AI News, Search results: Informatics

Evostar 2019

EvoApplications will focus on the following Thematic Areas: Accepted papers will appear in the proceedings of EvoApplications,

As all EDAs, compact algorithms build and update a probabilistic model of the distribution of solutions within the search space, as opposed to population-based algorithms that instead make use of an explicit population of solutions.

In addition to that, to keep their memory consumption low, compact algorithms purposely employ simple probabilistic models that can be described with a small number of parameters.

To overcome these limitations, here we investigate a possible memetic computing approach obtained by combining compact algorithms with a non-disruptive restart mechanism taken from the literature, named Re-Sampled Inheritance (RI).

The effects of different probabilistic distributions (normal, uniform, triangular and qui-squared ones) to initialize the decision variables (time dials, pickup currents, curves and indexes) are also discussed in this work.

The usage of renewable energy sources, storage devices, and flexible loads has the potential to greatly improve the overall efficiency of a building complex or factory.

We therefore formulated this taskas a many-objective optimization problem with 10 design parameters and 5 objectives (investment cost, yearly energy costs, CO2 emissions, system resilience, and battery lifetime).

We investigated if different optimization algorithms might produce different results, we therefore tested several different many-objective optimization algorithms in terms of their hypervolume performance and the practical relevance of their results.

This paper presents an original approach for building structures that are stable under gravity for the physics-based puzzle game Angry Birds, with the ultimate objective of creating fun and aesthetically pleasing Angry Birds levels with the minimum number of constraints.

It's has been recognized that DRL process is a high-dynamic and non-stationary optimization process even in the static environments, their performance is notoriously sensitive to the hyperparameter configuration which includes learning rate, discount coefficient and step size, etc.

The most ideal state of hyperparameter configuration in DRL is that the hyperparameter can self-adapt to the best values promptly for their current learning state, rather than using a fixed set of hyperparameters for the whole course of training like most previous works did.

The large number of patients and the uniqueness of their diseases demand a considerable amount of diverse and highly personalized therapies, that are nowadays largely managed manually.

This paper aims at catering for the emergent need of efficient and effective artificial intelligence systems for the support of the everyday activities of centers like Casa dos Marcos.

We present six predictive data models developed with a genetic programming based system which, integrated into a web-application, enabled data-driven support for the therapists in Casa dos Marcos.

In this way, we analyze four standard mutation mechanisms present in Differential Evolution algorithms using the Angle Probability List as a source of information to predict tertiary protein structures, something not explored yet with Differential Evolution.

Furthermore, combining experimental data in the optimization process can help the algorithm to avoid premature convergence, maintaining population diversity during the whole process and, consequently, reaching better conformational results.

Distributed Embodied Evolution (dEE) is a powerful approach to learn behaviors in robot swarms by exploiting their intrinsic parallelism: each robot runs an evolutionary algorithm, and locally shares its learning experience with other nearby robots.

In this paper, we investigate the distributed evolution of Gene Regulatory Networks (GRNs) as controller representation to learn swarm robot behavior, which have been extensively used for the evolution of single- robot behavior with remarkable success.

To improve our understanding of such distributed GRN evolution, we analyze the fitness and the behavioral diversity of the swarm over generations when using 5 levels of increasing local selection pressure and 4 different swarm sizes, from 25 to 200 robots.

Our experiments reveal that there exist different regimes, depending on the swarm size, in the relationship between local selection pressure, and both behavioral diversity and overall swarm performance, providing several insights on distributed evolution.

We further use a metric to quantify selection pressure in evolutionary systems, which is based on the correlation between number of offspring and fitness of the behaviors.

This reveals a complex relationship on the overall selection pressure between the ability or ease to spread genomes (or environmental pressure), and the fitness of the behavior (or task-oriented (local) pressure), opening new research questions.

We conclude the paper by discussing the need for developing specialized statistical tools to facilitate the analysis of the large and diverse amount of data relevant to distributed Embodied Evolution.

The procedure for finding the best energy affinity between ligand-receptor molecules is a very computationally expensive optimization process because of the roughness of the search space and the thousands of possible conformations of ligand.

Powered floor systems, i.e., surfaces with conductive strips alternatively connected to the two poles of a power source, are a practical and effective way for supplying power to robots without interruptions, by means of sliding contacts.

We validate experimentally our proposed solution by applying it to three real robots and by studying the impact of the main problem parameters on the effectiveness of the evolved designs for the sliding contacts.

We find that, although there is only a weak correlation between symmetry and fitness over the course of a single evolutionary run, there is a positive correlation between the level of symmetry and maximum fitness when a set of runs is taken into account.

In this paper, we use evolutionary algorithm to evolve customized quantum key distribution (QKD) protocols designed to counter attacks against the system in order to optimize the speed of the secure communication.

The experimental study showed that our memetic algorithm retrieves high-quality heterogeneous ensembles, and can effectively deal with small training sets in multi-class classification.

