Pawel Lichocki
Paweł Lichocki is a Software Engineer at Google in Operations Research team where he works on combinatorial optimization. He received his PhD in Computer Science from École Polytechnique Fédérale de Lausanne for work on evolution of division of labor in multi-agent systems. Prior, he was a researcher in Supercomputer and Networking Center in Poznań where he worked on parallel and distributed processing algorithms.
Authored Publications
Google Publications
Other Publications
Sort By
Mechanosensory interactions drive collective behaviour in Drosophila
Pavan Ramdya
Steeve Cruchet
Lukas Frisch
Winnie Tse
Dario Floreano
Richard Benton
Nature, 519(2015), 233–236
Preview abstract
Collective behaviour enhances environmental sensing and decision-making in groups of animals. Experimental and theoretical investigations of schooling fish, flocking birds and human crowds have demonstrated that simple interactions between individuals can explain emergent group dynamics. These findings indicate the existence of neural circuits that support distributed behaviours, but the molecular and cellular identities of relevant sensory pathways are unknown. Here we show that Drosophila melanogaster exhibits collective responses to an aversive odour: individual flies weakly avoid the stimulus, but groups show enhanced escape reactions. Using high-resolution behavioural tracking, computational simulations, genetic perturbations, neural silencing and optogenetic activation we demonstrate that this collective odour avoidance arises from cascades of appendage touch interactions between pairs of flies. Inter-fly touch sensing and collective behaviour require the activity of distal leg mechanosensory sensilla neurons and the mechanosensory channel NOMPC. Remarkably, through these inter-fly encounters, wild-type flies can elicit avoidance behaviour in mutant animals that cannot sense the odour - a basic form of communication. Our data highlight the unexpected importance of social context in the sensory responses of a solitary species and open the door to a neural-circuit-level understanding of collective behaviour in animal groups.
View details
Selection methods regulate evolution of cooperation in digital evolution
Dario Floreano
Laurent Keller
Journal of The Royal Society Interface, 11(2014), pp. 20130743
Preview abstract
A key, yet often neglected, component of digital evolution and evolutionary models is the ‘selection method’ which assigns fitness (number of offspring) to individuals based on their performance scores (efficiency in performing tasks). Here, we study with formal analysis and numerical experiments the evolution of cooperation under the five most common selection methods (proportionate, rank, truncation-proportionate, truncation-uniform and tournament). We consider related individuals engaging in a Prisoner's Dilemma game where individuals can either cooperate or defect. A cooperator pays a cost, whereas its partner receives a benefit, which affect their performance scores. These performance scores are translated into fitness by one of the five selection methods. We show that cooperation is positively associated with the relatedness between individuals under all selection methods. By contrast, the change in the performance benefit of cooperation affects the populations’ average level of cooperation only under the proportionate methods. We also demonstrate that the truncation and tournament methods may introduce negative frequency-dependence and lead to the evolution of polymorphic populations. Using the example of the evolution of cooperation, we show that the choice of selection method, though it is often marginalized, can considerably affect the evolutionary dynamics.
View details
The hourglass and the early conservation models - co-existing patterns of developmental constraints in vertebrates
Barbara Piasecka
Sébastien Moretti
Sven Bergmann
Marc Robinson-Rechavi
PLOS Genetics, 9(2013), e1003476
Preview abstract
Developmental constraints have been postulated to limit the space of feasible phenotypes and thus shape animal evolution. These constraints have been suggested to be the strongest during either early or mid-embryogenesis, which corresponds to the early conservation model or the hourglass model, respectively. Conflicting results have been reported, but in recent studies of animal transcriptomes the hourglass model has been favored. Studies usually report descriptive statistics calculated for all genes over all developmental time points. This introduces dependencies between the sets of compared genes and may lead to biased results. Here we overcome this problem using an alternative modular analysis. We used the Iterative Signature Algorithm to identify distinct modules of genes co-expressed specifically in consecutive stages of zebrafish development. We then performed a detailed comparison of several gene properties between modules, allowing for a less biased and more powerful analysis. Notably, our analysis corroborated the hourglass pattern at the regulatory level, with sequences of regulatory regions being most conserved for genes expressed in mid-development but not at the level of gene sequence, age, or expression, in contrast to some previous studies. The early conservation model was supported with gene duplication and birth that were the most rare for genes expressed in early development. Finally, for all gene properties, we observed the least conservation for genes expressed in late development or adult, consistent with both models. Overall, with the modular approach, we showed that different levels of molecular evolution follow different patterns of developmental constraints. Thus both models are valid, but with respect to different genomic features.
