Yoshihiko Hasegawa

A Bayesian Network Approach to Program Generation

Genetic programming (GP) is a powerful optimization algorithm that has been applied to a variety of problems. This algorithm can, however, suffer from problems arising from the fact that a crossover, which is a main genetic operator in GP, randomly selects crossover points, and so building blocks may be destroyed by the action of this operator. In recent years, evolutionary algorithms based on probabilistic techniques have been proposed in order to overcome this problem. In the present study, we propose a new program evolution algorithm employing a Bayesian network for generating new individuals. It employs a special chromosome called the expanded parse tree , which significantly reduces the size of the conditional probability table (CPT). Prior prototype tree-based approaches have been faced with the problem of huge CPTs, which not only require significant memory resources, but also many samples in order to construct the Bayesian network. By applying the present approach to three distinct computational experiments, the effectiveness of this new approach for dealing with deceptive problems is demonstrated.

Applied Genetic Programming and Machine Learning

Reflecting rapidly developing concepts and newly emerging paradigms in intelligent machines, this text is the first to integrate genetic programming and machine learning techniques to solve diverse real-world tasks.These tasks include financial data prediction, day-trading rule development; and bio-marker selection. Written by a leading authority, this text will teach readers how to use machine learning techniques, make learning operators that efficiently sample a search space, navigate the search process through the design of objective fitness functions, and examine the search performance of the evolutionary system. All source codes and GUIs are available for download from the authors website.

Latent Variable Model for Estimation of Distribution Algorithm Based on a Probabilistic Context-Free Grammar

Estimation of distribution algorithms are evolutionary algorithms using probabilistic techniques instead of traditional genetic operators. Recently, the application of probabilistic techniques to program and function evolution has received increasing attention, and this approach promises to provide a strong alternative to the traditional genetic programming techniques. Although a probabilistic context-free grammar (PCFG) is a widely used model for probabilistic program evolution, a conventional PCFG is not suitable for estimating interactions among nodes because of the context freedom assumption. In this paper, we have proposed a new evolutionary algorithm named programming with annotated grammar estimation based on a PCFG with latent annotations, which allows this context freedom assumption to be weakened. By applying the proposed algorithm to several computational problems, it is demonstrated that our approach is markedly more effective at estimating building blocks than prior approaches.

Classification of Gene Expression Data by Majority Voting Genetic Programming Classifier

Recently, genetic programming (GP) has been applied to the classification of gene expression data. In its typical implementation, using training data, a single rule or a single set of rules is evolved with GP, and then it is applied to test data to get generalized test accuracy. However, in most cases, the generalized test accuracy is not higher. In this paper, we propose a majority voting technique for prediction of the labels of test samples. Instead of a single rule or a single set of rules, we evolve multiple rules with GP and then apply those rules to test samples to determine their labels by using the majority voting technique. We demonstrate the effectiveness of our proposed method by performing different types of experiments on two microarray data sets.

Estimation of distribution algorithm based on probabilistic grammar with latent annotations

Genetic Programming (GP) which mimics the natural evolution to optimize functions and programs, has been applied to many problems. In recent years, evolutionary algorithms are seen from the viewpoint of the estimation of distribution. Many algorithms called EDAs (Estimation of Distribution Algorithms) based on probabilistic techniques have been proposed. Although probabilistic context free grammar (PCFG) is often used for the function and program evolution, it assumes the independence among the production rules. With this simple PCFG, it is not able to induce the building-blocks from promising solutions. We have proposed a new function evolution algorithm based on PCFG using latent annotations which weaken the independence assumption. Computational experiments on two subjects (the royal tree problem and the DMAX problem) demonstrate that our new approach is highly effective compared to prior approaches.

Optimal implementations for reliable circadian clocks.

