Search

Semantic segmentation consists of classifying each pixel of an image and constitutes an essential step towards scene recognition and understanding. Deep convolutional encoder–decoder neural networks now constitute state-of-the-art methods in the field of semantic segmentation. The problem of street scenes’ segmentation for automotive applications constitutes an important application field of such networks and introduces a set of imperative exigencies. Since the models need to be executed on self-driving vehicles to make fast decisions in response to a constantly changing environment, they are not only expected to operate reliably but also to process the input images rapidly. In this paper, we explore genetic programming (GP) as a meta-model that combines four different efficiency-oriented networks for the analysis of urban scenes. Notably, we present and examine two approaches. In the first approach, we represent solutions as GP trees that combine networks’ outputs such that each output class’s prediction is obtained through the same meta-model. In the second approach, we propose representing solutions as lists of GP trees, each designed to provide a unique meta-model for a given target class. The main objective is to develop efficient and accurate combination models that could be easily interpreted, therefore allowing gathering some hints on how to improve the existing networks. The experiments performed on the Cityscapes dataset of urban scene images with semantic pixel-wise annotations confirm the effectiveness of the proposed approach. Specifically, our best-performing models improve systems’ generalization ability by approximately 5% compared to traditional ensembles, 30% for the less performing state-of-the-art CNN and show competitive results with respect to state-of-the-art ensembles. Additionally, they are small in size, allow interpretability, and use fewer features due to GP’s automatic feature selection.

Flux Variability Analysis (FVA) is an important method to analyze the range of fluxes of a metabolic network. FVA consists in performing a large number of independent optimization problems, to obtain the maximum and minimum flux through each reaction in the network. Although several strategies to make the computation more efficient have been proposed, the computation time of an FVA can still be limiting. We present a two-step procedure to accelerate the FVA computational time that exploits the large presence within metabolic networks of sets of reactions that necessarily have an identical optimal flux value or only differ by a multiplication constant. The first step identifies such sets of reactions. The second step computes the maximum and minimum flux value for just one element of each of set, reducing the total number of optimization problems compared to the classical FVA. We show that, when applied to any metabolic network model included in the BiGG database, our FVA algorithm reduces the total number of optimization problems of about 35 %, and the computation time of FVA of about 30%.

In recent years, many algorithmic strategies have been developed to exploit single-cell mutational profiles generated via sequencing experiments of cancer samples and return reliable models of cancer evolution. Here, we introduce the COB-tree algorithm, which summarizes the solutions explored by state-of-the-art methods for clonal tree inference, to return a unique consensus optimum branching tree. The method proves to be highly effective in detecting pairwise temporal relations between genomic events, as demonstrated by extensive tests on simulated datasets. We also provide a new method to visualize and quantitatively inspect the solution space of the inference methods, via Principal Coordinate Analysis. Finally, the application of our method to a single-cell dataset of patient-derived melanoma xenografts shows significant differences between the COB-tree solution and the maximum likelihood ones.

Given the implicitly parallel nature of population-based heuristics, many contributions reporting on parallel and distributed models and implementations of these heuristics have appeared so far. They range from the most natural and simple ones, i.e. fitness-level embarrassingly parallel implementations (where, for instance, each candidate solution is treated as an independent agent and evaluated on a dedicated processor), to many more sophisticated variously interacting multi-population systems. In the last few years, researchers have dedicated a growing attention to Particle Swarm Optimization (PSO), a bio-inspired population based heuristic inspired by the behavior of flocks of birds and shoals of fish, given its extremely simple implementation and its high intrinsical parallelism. Several parallel and distributed models of PSO have been recently defined, showing interesting performances both on benchmarks and real-life applications. In this chapter we report on four parallel and distributed PSO methods that have recently been proposed. They consist in a genetic algorithm whose individuals are co-evolving swarms, an “island model”- based multi-swarm system, where swarms are independent and interact by means of particle migrations at regular time steps, and their respective variants enriched by adding a repulsive component to the particles. We show that the proposed repulsive multi-swarm system has a better optimization ability than all the other presented methods on a set of hand-tailored benchmarks and complex real-life applications.