Quality-diversity algorithms such as MAP-Elites provide a means of supporting the users when finding and choosing solutions to a problem by returning a set of solutions which are diverse according to set of user-defined features.

For a fixed evaluation budget, increasing the number of bins increases user-choice, but at the same time, can lead to a reduction in overall quality of solutions while vice-versa, decreasing the number of bins can lead to higher-quality solutions at the expense of reducing choice.

We note that for the problems under consideration 30 bins or above maximises coverage (and therefore the choice to the end user), whilst fewer bins maximise performance.

Our implementation of GA algorithm showed the best results on a number of test instances, in the role of which we used the cryptanalysis problems of several stream ciphers (cryptographic keystream generators).

The internet and computer networks have become an important asset in distributed computing organisations especially through enabling the collaboration between components of heterogeneous systems.

A profitable strand of literature has lately capitalized on the exploitation of the collaborative capabilities of robotic swarms for efficiently undertaking diverse tasks without any human intervention, ranging from the blind exploration of devastated areas after massive disasters to mechanical repairs of industrial machinery in hostile environments, among others.

In this paper multi-objective heuristic solvers are used to command and route a set of robots towards efficiently reconstructing a scene using simple camera sensors and stereo vision processing techniques.

Most work focuses on evolving learning agents in separate environments, this means each agent experiences its own environment (mostly similar), and has no interactive effect on others (e.g., the more one gains, the more another loses).

Inspired by Neuroevolution of Augmenting Toplogies (NEAT), which is adopted in Artificial Neural Networks (ANNs), our framework includes a genotype to phenotype mapping based on the CPN incidence matrix, and a fitness function, which consider both the behaviour of the evolving CPN and its emerging structural complexity.

Low controller complexity results in simple and conservative gaits, while higher complexity allows the more extreme controllers needed to achieve high performance in demanding environments or tasks, at the cost of a longer period of optimization.

This paper describes a multiobjective differential evolution approach to the optimization of the design of alternating current distributed stator windings of electric motors.

The characterization of the nondominated fronts conveys helpful information for aiding design engineers to choose the most suitable compromise solution for a specific machine, embodying a balanced trade-off between machine efficiency and manufacturing cost.

It introduces a new open data set from a coal-fired power plant, consisting of 10 days of per minute sensor recordings from 12 different burners at the plant.

EXALT provides interesting new techniques for evolving neural networks, including epigenetic weight initialization, where child neural networks re-use parental weights as a starting point to backpropagation, as well as node-level mutation operations which can improve evolutionary progress.

Preliminary results were gathered predicting the Main Flame Intensity data parameter, with EXALT strongly outperforming five traditional neural network architectures on the best, average and worse cases across 10 repeated training runs per test case;

Further, EXALT achived these results 2 to 10 times faster than the traditional methods, in part due to its scalability, showing strong potential to beat traditional architectures given additional runtime.

In an optimization problem, a coreset can be defined as a subset of the input points, such that a good approximation to the optimization problem can be obtained by solving it directly on the coreset, instead of using the whole original input.

In machine learning, coresets are exploited for applications ranging from speeding up training time, to helping humans understand the fundamental properties of a class, by considering only a few meaningful samples.

The problem of discovering coresets, starting from a dataset and an application, can be defined as identifying the minimal amount of samples that do not significangly lower performance with respect to the performance on the whole dataset.

Starting from the consideration that finding coresets is an intuitively multi-objective problem, as minimizing the number of points goes against maintaining the original performance, in this paper we propose a multi-objective evolutionary approach to identifying coresets for classification.

The proposed approach is tested on classical machine learning classification benchmarks, using 6 state-of-the-art classifiers, comparing against 7 algorithms for coreset discovery.

Results show that not only the proposed approach is able to find coresets representing different compromises between compactness and performance, but that different coresets are identified for different classifiers, reinforcing the assumption that coresets might be closely linked to the specific application.

From the performance point of view, the main challenge is to balance computation with communication, but from the algorithmic point of view we have to keep diversity high so that the algorithm is not stuck in local minima.

The common method for identifying proteins and characterising their amino acid sequences is to digest the proteins into peptides, analysing the peptides using mass spectrometry and assign the resulting tandem mass spectra (MS/MS) to peptides using database search tools.

In this study, we pro- pose a genetic algorithm based method, GA-Novo, to solve the complex optimisation task of de novo peptide sequencing, aiming at constructing full length sequences.

On the test- ing dataset, GA-Novo outperforms PEAKS, the the most commonly used software for this task, by constructing 4% higher number of fully matched peptide sequences, and 8% higher recall at partially matched sequences.

Salient Object Detection (SOD) aims to model human visual attention system to cope with the complex natural scene which contains various objects at different scales.

Over the past two decades, a wide range of saliency features have been introduced in the SOD field, however feature selection has not been widely investigated regarding selecting informative, non-redundant, and complementary features from the exciting features.

