Vision-based decision-making tasks encompass a wide range of applications, including safety-critical domains where trustworthiness is as key as performance. These tasks are often addressed using Deep Reinforcement Learning (DRL) techniques, based on Artificial Neural Networks (ANNs), to automate sequential decision making. However, the "black-box" nature of ANNs limits their applicability in these settings, where transparency and accountability are essential. To address this, various explanation methods have been proposed; however, they often fall short in fully elucidating the decision-making pipeline of ANNs, a critical aspect for ensuring reliability in safety-critical applications. To bridge this gap, we propose an approach based on Graph-based Genetic Programming (GGP) to generate transparent policies for vision-based control tasks. Our evolved policies are constrained in size and composed of simple and well-understood operational modules, enabling inherent interpretability. We evaluate our method on three Atari games, comparing explanations derived from common explainability techniques to those derived from interpreting the agent's true computational graph. We demonstrate that interpretable policies offer a more complete view of the decision process than explainability methods, enabling a full comprehension of competitive game-playing policies.

With new technologies and advances in biology, practitioners are faced with an ever-growing amount of data. Among this data, images are of particular interest thanks to the ability to segment and annotate directly on the image. These annotations are time-consuming, but indispensable for machine learning methods. In this paper, for the sake of a project on oral cancer detection, we developed an application that aims to enable easy-to-use and customizable annotation of data. It also has the advantage to consider multimodal annotation, to be able to read complex formats which are useful in the medical field such as MIRAX for example and to be compatible with graphic tablets. This software is completely free to use and the whole implementation, along with the ready-to-use executables are available on Github. It can be useful for many others data-driven projects that need accurate annotation on images especially while having time constraints for the annotator.

In this contribution, we explore Cartesian Genetic Programming for image analysis of biomedical data. Producing large quantities of human-labeled biomedical data is an expensive task. Here, we introduce a way for CGP to use a small amount of training data, without loss in performance. To define the size of the training data, we utilize an Active Learning method to direct the algorithm towards informative samples. We examine how sampling a small set of data from the CELLPOSE dataset affects the performance of CGP. We also study the effects of restarting CGP with Active Learning. We found that using several restarts can lead to a more diverse set of the highest-performing solutions with fewer active nodes while maintaining similar performance to standard CGP.

While interest in artificial neural networks (ANNs) has been renewed by the ubiquitous use of deep learning to solve high-dimensional problems, we are still far from general artificial intelligence. In this article, we address the problem of emergent cognitive capabilities and, more crucially, of their detection, by relying on co-evolving creatures with mutable morphology and neural structure. The former is implemented via both static and mobile structures whose shapes are controlled by cubic splines. The latter uses ESHyperNEAT to discover not only appropriate combinations of connections and weights but also to extrapolate hidden neuron distribution. The creatures integrate low-level perceptions (touch/pain proprioceptors, retina-based vision, frequency-based hearing) to inform their actions. By discovering a functional mapping between individual neurons and specific stimuli, we extract a high-level module-based abstraction of a creature's brain. This drastically simplifies the discovery of relationships between naturally occurring events and their neural implementation. Applying this methodology to creatures resulting from solitary and tag-team co-evolution showed remarkable dynamics such as range-finding and structured communication. Such discovery was made possible by the abstraction provided by the modular ANN which allowed groups of neurons to be viewed as functionally enclosed entities.

Cartesian Genetic Programming (CGP) has previously shown capabilities in image processing tasks by evolving programs with a function set specialized for computer vision. A similar approach can be applied to Atari playing. Programs are evolved using mixed type CGP with a function set suited for matrix operations, including image processing, but allowing for controller behavior to emerge. While the programs are relatively small, many controllers are competitive with state of the art methods for the Atari benchmark set and require less training time. By evaluating the programs of the best evolved individuals, simple but effective strategies can be found.

In nature, gene regulatory networks are a key mediator between the information stored in the DNA of living organisms (their genotype) and the structural and behavioral expression this finds in their bodies, surviving in the world (their phenotype). They integrate environmental signals, steer development, buffer stochasticity, and allow evolution to proceed. In engineering, modeling and implementations of artificial gene regulatory networks have been an expanding field of research and development over the past few decades. This review discusses the concept of gene regulation, describes the current state of the art in gene regulatory networks, including modeling and simulation, and reviews their use in artificial evolutionary settings. We provide evidence for the benefits of this concept in natural and the engineering domains.

