Members of the SPOTSeven Lab are speaking at or organizing the following GECCO Events.
Sunday July 11th 2021
Workshop Keynote: Surrogate Model Based Hyperparameter Tuning for Deep Learning with SPOT
July 11, 2021, 8:30 AM – 9:20 AM (Europe/Paris)
Workshop Room 3
Speakers: Thomas Bartz-Beielstein
Abstract: A surrogate model based (SMB) hyperparameter tuning approach for deep learning is presented. Typical questions regarding hyperparameters in deep-learning models are as follows:
– How many layers should be stacked?
– Which dropout rate should be used?
– How many filters (units) should be used in each layer?
To date, there is no comprehensive theory that adequately explains how to answer these questions. We will demonstrate how the architecture-level parameters (hyperparameters) of deep learning models that were implemented in Keras/tensorflow can be optimized.
R, the software environment for statistical computing, will be used in this talk: with a few lines of code, functions from available R packages can be combined to perform hyperparameter tuning. An elementary hyperparameter tuning task (neural network and the MNIST data) is used to exemplify our SMB approach. We will use statistical tools for understanding hyperparameter importance and interactions between several hyperparameters. Recommendations, problems, and pitfalls of this experimental approach will be discussed.
References: Keynote talk is accompanied by the arXiv publication: Surrogate Model Based Hyperparameter Tuning for Deep Learning with SPOT
Competition on Optimization of a simulation model for a capacity and resource planning task for hospitals under special consideration of the COVID-19 pandemic
July 11, 2021, 2:30 PM – 2:45 PM (Europe/Paris)
Competition Room
Speakers: Margarita Rebolledo, Frederik Rehbach, Sowmya Chandrasekaran, Thomas Bartz-Beielstein
Abstract: Similar to the many previous competitions, the team of the Institute of Data Science, Engineering, and Analytics at the TH Cologne (IDE+A), hosts the ‘Industrial Challenge’ at the GECCO 2021. This year’s industrial challenge is posed in cooperation with an IDE+A partner from health industry and with Bartz & Bartz GmbH.
Simulation models are valuable tools for resource usage estimation and capacity planning. Your goal is to determine improved simulation model parameters for a capacity and resource planning task for hospitals. The simulator, babsim.hospital, explicitly covers difficulties for hospitals caused by the COVID-19 pandemic. The simulator can handle many aspects of resource planning in hospitals: – various resources such as ICU beds, ventilators, personal protection equipment, staff, pharmaceuticals – several cohorts (based on age, health status, etc.).
The task represents an instance of an expensive, high-dimensional computer simulation-based optimization problem and provides an easy evaluation interface that will be used for the setup of our challenge. The simulation will be executed through an interface and hosted on one of our servers (similar to our last year’s challenge). The task is to find an optimal parameter configuration for the babsim.hospital simulator with a very limited budget of objective function evaluations. The best-found objective function value counts. There will be multiple versions of the babsim.hospital simulations, with slightly differing optimization goals, so that algorithms can be developed and tested before they are submitted for the final evaluation in the challenge. The participants will be free to apply one or multiple optimization algorithms of their choice. Thus, we enable each participant to apply his/her algorithms to a real problem from health industry, without software setup or licensing that would usually be required when working on such problems.
Behavior-based Neuroevolutionary Training in Reinforcement Learning
July 11, 2021, 4:00 PM – 4:25 PM (Europe/Paris)
Workshop Room 5
Speakers: Jörg Stork, Martin Zaefferer, Thomas Bartz-Beielstein, Gusz Eiben
Abstract: In addition to their undisputed success in solving classical optimization problems, neuroevolutionary and population-based algorithms have become an alternative to standard reinforcement learning methods. However, evolutionary methods often lack the sample efficiency of standard value-based methods that leverage gathered state and value experience. If reinforcement learning for real-world problems with significant resource cost is considered, sample efficiency is essential. The enhancement of evolutionary algorithms with experience exploiting methods is thus desired and promises valuable insights. This work presents a hybrid algorithm that combines topology-changing neuroevolutionary optimization with value-based reinforcement learning. We illustrate how the behavior of policies can be used to create distance and loss functions, which benefit from stored experiences and calculated state values. They allow us to model behavior and perform a directed search in the behavior space by gradient-free evolutionary algorithms and surrogate-based optimization. For this purpose, we consolidate different methods to generate and optimize agent policies, creating a diverse population. We exemplify the performance of our algorithm on standard benchmarks and a purpose-built real-world problem. Our results indicate that combining methods can enhance the sample efficiency and learning speed for evolutionary approaches.
References: https://doi.org/10.1145/3449726.3463171
Monday, July 12th 2021
Evolutionary Computation in Practice (ECiP)
Discussion
July 12, 2021, 11:20 AM – 11:40 AM (Europe/Paris)
Chairs: Thomas Bartz-Beielstein, Bogdan Filipic, Sowmya Chandrasekaran
Abstract: In the Evolutionary Computation in Practice (ECiP) track, well-known speakers with outstanding reputation in academia and industry present background and insider information on how to establish reliable cooperation with industrial partners. They actually run companies or are involved in cooperations between academia and industry.