In this paper, we propose a genetic programming (GP) based method that is able to automatically select the complementary saliency features and generate mathematical function to combine those features.

We investigate the problem of optimally placing virtual network functions in 5G virtualized infrastructures according to a green paradigm that pursues energy efficiency.

In the worldwide expansion of renewable energies, there is not only a need for weather-dependent plants, but also for plants with flexible power generation that have the potential to reduce storage requirements by working against fluctuations.

The planning of a CHP plant, whose generated heat always finds a consumer and the generated electricity is simultaneously optimized with regard to an optimization objective, requires nonlinear optimization approaches due to the physical effects in the heat storage.

Evolutionary Computation based techniques and in particular genetic algorithms, because of their ability to explore large and complex search spaces, have proven to be effective in solving such kind of problems.

Though genetic algorithms binary strings provide a natural way to represent feature subsets, several different representation schemes have been proposed to improve the performance, with most of them needing to a priori set the number of features.

In NeuroEvolution, the topologies of artificial neural networks are optimized with evolutionary algorithms to solve tasks in data regression, data classification, or reinforcement learning.

The obtained results, in terms of average score and win percentage, seem quite satisfactory and highlight the advantages of the suggested technique, especially when compared to a rolling horizon GA implementation of the aforementioned framework;

With this recombination scheme, solutions at the local populations are recombined using a weighted aver- age that favors fitter solutions to produce a new solution.

The new measure was compared against the previously introduced measure in terms of triangle inequality satisfiability, changes in raw measure values and the computational cost.

The investigation of global convexity of the fitness landscape, involving the fitness--distance correlation analysis, revealed positive correlation between the similarity of structures and their fitness for most of the investigated cases.

Wireless Underground Sensor Networks (WUSNs) have received attention in the past years because of their popularity and cost effectiveness when they are used in many real fields such as military applications, environmental applications, and home applications.

Several approaches have been proposed to keep the sensor nodes active, one of which is deploying relay nodes above ground to transfer data from sensor nodes to the base station.

This paper addresses this concern and proposes heuristics for relay nodes placement problem to guarantee load balance among relay nodes and maximize network lifetime.The experimental results show that the proposed methods result in quality solutions for the problem, when compared to existing methods

Browser Upgrade Recommended

The successful candidate will work on the EPSRC project 'Combining Viewpoints in Quantum Theory' EP/R044759/1 in one of two areas: optimisation of quantum programs, or developing a categorical framework for spatially distributed quantum protocols.

It was founded in 1987 and is a community of theoretical computer scientists with interests in concurrency, semantics, categories, algebra, types, logic, algorithms, complexity, databases and modelling.

Vacancy Ref: 047668 Closing Date:17-MAY-2019  at 5pm GMT For further particulars and to apply for this post please click on the 'apply' button below https://www.vacancies.ed.ac.uk/pls/corehrrecruit/erq_jobspec_version_4.jobspec?p_id=047668 The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336.

Master Artificial Intelligence

1.3a. AIPLAN - Planning and Search

Week one of the University of Edinburgh's "Artificial Intelligence Planning" (AIPLAN) open online course. Dr. Gerhard Wickler and Prof. Austin Tate School of ...

3.3. AIPLAN - Plan Space Search

Week three of the University of Edinburgh's "Artificial Intelligence Planning" (AIPLAN) open online course. Dr. Gerhard Wickler and Prof. Austin Tate School of ...

[Feature] AIPLAN - Joerg Hoffmann on Heuristic Search

Feature video (Heuristic Search: Joerg Hoffmann) of the University of Edinburgh's "Artificial Intelligence Planning" (AIPLAN) open online course. Dr. Gerhard ...

2.4b. AIPLAN - A* Graph Search

Week two of the University of Edinburgh's "Artificial Intelligence Planning" (AIPLAN) open online course. Dr. Gerhard Wickler and Prof. Austin Tate School of ...

2.10. AIPLAN - Backward Search

Week two of the University of Edinburgh's "Artificial Intelligence Planning" (AIPLAN) open online course. Dr. Gerhard Wickler and Prof. Austin Tate School of ...

MSc Artificial Intelligence Computer vision

1.3b. AIPLAN - Planning and Search

Week one of the University of Edinburgh's "Artificial Intelligence Planning" (AIPLAN) open online course. Dr. Gerhard Wickler and Prof. Austin Tate School of ...

[Feature] AIPLAN - Alex Champandard on AI Planning in Games

Feature video (AI Planning in Games: Alex Champandard) of the University of Edinburgh's "Artificial Intelligence Planning" (AIPLAN) open online course.

4.10. AIPLAN - The FF Planner

Week four of the University of Edinburgh's "Artificial Intelligence Planning" (AIPLAN) open online course. Dr. Gerhard Wickler and Prof. Austin Tate School of ...