Catálogo de publicaciones - libros

We present the dynamical analysis of embodied RNNs evolved to control a cybernetic device that solves a tracking problem. From the Neurodynamics perspective, we analyze the networks with focus on a characterization of the attractors and attractor sequences, guiding the transients. Projections of these attractors to motor space help visualizing the shape of the attractors, thus pointing to the underpinnings of behavior. Among the different attractors found are fixed points, periodic and quasi-periodic attractors of different periods, as well as chaos. Further analysis of the attractors relates changes of shape, size and period to motor control. Interesting characteristic behaviors arise, such as chaotic transitory regimes and implicit mapping of environmental assymmetricities in the network’s response, (as for example attractor hops that implicitly code for gravity). We discuss autonomy, capacity and some issues relating to a possible theory of transients.

This paper propose a linearizing scheme based on wavelet networks to reduce nonlinear distortion introduced by a high power amplifier over 256QAM signals. Parameters of the proposed linearizer are estimated by using a hybrid algorithm, namely least square and gradient descent. Computer simulation results confirm that once the 256QAM signals are amplified at an input back off level of 0 dB, there is a reduction of 29 dB spectrum re-growth. In addition proposed linearizing scheme has a low complexity and fast convergence.

Behavior-based artificial systems, e.g. mobile robots, are frequently designed using (various degrees and levels of) biology as inspiration, but rarely modeled based on actual quantitative empirical data. This paper presents a data-driven behavior-based model of a simple biological organism, the hydra. Four constituent behaviors were implemented in a simulated animal, and the overall behavior organization was accomplished using a colony-style architecture (CSA). The results indicate that the CSA, using a priority-based behavioral hierarchy suggested in the literature, can be used to model behavioral properties like latency, activation threshold, habituation, and duration of the individual behaviors of the hydra. Limitations of this behavior-based approach are also discussed.

We have developed a software system that simulates chemotaxis-based cell aggregation in 2D. The model implemented within the system consists of such cell behaviors as chemical diffusion/detection, motility, proliferation, adhesion and life cycle stages. Each virtual cell detects the state of the environment, and responds to the environment based on a pre-defined “program” and its own internal state. Cells are discrete units that are located on a grid, exist in discrete states (e.g. active or dying) and perform discrete tasks (e.g. divide and attach), but they also contain and are affected by continuous quantities (e.g. chemoattractant concentrations, gradients, age and velocities). This paper provides an overview of our chemotaxis-based aggregation model and details the algorithms required to perform chemotaxis-based cell aggregation simulation. A number of biological studies are being conducted with the system. They include fine-tuning the model parameters to reproduce PC12 cell aggregation experiments and parametric studies that demonstrate the effect that the model’s components have on cell aggregation dynamics.

In this paper we propose an approach to agent modeling that follows a signal based metaphor where agents are modeled as dissipative structures and their cognitive structures are modeled as compositions of multiple energetic potentials. This uniform representational support is used to model both reactive and deliberative processes. To illustrate the descriptive adequacy of the model, two experimental cases are presented where reactive and deliberative processes are modeled based on the proposed approach.

The concept of hypercycles was proposed by M. Eigen and P. Schuster to study the origin-of-life problem. A hypercycle is a simple self-reproducing system modeling molecular evolution in the abiotic period. Typogenetics is a formal system of strings originally devised by D. Hofstadter to explain the connection between computation and molecular genetics. It was later established by H. Morris as a formal system to study artificial life. Evolutionary algorithms were used by Kvasnicka et al. to find a small hypercycle in typogenetics. We improve upon their algorithm and construct many hypercycles of large sizes. We also experimented with enzymes of different lengths and various mappings between enzymes and their functions.

This paper proposes a methodology to estimate the information content of protein structures by using: an alphabet of local microenvironments obtained from a set of protein domains with equivalent function, a modification of the physical complexity concept [1], the measures of mutual information ((;)) and conditional entropy ((|)) between sequence and structure. The kinase domain catalytic subunit was used as a specific example. Our results are in accord with the hypothesis that proteins are information gathering and using systems [3], and suggest that protein structure depends less on protein sequence than biologists have historically supposed.

We propose a theoretical framework for the design of multi-agent systems with the capacity to surprise a human observer. Reasons for the difficulty in creating simulations capable of perpetual innovation are addressed. Artificial life concepts and biological inspiration from different abstraction layers serve as base for a three part model. These parts are: world modelling with artificial chemistries, agent brains and population dynamics. The parts of the model are theoretically discussed and some academic examples provided. Possible applications to biological research in the problem of speciation are considered.

Virtual neurons are digitized representations of biological neurons, with an emphasis on their morphology. In previous research we presented a proof of principle of reconstructing virtual neuronal morphologies by means of Genetic L-Systems Programming (GLP) [13]. However, the results were limited due to a hard evolutionary search process and a minimalistic fitness function. In this work we analyzed the search process and optimized the GLP configuration to enhance the search process. In addition, we designed a neuron type-specific fitness function which provides an incremental assessment of the evolved structures. The results are significantly better and relevant issues are discussed.

Proteins, under native conditions, fold to specific 3D structures according to their 1D amino acid sequence, which in turn is defined by the genetic code. The specific shape of a folded protein is a strong indicator of its function in the cell. The mechanisms involved in protein folding are not well understood and predicting the final conformation of a folded protein from its amino acid sequence alone is not yet achievable despite extensive research efforts, both theoretical and experimental. The protein folding process may be viewed as an emergent phenomenon, a result of underlying physics controlling the interaction of amino acids with their local environment, leading to the complex global fold. In this spirit we present a model for investigating protein folding using open L-systems, local rewriting rules with environmental interaction.