Recent Work in Applied Genetic Programming

Synthetic-analytic behavior-based control framework: Constraining velocity in tracking for nonholonomic wheeled mobile robot
Information Sciences. Volume 501, October 2019, Pages 436-459 – by MarlenMeza-Sánchez, Clemente, Rodríguez-Liñán, and Olague.
Abstract: This work presents a genetic programming control design methodology that extends the traditional behavior–based control strategy towards a synthetic-analytic perspective. The proposed approach considers the internal and external dynamics of the system, providing solutions to a general structure, and including analytic functions, which can be studied within the Control Theory framework. The method is illustrated for the tracking control problem under bounded velocity restrictions of a nonholonomic wheeled mobile robot.

New Pathways in Coevolutionary Computation
GPTP 2019 – by Sipper, Moore, and Urbanowicz
Abstract: The simultaneous evolution of two or more species with coupled fitness—coevolution—has been put to good use in the field of evolutionary computation. Herein, we present two new forms of coevolutionary algorithms, which we have recently designed and applied with success. OMNIREP is a cooperative coevolutionary algorithm that discovers both a representation and an encoding for solving a particular problem of interest. SAFE is a commensalistic coevolutionary algorithm that maintains two coevolving populations: a population of candidate solutions and a population of candidate objective functions needed to measure solution quality during evolution.

Artificial Visual Cortex and Random Search for Object Categorization
April 2019 – by Olague, Clemente, Hernández, Barrera, Chan-Ley, and Bakshi
Abstract: This paper presents results about a model of an artificial brain that is used to recreate the ventral and dorsal stream. Brain programming is a kind of deep learning approach where neuroscience knowledge is merged with genetic programming to synthesize artificial models of the brain. Brain programming applies the paradigm of artificial evolution by combining hierarchical networks with multi-tree programs. This paper shows that a simple random search can find optimal programs that solve non-trivial object categorization problems. The approach follows a function-driven paradigm in contrast to popular data-driven models.

Local search in speciation-based bloat control for genetic programming
March 2019 – by Juárez-Smith, Trujillo, García-Valdez, Fernández de Vega, and Chávez
Abstract: This work presents a unique genetic programming (GP) approach that integrates a numerical local search method and a bloat-control mechanism to address some of the main issues with traditional GP. The former provides a directed search operator to work in conjunction with standard syntax operators that perform more exploration in design space, while the latter controls code growth by maintaining program diversity through speciation. The system can produce highly parsimonious solutions, thus reducing the cost of performing the local optimization process. Results show that the proposed approach compares favorably with a standard approach, and that the hybrid algorithm can be used as a viable alternative for solving real-world symbolic regression problems.

Synthesis of Vegetation Indices Using Genetic Programming for Soil Erosion Estimation
January 2019 – by Puente, Olague, Trabucchi, Arjona-Villicaña, and Soubervielle-Montalvo
Abstract: The aim of this research is to propose a novel approach for calculating new Vegetation Indices (VIs) that are better correlated with C, using field and satellite information. The approach followed by this research is to state the generation of new VIs in terms of a computer optimization problem and then applying a machine learning technique, named Genetic Programming (GP), which builds new indices by iteratively recombining a set of numerical operators and spectral channels until the best composite operator is found.

Emergent solutions to high-dimensional multitask reinforcement learning
September 2018 – by Kelly and Heywood
Algorithms that learn through environmental interaction and delayed rewards, or reinforcement learning (RL), increasingly face the challenge of scaling to dynamic, high-dimensional, and partially observable environments. In this work, we propose a framework based on genetic programming which adaptively complexifies policies through interaction with the task. We make a direct comparison with multiple deep reinforcement learning frameworks in the challenging arcade video game environment as well as more traditional reinforcement learning frameworks based on a priori engineered features. Results indicate that the proposed approach matches the quality of deep learning while being a minimum of three orders of magnitude simpler with respect to model complexity. This results in real-time operation of the champion RL agent without recourse to specialized hardware support. Moreover, the approach is capable of evolving solutions to multiple game titles simultaneously with no additional computational cost.

Evolving Head Tracking Routines With Brain Programming
April 2018 – by Olague, Hernández, Clemente, and Chan-Ley
Abstract: Brain programming is a kind of deep learning approach where evolutionary computing is merged with neuroscience knowledge to create artificial models of the brain. Brain programming applies the paradigm of artificial evolution by merging hierarchical networks with multi-tree programs. This approach could be seen as a kind of deep genetic programming. Here is applied to the problem of head tracking.

Enhanced Optimization with Composite Objectives and Novelty Selection
March 2018 – by Shahrzad, Fink, and Miikkulainen
Abstract: An important benefit of multi-objective search is that it maintains a diverse population of candidates, which helps in deceptive problems in particular. Not all diversity is useful, however: candidates that optimize only one objective while ignoring others are rarely helpful. This paper proposes a solution: The original objectives are replaced by their linear combinations, thus focusing the search on the most useful tradeoffs between objectives. To compensate for the loss of diversity, this transformation is accompanied by a selection mechanism that favors novelty. In the highly deceptive problem of discovering minimal sorting networks, this approach finds better solutions, and finds them faster and more consistently than standard methods. It is therefore a promising approach to solving deceptive problems through multi-objective optimization.

An evolutionary learning and network approach to identifying key metabolites for osteoarthritis
March 2018 – by Hu, Oksanen, Zhang, Randell, Furey, Sun, and Zhai
Abstract: Biomedical research has entered a new era where a large number of molecules and different components in biological systems can be quantitatively examined to investigate the causes of common human diseases. However, given the complexity of biological systems, those causes may not contribute to diseases individually but through interactions. The identification of those interactions, or the synergy of multiple factors, is a very challenging task due to the computational limitation, as well as the lack of effective methodologies for investigating multiple factors simultaneously. In this study, we proposed to model such an interaction effect through a self-learning algorithm using mechanisms inspired by natural evolution, termed genetic programming (GP). Moreover, by constructing a synergy network using those evolved models, we were able to identify a set of interacting factors associated with a particular disease.

Finding the origin of noise transients in LIGO data with machine learning
December 2018 – by Cavaglia, Staats, and Gill
Abstract: Quality improvement of data collected at Advanced LIGO and Virgo is mission critical for the success of gravitational-wave astrophysics. Removing non-astrophysical artifacts that corrupt the data stream is crucial for increasing the number and statistical significance of gravitational-wave detections and enabling refined astrophysical interpretations of the data. Machine learning has proved to be a powerful tool for analysis of massive quantities of complex data in astronomy and related fields of study. We present two machine learning methods, based on random forest and genetic programming algorithms, that can be used to determine the origin of non-astrophysical transients in the LIGO detectors.

TensorFlow Enabled Genetic Programming
August 2017 – by Staats, Pantridge, Cavaglia, Milovanov, and Aniyan
Abstract: Genetic Programming, a kind of evolutionary computation and machine learning algorithm, is shown to benefit significantly from the application of vectorized data and the TensorFlow numerical computation library on both CPU and GPU architectures. The open source, Python Karoo GP is employed for a series of 190 tests across 6 platforms, with real-world datasets ranging from 18 to 5.5M data points. This body of tests demonstrates that datasets measured in tens and hundreds of data points see 2-15x improvement when moving from the scalar/SymPy configuration to the vector/TensorFlow configuration, with a single core performing on par or better than multiple CPU cores and GPUs. A dataset composed of 90,000 data points demonstrates a single vector/TensorFlow CPU core performing 875x better than 40 scalar/Sympy CPU cores. And a dataset containing 5.5M data points sees GPU configurations out-performing CPU configurations on average by 1.3x.