View details
Evolving team compositions by agent swapping
Steffen Wischmann
Laurent Keller
Dario Floreano
IEEE Transactions on Evolutionary Computation, 17(2013), pp. 282-298
Preview abstract
Optimizing collective behavior in multiagent systems requires algorithms to find not only appropriate individual behaviors but also a suitable composition of agents within a team. Over the last two decades, evolutionary methods have emerged as a promising approach for the design of agents and their compositions into teams. The choice of a crossover operator that facilitates the evolution of optimal team composition is recognized to be crucial, but so far, it has never been thoroughly quantified. Here, we highlight the limitations of two different crossover operators that exchange entire agents between teams: restricted agent swapping (RAS) that exchanges only corresponding agents between teams and free agent swapping (FAS) that allows an arbitrary exchange of agents. Our results show that RAS suffers from premature convergence, whereas FAS entails insufficient convergence. Consequently, in both cases, the exploration and exploitation aspects of the evolutionary algorithm are not well balanced resulting in the evolution of suboptimal team compositions. To overcome this problem, we propose combining the two methods. Our approach first applies FAS to explore the search space and then RAS to exploit it. This mixed approach is a much more efficient strategy for the evolution of team compositions compared to either strategy on its own. Our results suggest that such a mixed agent-swapping algorithm should always be preferred whenever the optimal composition of individuals in a multiagent system is unknown.
View details
Neural networks as mechanisms to regulate division of labor
Danesh Tarapore
Laurent Keller
Dario Floreano
The American Naturalist, 179(2012), pp. 391-400
The ethical landscape of robotics
Preview abstract
In this article, we highlight the possible benefits, as well potential threats, related to the widespread use of robots. We follow the view that a robot cannot be analyzed on its own without taking into consideration the complex sociotechnical nexus of today's societies and that high-tech devices, such as robots, may influence how societies develop in ways could not be foreseen during the design of the robots. In our survey, we limit ourselves to presenting the ethical issues delineated by other authors and relay their lines of reasoning for raising the public's concerns. We show that disagree ments on what is ethical or not in robotics stem often from different beliefs on human nature and different expectations on what technology may achieve in the future. We do not offer a personal stance to these issues, so as to allow the reader to form his/her opinion.
View details
Two-dimensional discrete wavelet transform on large images for hybrid computing architectures: GPU and CELL
Marek Błażewicz
Miłosz Ciżnicki
Piotr Kopta
Krzysztof Kurowski
European Conference on Parallel Processing, Springer, Berlin, Heidelberg(2011), pp. 481-490
Preview abstract
The Discrete Wavelet Transform (DWT) has gained the momentum in signal processing and image compression over the last decade bringing the concept up to the level of new image coding standard JPEG2000. Thanks to many added values in DWT, in particular inherent multi-resolution nature, wavelet-coding schemes are suitable for various applications where scalability and tolerable degradation are relevant. Moreover, as we demonstrate in this paper, it can be used as a perfect benchmarking procedure for more sophisticated data compression and multimedia applications using General Purpose Graphical Processor Units (GPGPUs). Thus, in this paper we show and compare experiments performed on reference implementations of DWT on Cell Broadband Engine Architecture (Cell B.E) and nVidia Graphical Processing Units (GPUs). The achieved results show clearly that although both GPU and Cell B.E. are being considered as representatives of the same hybrid architecture devices class they differ greatly in programming style and optimization techniques that need to be taken into account during the development. In order to show the speedup, the parallel algorithm has been compared to sequential computation performed on the x86 architecture.
View details
Using co-solvability to model and exploit synergetic effects in evolution
Krzysztof Krawiec
International Conference on Parallel Problem Solving from Nature, Springer, Berlin, Heidelberg(2010), pp. 492-501
Preview abstract
We introduce, analyze, and experimentally examine co-solvability, an ability of a solution to solve a pair of fitness cases (tests). Based on this concept, we devise a co-solvability fitness function that makes solutions compete for rewards granted for solving pairs of tests, in a way analogous to implicit fitness sharing. We prove that co-solvability fitness function is by definition synergistic and imposes selection pressure which is qualitatively different from that of standard fitness function or implicit fitness sharing. The results of experimental verification on eight genetic programming tasks demonstrate that evolutionary runs driven by co-solvability fitness function usually converge faster to well-performing solutions and are more likely to reach global optima.
View details
Approximating geometric crossover in semantic space
Krzysztof Krawiec
Proceedings of the 11th Annual Conference on Genetic and Evolutionary Computation, ACM(2009), pp. 987-994
Preview abstract
We propose a crossover operator that works with genetic programming trees and is approximately geometric crossover in the semantic space. By defining semantic as program's evaluation profile with respect to a set of fitness cases and constraining to a specific class of metric-based fitness functions, we cause the fitness landscape in the semantic space to have perfect fitness-distance correlation. The proposed approximately geometric semantic crossover exploits this property of the semantic fitness landscape by an appropriate sampling. We demonstrate also how the proposed method may be conveniently combined with hill climbing. We discuss the properties of the methods, and describe an extensive computational experiment concerning logical function synthesis and symbolic regression.
View details
Evolving teams of cooperating agents for real-time strategy game
Krzysztof Krawiec
Wojciech Jaśkowski
Proceedings of the EvoWorkshops 2009 on Applications of Evolutionary Computing, Springer Berlin/Heidelberg(2009), pp. 333-342
Preview abstract
We apply gene expression programing to evolve a player for a real-time strategy (RTS) video game. The paper describes the game, evolutionary encoding of strategies and the technical implementation of experimental framework. In the experimental part, we compare two setups that differ with respect to the used approach of task decomposition. One of the setups turns out to be able to evolve an effective strategy, while the other leads to more sophisticated yet inferior solutions. We discuss both the quantitative results and the behavioral patterns observed in the evolved strategies.
View details