Circadian rhythms are acquired through evolution to increase the chances for survival through synchronizing with the daylight cycle. Reliable synchronization is realized through two trade-off properties: regularity to keep time precisely, and entrainability to synchronize the internal time with daylight. We find by using a phase model with multiple inputs that achieving the maximal limit of regularity and entrainability entails many inherent features of the circadian mechanism. At the molecular level, we demonstrate the role sharing of two light inputs, phase advance and delay, as is well observed in mammals. At the behavioral level, the optimal phase-response curve inevitably contains a dead zone, a time during which light pulses neither advance nor delay the clock. We reproduce the results of phase-controlling experiments entrained by two types of periodic light pulses. Our results indicate that circadian clocks are designed optimally for reliable clockwork through evolution.

Circadian clocks optimally adapt to sunlight for reliable synchronization.

Circadian oscillation provides selection advantages through synchronization to the daylight cycle. However, a reliable clock must be designed through two conflicting properties: entrainability to synchronize internal time with periodic stimuli such as sunlight, and regularity to oscillate with a precise period. These two aspects do not easily coexist, because better entrainability favours higher sensitivity which may sacrifice regularity. To investigate conditions for satisfying the two properties, we analytically calculated the optimal phase–response curve with a variational method. Our results indicate an existence of a dead zone, i.e. a time period during which input stimuli neither advance nor delay the clock. A dead zone appears only when input stimuli obey the time course of actual solar radiation, but a simple sine curve cannot yield a dead zone. Our calculation demonstrates that every circadian clock with a dead zone is optimally adapted to the daylight cycle.

Enhanced entrainability of genetic oscillators by period mismatch

Biological oscillators coordinate individual cellular components so that they function coherently and collectively. They are typically composed of multiple feedback loops, and period mismatch is unavoidable in biological implementations. We investigated the advantageous effect of this period mismatch in terms of a synchronization response to external stimuli. Specifically, we considered two fundamental models of genetic circuits: smooth and relaxation oscillators. Using phase reduction and Floquet multipliers, we numerically analysed their entrainability under different coupling strengths and period ratios. We found that a period mismatch induces better entrainment in both types of oscillator; the enhancement occurs in the vicinity of the bifurcation on their limit cycles. In the smooth oscillator, the optimal period ratio for the enhancement coincides with the experimentally observed ratio, which suggests biological exploitation of the period mismatch. Although the origin of multiple feedback loops is often explained as a passive mechanism to ensure robustness against perturbation, we study the active benefits of the period mismatch, which include increasing the efficiency of the genetic oscillators. Our findings show a qualitatively different perspective for both the inherent advantages of multiple loops and their essentiality.

Escape Process and Stochastic Resonance Under Noise-Intensity Fluctuation

Abstract We study the effects of noise intensity fluctuations on the stationary and dynamical properties of an overdamped Langevin model with a bistable potential and external periodical driving force. We calculated the stationary distributions, mean-first passage time (MFPT) and the spectral amplification factor using a complete set expansion (CSE) technique. We found resonant activation (RA) and stochastic resonance (SR) phenomena in the system under investigation. Moreover, the strength of RA and SR phenomena exhibit non-monotonic behavior and their trade-off relation as a function of the squared variation coefficient of the noise intensity process. The reliability of CSE is verified with Monte Carlo simulations.

Bistable stochastic processes in the q-exponential family

Stochastic bistable systems whose stationary distributions belong to the q-exponential family are investigated using two approaches: (i) the Langevin model subjected to additive and quadratic multiplicative noise, and (ii) the superstatistical model. Previously, the bistable Langevin model has been analyzed under linear multiplicative noise, whereas this paper reports on quadratic multiplicative noise, which is more physically meaningful. The stationary distribution of the Langevin model under quadratic multiplicative noise, which agrees with that derived by the maximum Tsallis entropy method, is found to be qualitatively different from its counterpart under linear multiplicative noise. We also show that the stationary distribution of the superstatistical model is the same as that of the Langevin model, whereas their transient properties, described in terms of mean first passage times (MFPTs), are qualitatively different.