Marc-André Gardner

DEAP: a python framework for evolutionary algorithms

DEAP (Distributed Evolutionary Algorithms in Python) is a novel volutionary computation framework for rapid prototyping and testing of ideas. Its design departs from most other existing frameworks in that it seeks to make algorithms explicit and data structures transparent, as opposed to the more common black box type of frameworks. It also incorporates easy parallelism where users need not concern themselves with gory implementation details like synchronization and load balancing, only functional decomposition. Several examples illustrate the multiple properties of DEAP.

Learning to predict indoor illumination from a single image

We propose an automatic method to infer high dynamic range illumination from a single, limited field-of-view, low dynamic range photograph of an indoor scene. In contrast to previous work that relies on specialized image capture, user input, and/or simple scene models, we train an end-to-end deep neural network that directly regresses a limited field-of-view photo to HDR illumination, without strong assumptions on scene geometry, material properties, or lighting. We show that this can be accomplished in a three step process: 1) we train a robust lighting classifier to automatically annotate the location of light sources in a large dataset of LDR environment maps, 2) we use these annotations to train a deep neural network that predicts the location of lights in a scene from a single limited field-of-view photo, and 3) we fine-tune this network using a small dataset of HDR environment maps to predict light intensities. This allows us to automatically recover high-quality HDR illumination estimates that significantly outperform previous state-of-the-art methods. Consequently, using our illumination estimates for applications like 3D object insertion, produces photo-realistic results that we validate via a perceptual user study.

DEAP: enabling nimbler evolutions

DEAP is a Distributed Evolutionary Algorithm (EA) framework written in Python and designed to help researchers developing custom evolutionary algorithms. Its design philosophy promotes explicit algorithms and transparent data structures, in contrast with most other evolutionary computation softwares that tend to encapsulate standardized algorithms using the black-box approach. This philosophy sets it apart as a rapid prototyping framework for testing of new ideas in EA research. An executable notebook version of this paper is available at https://github.com/DEAP/notebooks.

ros4mat: A Matlab Programming Interface for Remote Operations of ROS-Based Robotic Devices in an Educational Context

More and more, robotics is perceived in education as being an excellent way to promote higher quality learning among students, by grounding theoretical concepts into reality. In order to maximize the learning throughput, the focus of any robotics software platform should be on ease of use, with little time spent integrating the components together. To this effect, we introduce ros4mat, an open source library which provides a simple and flexible interface between ROS (Robot Operating System) and Matlab®. The conception is focused on academic use, and allows a very simple integration of sensors and actuators to existing Matlab code. The library is designed to provide an easy, platform-independent, and fast connection between a robot (running ROS) and multiple clients (running only Matlab). Moreover, it is very versatile and can be used with many common types of sensors in robotics, including low-cost ones. We report the results of ros4mat use in a robotics course to provide more real world experimentation, along with some code samples illustrating the simplicity of our approach.

Controlling code growth by dynamically shaping the genotype size distribution

Genetic programming is a hyperheuristic optimization approach that seeks to evolve various forms of symbolic computer programs, in order to solve a wide range of problems. However, the approach can be severely hindered by a significant computational burden and stagnation of the evolution caused by uncontrolled code growth. This paper introduces HARM-GP, a novel operator equalization method that conducts an adaptive shaping of the genotype size distribution of individuals in order to effectively control code growth. Its probabilistic nature minimizes the computational overheads on the evolutionary process while its generic formulation allows it to remain independent of both the problem and the genetic variation operators. Comparative results over twelve problems with different dynamics, and over nine other algorithms taken from the literature, show that HARM-GP is excellent at controlling code growth while maintaining good overall performance. Results also demonstrate the effectiveness of HARM-GP at limiting overfitting in real-world supervised learning problems.

Estimation of distribution algorithm based on hidden Markov models for combinatorial optimization

Estimation of Distribution Algorithms (EDAs) have been successfully applied to a wide variety of problems. The algorithmic model of EDA is generic and can virtually be used with any distribution model, ranging from the mere Bernoulli distribution to the sophisticated Bayesian network. The Hidden Markov Model (HMM) is a well-known graphical model useful for modelling populations of variable-length sequences of discrete values. Surprisingly, HMMs have not yet been used as distribution estimators for an EDA, even though it is a very powerful tool especially designed for modelling sequences. We thus propose a new method, called HMM-EDA, implementing this idea. Preliminary comparative results on two classical combinatorial optimization problems show that HMM-EDA is indeed a promising approach for problems that have sequential representations.

Combinatorial optimization EDA using hidden Markov models

Estimation of Distribution Algorithms (EDAs) have been successfully applied to a wide variety of problems. The algorithmic model of EDA is generic and can virtually be used with any distribution model, ranging from the mere Bernoulli distribution to the sophisticated Bayesian network. The Hidden Markov Model (HMM) is a well-known graphical model useful for modelling populations of variable-length sequences of discrete values. Surprisingly, HMMs have not yet been used as distribution estimators for an EDA, although they are a very powerful tool for estimating sequential samples. This paper thus proposes a new method, called HMM-EDA, implementing this idea, along with some preliminary experimental results.

Bloat control in genetic programming with a histogram-based accept-reject method

Recent bloat control methods such as dynamic depth limit (DynLimit) and Dynamic Operator Equalization (DynOpEq) aim at modifying the tree size distribution in a population of genetic programs. Although they are quite efficient for that purpose, these techniques have the disadvantage of evaluating the fitness of many bloated Genetic Programming (GP) trees, and then rejecting most of them, leading to an important waste of computational resources. We are proposing a method that makes a histogram-based model of current GP tree size distribution, and uses the so-called accept-reject method for generating a population with the desired target size distribution, in order to make a stochastic control of bloat in the course of the evolution. Experimental results show that the method is able to control bloat as well as other state-of-the-art methods, with minimal additionnal computational efforts compared to standard tree-based GP.