We present a new reverse-engineering framework for gene regulatory network reconstruction. It works on temporal series of gene activation data and, using genetic programming, it extracts the activation functions of the different genes from those data. Successively, the gene regulatory network is reconstructed exploiting the automatic feature selection performed by genetic programming and its dynamics can be simulated using the previously extracted activation functions. The framework was tested on the well-known IRMA gene regulatory network, a simple network composed by five genes in the yeast Saccharomyces cerevisiae, defined in 2009 as a simplified biological model to benchmark reverse-engineering approaches. We show that the performances of the proposed framework on this benchmark network are encouraging

Generalization is an important issue in colorimetric characterization of devices. We propose a framework based on Genetic Algorithms to select training samples from large datasets. Even though the framework is general, and can be used in principle for any dataset, we use two well known datasets as case studies: training samples are selected from the Macbeth ColorCheckerDC dataset and the trained models are tested on the Kodak Q60 photographic standard dataset. The presented experimental results show that the proposed framework has better, or at least comparable, performances than a set of other computational methods defined so far for the same goal (Hardeberg, Cheung, CIC and Schettini). Even more importantly, the proposed framework has the ability to optimize the training samples and the characterizing polynomial's coefficients at the same time. © 2010 Springer-Verlag Berlin Heidelberg.

We present a framework based on genetic algorithms to automatically generate cellular automata rules under four different asynchronous update models (fixed random sweep, random new sweep, clock and independent random ordering). We consider four different rules with well known dynamics under synchronous update scheme. We try to reconstruct the same dynamics by means of a genetic algorithm using asynchronous update schemes. We show that in many cases it is impossible, by means of an asynchronous update scheme, to perfectly reconstruct these dynamics. Nevertheless, we show that the genetic algorithm finds the rules that more closely approximate the target behavior and the dynamics of the rules found by the genetic algorithm are rather similar to the target ones. In particular, we can always recognize a similar pattern and we can also identify some differences in small details. This paves the way to a deeper investigation on this track: does using asynchronous updates allow us to find more stable rules, i.e. rules that are less affected by noise, and thus do not overfit training data? This question remains open and answering it is one of the main goals of our current research.

Predicting the toxicity of new potential drugs is a fundamental step in the drug design process. Recent contributions have shown that, even though Genetic Programming is a promising method for this task, the problem of predicting the toxicity of molecular compounds is complex and difficult to solve. In particular, when executed for predicting drug toxicity, Genetic Programming undergoes the well-known phenomenon of bloat, i.e. the growth in code size during the evolutionary process without a corresponding improvement in fitness. We hypothesize that this might cause overfitting and thus prevent the method from discovering simpler and potentially more general solutions. For this reason, in this paper we investigate two recently defined variants of the operator equalization bloat control method for Genetic Programming. We show that these two methods are bloat free also when executed on this complex problem. Nevertheless, overfitting still remains an issue. Thus, contradicting the generalized idea that bloat and overfitting are strongly related, we argue that the two phenomena are independent from each other and that eliminating bloat does not necessarily eliminate overfitting. © 2009 Springer Berlin Heidelberg.

Among the variants of the basic Particle Swarm Optimization (PSO) algorithm as first proposed in 1995, EPSO (Evolutionary PSO), proposed by Miranda and Fonseca, seems to produce significant improvements. We analyze the effects of two modifications introduced in that work (adaptive parameter setting and selection based on an evolution strategies-like approach) separately, reporting results obtained on a set of multimodal benchmark functions, which show that they may have opposite and complementary effects. In particular, using only parameter adaptation when optimizing 'harder' functions yields better results than when both modifications are applied. We also propose a justification for this, based on recent analyses in which particle swarm optimizers are studied as dynamical systems. © 2008 Springer-Verlag Berlin Heidelberg.