Motivation: Segmental Duplications (SDs) are DNA fragments longer than 1 kbp, distributed within and between chromosomes and sharing more than 90% identity. Although they hold a significant role in genomic fluidity and adaptability, many key questions about their intrinsic characteristics and mutability remain unsolved due to the persistent difficulty of sequencing highly duplicated genomic regions. The recent development of long and linked-read NGS technologies will increase the need to search for SDs in genomes newly sequenced with these technics. The main limitation of SD analysis will soon be the availability of efficient detection software, to retrieve and compare SD genomic component between species or lineages. Results: In this paper, we present the open-source ASGART, 'A Segmental duplications Gathering And Refining Tool', developed to search for segmental duplications (SDs) in any assembled sequence. We have tested and benchmarked ASGART on five models organisms. Our results demonstrate ASGART's ability to extract SDs from any genome-wide sequence, regardless of genomic size or organizational complexity and quicker than any other software available. Availability and implementation: The online version of ASGART is available at http://asgart.irit.fr. The source code of ASGART is available both on the ASGART website and at https://github.com/delehef/asgart. Supplementary information: Supplementary data are available at Bioinformatics online.

Genetic Regulatory Networks (GRNs) implementations have a high degree of variability in their details. Parameters, encoding methods, and dynamics formulas all differ in the literature, and some GRN implementations have a high degree of model complexity. In this paper, we present a comparative study of different implementations of a GRN and introduce new variants for comparison. We use a modified Genetic Algorithm (GA) to evaluate GRN performance on a number of common benchmark tasks, with a focus on real-time control problems. We propose an encoding scheme and set of dynamics equations that simplifies implementation and evaluate the evolutionary fitness of this proposed method. Lastly, we use the comparative modifications study to demonstrate overall enhancements for GRN models.

International audience

In this paper, we use a gene regulatory network (GRN) to regulate a reinforcement learning controller, the State- Action-Reward-State-Action (SARSA) algorithm. The GRN serves as a neuromodulator of SARSA's learning parame- ters: learning rate, discount factor, and memory depth. We have optimized GRNs with an evolutionary algorithm to regulate these parameters on specific problems but with no knowledge of problem structure. We show that genetically- regulated neuromodulation (GRNM) performs comparably or better than SARSA with fixed parameters. We then ex- tend the GRNM SARSA algorithm to multi-task problem generalization, and show that GRNs optimized on multi- ple problem domains can generalize to previously unknown problems with no further optimization.

Artificial gene regulatory networks (GRNs) are biologically inspired dynamical systems used to control various kinds of agents, from the cells in developmental models to embodied robot swarms. Most recent work uses a genetic algorithm (GA) or an evolution strategy in order to optimize the network for a specific task. However, the empirical performances of these algorithms are unsatisfactory. This paper presents an algorithm that primarily exploits a network distance metric, which allows genetic similarity to be used for speciation and variation of GRNs. This algorithm, inspired by the successful neuroevolution of augmenting topologies algorithm's use in evolving neural networks and compositional pattern-producing networks, is based on a specific initialization method, a crossover operator based on gene alignment, and speciation based upon GRN structures. We demonstrate the effectiveness of this new algorithm by comparing our approach both to a standard GA and to evolutionary programming on four different experiments from three distinct problem domains, where the proposed algorithm excels on all experiments.

This paper presents a new model for the development of artificial creatures from a single cell. The model aims at providing a more biologically plausible abstraction of the morphogenesis and the specialization process, which the organogenesis follows. It is built upon three main elements: a cellular physics system that simulates division and intercellular adhesion dynamics, a simplified cell cycle offering to the cells the possibility to select actions such as division, quiescence, differentiation or apoptosis and, finally, a cell specialization mechanism quantifying the ability to perform different functions. An evolved artificial gene regulatory network is employed as a cell controller. As a proof-of-concept, we present two experiments where the morphology of a multicellular organism is guided by cell weaknesses and efficiency at performing different functions under environmental stress.

We present DevoII, an improved cell-based developmental model for wind farm layout optimization. To address the shortcomings of discretization, DevoII's gene regulatory networks control cells that act in a continuous rather than discretized grid space. We find that DevoII is competitive, and in some cases, superior with respect to state-of-the-art global, stochastic search approaches when a suite of algorithms is evaluated on different wind scenarios. The modularity of the genetic regulatory network computational paradigm in terms of isolating its search algorithm, the regulatory network simulation and the cell simulation, allowed this improvement to largely focus upon cell simulation. This indicates a robustness property of the paradigm's design. As well, wflo highlights how developmental models can be considered more efficient than other optimization methods because of their "optimize once, use-many" adaptability.

All multicellular living beings are created from a single cell. A developmental process, called embryogenesis, takes this first fertilized cell down a complex path of reproduction, migration, and specialization into a complex organism adapted to its environment. In most cases, the first steps of the embryogenesis take place in a protected environment such as in an egg or in utero. Starting from this observation, we propose a new approach to the generation of real robots, strongly inspired by living systems. Our robots are composed of tens of specialized cells, grown from a single cell using a bio-inspired virtual developmental process. Virtual cells, controlled by gene regulatory networks, divide, migrate, and specialize to produce the robot's body plan (morphology), and then the robot is manually built from this plan. Because the robot is as easy to assemble as Lego, the building process could be easily automated.

