knowledge required for the learning process is a part of it. There is also the possibility of including capabilities for inferential level processing in the elements of the network.

Evolving Neural Controllers for Temporally Dependent Behaviors in Autonomous Robots

The objective of this work is to study neural control architectures for autonomous robots that explicitly handle time in tasks that require reasoning with the temporal component. The controllers are generated and trained through the methodology of evolutionary robotics. In this study, the reasoning processes are circumscribed to data provided by light sensors, as a first step in the process of evaluating the requirements of control structures that can be extended to the processing of visual information provided by cameras.

ECG Beat Classification with Synaptic Delay Based Artificial Neural Networks

In this work we present an application of Synaptic Delay Based Artificial Neural Networks to the classification of beats in ECG signal processing, both in terms of the ”shape” of the P-QRS-T complex and its position in time without any explicit windowing or thresholding process. The signal is simply introduced as it is to the network, sample by sample as time passes, and the network using internal delay terms modeling the length of the synaptic connections, learns to perform all the temporal reasoning processes required for the classification through the application of Discrete Time Backpropagation. We present an example of classification using real ECG patterns from the European ST-T Database.

Synaptic Modulation Based Artificial Neural Networks

This work introduces complex processing neural network topologies, based on the concept of modulating neuron, which induce higher order terms by means of the modulation of the synaptic weights. These structures present the advantages of being very easy to train, adapting easily to changing contexts and offer very good generalization capabilities along all the dimensions of the problems they are trained to solve. Finally, the function each modulation level or each module performs is very clear, making it simple to extend the model to multilevel hierarchies.

Parallelization of Connectionist Models Based on a Symbolic Formalism

In this paper we study the parallelization of the inference process for connectionist models. We use a symbolic formalism for the representation of the connectionist models. With this translation, the training mechanism is local in the elements of the network, the computing power is improved in the network nodes and a local hybridization with symbolic parts is achieved. The inference in the final knowledge network can be parallelized, whether the knowledge corresponds to a symbolic module, a connectionist model or a hybrid connectionist-symbolic module. Besides, the concurrency for knowledge networks corresponding to connectionist models is presented for the phases of processing and training. The parallelization is studied for a multiprocessor architecture with shared memory.

Connectionist Models for Syllabic Recognition in the Time Domain

In this work we study the possibility of identifying syllables in the temporal domain of the speech signal using connectionist models. We set out a distributed neural network for the separated recognition of vowels and consonants. The network will act as a speech to text translator performing the recognition at the minimum level neccesary for its use as input to a semantic system. We have studied the features that the network shows when the input is a temporal-based or a frecuency-based signal.