We introduce the concept of "minimal" search algorithm for a set of functions to optimize. We investigate the structure of closed under permutation (c.u.p.) sets and we calculate the performance of an algorithm applied to them. We prove that each set of functions based on the distance to a given optimal solution, among which trap functions, onemax or the recently introduced onemix functions, and the NK-landscapes are not c.u.p. and thus the thesis of the sharpened No Free Lunch Theorem does not hold for them. Thus, it makes sense to look for a specific algorithm for those sets. Finally, we propose a method to build a "good" (although not necessarily minimal) search algorithm for a specific given set of problems. The algorithms produced with this technique show better average performance than a genetic algorithm executed on the same set of problems, which was expected given that those algorithms are problem-specific. Nevertheless, in general they cannot be applied for real-life problems, given their high computational complexity that we have been able to estimate. © 2008 Springer-Verlag Berlin Heidelberg.

In this paper we present an empirical study of the Negative Slope Coefficient (NSC) hardness statistic to characterize the difficulty of 3-SAT fitness landscapes for randomly generated problem instances. NSC correctly classifies problem instances with a low ratio of clauses to variables as easy, while instances with a ratio close to the critical point are classified as hard, as expected. Together with previous results on many different problems and fitness landscapes, the present results confirm that NSC is a useful and reliable indicator of problem difficulty. © 2008 Springer-Verlag Berlin Heidelberg.

We define a set of measures that capture some different aspects of neutrality in evolutionary algorithms fitness landscapes from a qualitative point of view. If considered all together, these measures offer a rather complete picture of the characteristics of fitness landscapes bound to neutrality and may be used as broad indicators of problem hardness. We compare the results returned by these measures with the ones of negative slope coefficient, a quantitative measure of problem hardness that has been recently defined and with success rate statistics on a well known genetic programming benchmark: the multiplexer problem. In order to efficaciously study the search space, we use a sampling technique that has recently been introduced and we show its suitability on this problem

Computational methods allowing reliable pharmacokinetics predictions for newly synthesized compounds are critically relevant for drug discovery and development. Here we present an empirical study focusing on various versions of Genetic Programming and other well known Machine Learning techniques to predict Median Oral Lethal Dose (LD50) and Plasma Protein Binding (%PPB) levels. Since these two parameters respectively characterize the harmful effects and the distribution into human body of a drug, their accurate prediction is essential for the selection of effective molecules. The obtained results confirm that Genetic Programming is a promising technique for predicting pharmacokinetics parameters, both from the point of view of the accurateness and of the generalization ability.

Model checking is a way of analysing programs and program-like structures to decide whether they satisfy a list of temporal logic statements describing desired behaviour. In this paper we apply this to the fitness checking stage in an evolution strategy for learning finite state machines. We give experimental results consisting of learning the control program for a vending machine.

To analyse various properties of the search process of genetic programming it is useful to quantify the distance between two individuals. Using operator-based distance measures can make this analysis more accurate and reliable than using distance measures which have no relationship with the genetic operators. This paper extends a recent definition of a distance measure based on subtree crossover for genetic programming. Empirical studies are presented that show the suitability of this measure to dynamically calculate the fitness distance correlation coefficient during the evolution, to construct a fitness sharing system for genetic programming and to measure genotypic diversity in the population. These experiments confirm the accuracy of the new measure and its consistency with the subtree crossover genetic operator

The paper presents an empirical study aiming at evaluating and comparing several Machine Learning (ML) classification techniques in the automatic recognition of known patterns. The main motivations of this work is to select best performing classification techniques where target classes are based on the occurrence of known patterns in configurations of a forest system modeled according to Cellular Automata. Best performing ML classifiers will be adopted for the study of ecosystem dynamics within an interdisciplinary research collaboration between computer scientists biologists and ecosystem managers (Cellular Automata For Forest Ecosystems - CAFFE project). One of the main aims of the CAFFE project is the development of an analysis method based on recognition in CA state configurations of spatial patterns whose interpretations are inspired by the Chinese Go game.