If you attend, you will learn multiple ways to extend EC practice beyond the approaches found in textbooks. Experts in real-world optimization with decades of experience share their approaches to creating successful projects for real-world clients. Some of what they do is based on sound project management principles, and some is specific to our type of optimization projects. If you are working in academia and are interested in managing industrial projects, you will receive valuable hints for your own research projects.
In 2021, ECiP will be online (virtual) for the first time, which is a great challenge for a track that relies on personal interactions. We will do our best to enable opportunities for establishing contacts among participants.
More: Evolutionary Computation in Practice (ECiP)
Evolutionary Computation in Practice (ECiP)
Discussion
July 12, 2021, 3:20 PM – 3:40 PM (Europe/Paris)
Chairs: Thomas Bartz-Beielstein, Bogdan Filipic, Sowmya Chandrasekaran
Abstract: In the Evolutionary Computation in Practice (ECiP) track, well-known speakers with outstanding reputation in academia and industry present background and insider information on how to establish reliable cooperation with industrial partners. They actually run companies or are involved in cooperations between academia and industry.
If you attend, you will learn multiple ways to extend EC practice beyond the approaches found in textbooks. Experts in real-world optimization with decades of experience share their approaches to creating successful projects for real-world clients. Some of what they do is based on sound project management principles, and some is specific to our type of optimization projects. If you are working in academia and are interested in managing industrial projects, you will receive valuable hints for your own research projects.
In 2021, ECiP will be online (virtual) for the first time, which is a great challenge for a track that relies on personal interactions. We will do our best to enable opportunities for establishing contacts among participants.
More: Evolutionary Computation in Practice (ECiP)
Impact of Energy Efficiency on the Morphology and Behaviour of Evolved Robots
July 12, 2021, 4:00 PM – 6:00 PM (Europe/Paris)
Poster Room 1
Speakers: Margarita Rebolledo, Thomas Bartz-Beielstein, Gusz Eiben
Abstract: Most evolutionary robotics studies focus on evolving some targeted behavior without considering energy usage. In this paper, we extend our simulator with a battery model to take energy consumption into account in a system where robot morphologies and controllers evolve simultaneously. The results show that including the energy consumption in the fitness in a multi-objective fashion reduces the average size of robot bodies while reducing their speed. However, robots generated without size reduction can achieve speeds comparable to robots from the baseline set.
References: https://doi.org/10.1145/3449726.3459489
Resource Planning for Hospitals Under Special Consideration of the COVID-19 Pandemic: Optimization and Sensitivity Analysis
July 12, 2021, 4:00 PM – 6:00 PM (Europe/Paris)
Poster Room 2
Speakers: Thomas Bartz-Beielstein, Marcel Dröscher, Alpar Gür, Dessislava Peeva, Nicolas Rehbach, Frederik Rehbach, Amrita Sen, Aleksandr Subbotin, Martin Zaefferer
Abstract: Pandemics pose a serious challenge to health-care institutions. To support the resource planning of health authorities from the Cologne region, BaBSim.Hospital, a tool for capacity planning based on discrete event simulation, was created. The predictive quality of the simulation is determined by 29 parameters with reasonable default values obtained in discussions with medical professionals. We aim to investigate and optimize these parameters to improve BaBSim.Hospital using a surrogate-based optimization approach and an in-depth sensitivity analysis.
References: https://doi.org/10.1145/3449726.3459473
Tuesday, July 13th 2021
Impact of Energy Efficiency on the Morphology and Behaviour of Evolved Robots
July 13, 2021, 9:00 AM – 10:20 AM (Europe/Paris)
Poster Room 1
Speakers: Margarita Rebolledo, Thomas Bartz-Beielstein, Gusz Eiben
Abstract: Most evolutionary robotics studies focus on evolving some targeted behavior without considering energy usage. In this paper, we extend our simulator with a battery model to take energy consumption into account in a system where robot morphologies and controllers evolve simultaneously. The results show that including the energy consumption in the fitness in a multi-objective fashion reduces the average size of robot bodies while reducing their speed. However, robots generated without size reduction can achieve speeds comparable to robots from the baseline set.
References: https://doi.org/10.1145/3449726.3459489
Resource Planning for Hospitals Under Special Consideration of the COVID-19 Pandemic: Optimization and Sensitivity Analysis
July 13, 2021, 9:00 AM – 10:20 AM (Europe/Paris)
Poster Room 2
Speakers: Thomas Bartz-Beielstein, Marcel Dröscher, Alpar Gür, Dessislava Peeva, Nicolas Rehbach, Frederik Rehbach, Amrita Sen, Aleksandr Subbotin, Martin Zaefferer
Abstract: Pandemics pose a serious challenge to health-care institutions. To support the resource planning of health authorities from the Cologne region, BaBSim.Hospital, a tool for capacity planning based on discrete event simulation, was created. The predictive quality of the simulation is determined by 29 parameters with reasonable default values obtained in discussions with medical professionals. We aim to investigate and optimize these parameters to improve BaBSim.Hospital using a surrogate-based optimization approach and an in-depth sensitivity analysis.
References: https://doi.org/10.1145/3449726.3459473