Optimizing a wind farm layout is a very complex problem that involves many local and global constraints such as inter-turbine wind interference or terrain peculiarities. Existing methods are either inefficient or, when efficient, take days or weeks to execute. Solutions are contextually sensitive to the specific values of the problem variables; when one value is modified, the algorithm has to be re-run from scratch. This paper proposes the use of a developmental model to generate farm layouts. Controlled by a gene regulatory network, virtual cells have to populate a simulated environment that represents the wind farm. When the cells' behavior is learned, this approach has the advantage that it is re-usable in different contexts; the same initial cell is responsive to a variety of environments and the layout generation takes few minutes instead of days.

Metamorphic robots are robots that can change their shape by reorganizing the connectivity of their modules to adapt to new environments, perform new tasks, or recover from damages. In this paper we present a decentralized method for structural evolving of a class of lattice-based simulated metamorphic robots in a static environment. These robots are considered as a set of crystalline (compressible) modules that are able to connect or disconnect one from each another or even exchange information and energy with the neighbor modules in order to form various structures/patterns dynamically. Our approach is spited in two layers: in the first layer a genetic algorithm is used to generate a number of well suited target configurations based on current information perceived from environment, while in the second layer a PacMan-like algorithm is used to make a plan for modules movement to transform the robot from its current pattern to the target pattern emerged in first layer.

An important connection between evolution and learning was made over a century ago and is now termed as the Baldwin effect. Learning acts as a guide for an evolutionary search process. In this study reinforcement learning agents are trained to solve the robot coverage control problem. These agents are improved by evolving neuromodulatory gene regulatory networks (GRN) that influence the learning and memory of agents. Agents trained by these neuromodulatory GRNs can consistently generalize better than agents trained with fixed parameter settings. This work introduces evolutionary GRN models into the context of neuromodulation and illustrates some of the benefits that stem from neuromodulatory GRNs.

This paper proposes a new method to evaluate the complexity of a Gene Regulatory Network (GRN). It is based on the generation of pictures. In addition to being visually interesting, the pictures shows the capacity of the GRN to produce smooth and/or sudden transitions, fractal-like complexity and regularities. We also have studied the influence of the size of the GRN on the complexity of pictures generated.

This paper presents a new method to generate the body plans of modular robots. In this work, we use a developmental model where cells are controlled by a gene regulatory network. Instead of using morphogens as in many existing works, we evolve a more flexible "hormonal system" that controls the inputs of the regulatory network. By evolving the regulatory network and the hormonal system in parallel with a blind watchmaker, we have generated various virtual robots with interesting inherent properties such as regularity and symmetry. The prototypes of the robotic blocks that will be used to actually build the real machines are also presented in this paper.

With the purpose of populating virtual worlds with various adapted artificial organisms, we propose an ontogenetic and phylogenetic hybrid model to generate complete organisms possessing metabolism, morphology, and behavior from a single initial cell. The initial purpose of our work is to generate organisms that are thereafter used to define complete organisms. In this paper, we introduce a bio-inspired cellular developmental model that links different approaches of ontogenesis systems: grammatical and cell chemistry approaches. Thus, we propose an alternative to parametric L-systems (APL-systems) in order to simulate morphogenesis of organisms according to their internal states. The developed organisms have a metabolism using environmental substrates to grow and to act. Moreover, they are able to exhibit almost perfect self-healing characteristics afterwards or even during their development.

Artificial embryogeny aims to develop a complete organism starting from a unique cell. For years, plenty of developmental models have been introduced. The main interests are reported on morphogen positioning, differentiation mechanisms and cellular interactions. In this paper, we show how the developmental model Cell2Organ has been extended to become a multi-level simulator able to work both on morphogen positioning thanks to an hydrodynamic layer and on cellular interaction with a physical layer. Through different experiments, we show the capacities of such a model with a “muscular joint” able to move in a physical world and a small organism able to create substrate vortices thanks to the hydrodynamic engine. The inspiration of such a set of simulators is provided the gastrulation stage of vertebrate embryos. During this stage, the embryo reorganizes its environment to continue its growth.

For living organisms, the robustness property is capital. For almost all of them, robustness rhymes with self-repairing. Indeed, organisms are subject to various injuries brought by the environment. To maintain their integrity, organisms are able to regenerate dead parts of themselves. This mechanism, commonly named self-repairing, is interesting to reproduce. Many works exist about self-repairing in robotics and electronics but fewer are in our domain of interest, artificial embryogenesis. In this paper, we show the self-repairing abilities of our model, Cell2Organ, designed to generate artificial creatures for artificial worlds. This model has previously been presented in.

Genetic algorithms are very demanding in terms of computing time and, when the population size is large, they need days to complete or even fail due to memory restrictions. It is particularly the case for artificial life where each evaluation can take more than one minute to develop an artificial creature, plant or organism. Indeed, creatures are developed in physical and chemical simulators that require important computation resources. In order to create more and more realistic creatures, we propose a grid parallelized version of genetic algorithms. Two possibilities exist to increase them: supercomputers or computational grids. Because of their scalability, we choose computational grid in their works.