Negative slope coefficient has been recently introduced and empirically proven a suitable hardness indicator for some well known genetic programming benchmarks, such as the even parity problem, the binomial-3 and the artificial ant on the Santa Fe trail. Nevertheless, the original definition of this measure contains several limitations. This paper points out some of those limitations, presents a new and more relevant definition of the negative slope coefficient and empirically shows the suitability of this new definition as a hardness measure for some genetic programming benchmarks, including the multiplexer, the intertwined spirals problem and the royal trees.

We study in detail the fitness landscape of a difficult cellular automata computational task: the majority problem. Our results show why this problem landscape is so hard to search, and we quantify the large degree of neutrality found in various ways. We show that a particular subspace of the solution space, called the "Olympus", is where good solutions concentrate, and give measures to quantitatively characterize this subspace

This paper presents a study of fitness distance correlation and negative slope coefficient as measures of problem hardness for genetic programming. Advantages and drawbacks of both these measures are presented both from a theoretical and empirical point of view. Experiments have been performed on a set of well-known hand-tailored problems and “real-life-like” GP benchmarks

This paper explores distance measures based on genetic operators for genetic programming using tree structures. The consistency between genetic operators and distance measures is a crucial point for analytical measures of problem difficulty, such as fitness distance correlation, and for measures of population diversity, such as entropy or variance. The contribution of this paper is the exploration of possible definitions and approximations of operator-based edit distance measures. In particular, we focus on the subtree crossover operator. An empirical study is presented to illustrate the features of an operator-based distance. This paper makes progress toward improved algorithmic analysis by using appropriate measures of distance and similarity

This paper proposes the association of two approaches of GP which improve efficiency and reduce bloat. The first approach is to use a multi-population version of GP and the second one is to employ populations that can change size dynamically and adaptively. The latter approach consists in deleting or adding individuals in the population as a function of the current fitness and two other parameters. We test this approach on three well-known problems in GP, artificial ant, even parity 5 and one instance of the symbolic regression. We find that the combination of these two methods improves the quality of the individuals in the populations while keeping their size as small as possible and decreases the amount of resources required

This paper presents an investigation of genetic programming fitness landscapes. We propose a new indicator of problem hardness for tree-based genetic programming, called negative slope coefficient, based on the concept of fitness cloud. The negative slope coefficient is a predictive measure, i.e. it can be calculated without prior knowledge of the global optima. The fitness cloud is generated via a sampling of individuals obtained with the Metropolis-Hastings method. The reliability of the negative slope coefficient is tested on a set of well known and representative genetic programming benchmarks, comprising the binomial-3 problem, the even parity problem and the artificial ant on the Santa Fe trail

Experimentally, periodically released droplets in systems of widening pipes show clustering. This is surprising, as purely hydrodynamic interactions are repulsive so that agglomeration should be prevented. In the main part of this paper, we investigate the clustering of droplets under the influence of phenomenological hydrostatic forces and some hypothetical attraction. In two appendices, we explain why a direct numerical simulation for this system is rather more difficult (and probably not possible with current methods) than the “simple” geometry would suggest.

Within the scope of the European Horizon 2020 project ACDC – Artificial Cells with Distributed Cores to Decipher Protein Function, we aim at the development of a chemical compiler governing the three-dimensional arrangement of droplets, which are filled with various chemicals. Neighboring droplets form bilayers with pores which allow chemicals to move from one droplet to its neighbors. With an appropriate three-dimensional configuration of droplets, we can thus enable gradual biochemical reaction schemes for various purposes, e.g., for the production of macromolecules for pharmaceutical purposes. In this paper, we demonstrate with artificial chemistry simulations that the ACDC technology is excellently suitable to maximize the yield of desired reaction products or to minimize the relative output of unwanted side products.