Solving Deep Reinforcement Learning Benchmarks with Linear Policy Networks

Although Deep Reinforcement Learning (DRL) methods can learn effective policies for challenging problems such as Atari games and robotics tasks, algorithms are complex and training times are often long. This study investigates how evolution strategies (ES) perform compared to gradient-based deep reinforcement learning methods. We use ES to optimize the weights of a neural network via neuroevolution, performing direct policy search. We benchmark both regular networks and policy networks consisting of a single linear layer from observations to actions; for three classical ES methods and for three gradient-based methods such as PPO. Our results reveal that ES can find effective linear policies for many RL benchmark tasks, in contrast to DRL methods that can only find successful policies using much larger networks, suggesting that current benchmarks are easier to solve than previously assumed. Interestingly, also for higher complexity tasks, ES achieves results comparable to gradient-based DRL algorithms. Furthermore, we find that by directly accessing the memory state of the game, ES are able to find successful policies in Atari, outperforming DQN. While gradient-based methods have dominated the field in recent years, ES offers an alternative that is easy to implement, parallelize, understand, and tune.

Computing Star Discrepancies with Numerical Black-Box Optimization Algorithms

The $L_{\infty}$ star discrepancy is a measure for the regularity of a finite set of points taken from $[0,1)^d$. Low discrepancy point sets are highly relevant for Quasi-Monte Carlo methods in numerical integration and several other applications. Unfortunately, computing the $L_{\infty}$ star discrepancy of a given point set is known to be a hard problem, with the best exact algorithms falling short for even moderate dimensions around 8. However, despite the difficulty of finding the global maximum that defines the $L_{\infty}$ star discrepancy of the set, local evaluations at selected points are inexpensive. This makes the problem tractable by black-box optimization approaches. In this work we compare 8 popular numerical black-box optimization algorithms on the $L_{\infty}$ star discrepancy computation problem, using a wide set of instances in dimensions 2 to 15. We show that all used optimizers perform very badly on a large majority of the instances and that in many cases random search outperforms even the more sophisticated solvers. We suspect that state-of-the-art numerical black-box optimization techniques fail to capture the global structure of the problem, an important shortcoming that may guide their future development. We also provide a parallel implementation of the best-known algorithm to compute the discrepancy.

When to be Discrete: Analyzing Algorithm Performance on Discretized Continuous Problems

The domain of an optimization problem is seen as one of its most important characteristics. In particular, the distinction between continuous and discrete optimization is rather impactful. Based on this, the optimizing algorithm, analyzing method, and more are specified. However, in practice, no problem is ever truly continuous. Whether this is caused by computing limits or more tangible properties of the problem, most variables have a finite resolution. In this work, we use the notion of the resolution of continuous variables to discretize problems from the continuous domain. We explore how the resolution impacts the performance of continuous optimization algorithms. Through a mapping to integer space, we are able to compare these continuous optimizers to discrete algorithms on the exact same problems. We show that the standard $(\mu_W, \lambda)$-CMA-ES fails when discretization is added to the problem.

What Performance Indicators to Use for Self-Adaptation in Multi-Objective Evolutionary Algorithms

Parameter control has succeeded in accelerating the convergence process of evolutionary algorithms. Empirical and theoretical studies for classic pseudo-Boolean problems, such as OneMax, LeadingOnes, etc., have explained the impact of parameters and helped us understand the behavior of algorithms for single-objective optimization. In this work, by transmitting the techniques of single-objective optimization, we perform an extensive experimental investigation into the behavior of the self-adaptive GSEMO variants. We test three self-adaptive mutation techniques designed for single-objective optimization for the OneMinMax, COCZ, LOTZ, and OneJumpZeroJump problems. While adopting these techniques for the GSEMO algorithm, we consider different performance metrics based on the current non-dominated solution set. These metrics are used to guide the self-adaption process. Our results indicate the benefits of self-adaptation for the tested benchmark problems. We reveal that the choice of metrics significantly affects the performance of the self-adaptive algorithms. The self-adaptation methods based on the progress in one objective can perform better than the methods using multi-objective metrics such as hypervolume, inverted generational distance, and the number of the obtained Pareto solutions. Moreover, we find that the self-adaptive methods benefit from the large population size for OneMinMax and COCZ.

Benchmarking Algorithms for Submodular Optimization Problems Using IOHProfiler

Submodular functions play a key role in the area of optimization as they allow to model many real-world problems that face diminishing returns. Evolutionary algorithms have been shown to obtain strong theoretical performance guarantees for a wide class of submodular problems under various types of constraints while clearly outperforming standard greedy approximation algorithms. This paper introduces a setup for benchmarking algorithms for submodular optimization problems with the aim to provide researchers with a framework to enhance and compare the performance of new algorithms for submodular problems. The focus is on the development of iterative search algorithms such as evolutionary algorithms with the implementation provided and integrated into IOHprofiler which allows for tracking and comparing the progress and performance of iterative search algorithms. We present a range of submodular optimization problems that have been integrated into IOHprofiler and show how the setup can be used for analyzing and comparing iterative search algorithms in various settings.

Optimizing Stimulus Energy for Cochlear Implants with a Machine Learning Model of the Auditory Nerve

Performing simulations with a realistic biophysical auditory nerve fiber model can be very time consuming, due to the complex nature of the calculations involved. Here, a surrogate (approximate) model of such an auditory nerve fiber model was developed using machine learning methods, to perform simulations more efficiently. Several machine learning models were compared, of which a Convolutional Neural Network showed the best performance. In fact, the Convolutional Neural Network was able to emulate the behavior of the auditory nerve fiber model with extremely high similarity ($R^2 > 0.99$), tested under a wide range of experimental conditions, whilst reducing the simulation time by five orders of magnitude. In addition, we introduce a method for randomly generating charge-balanced waveforms using hyperplane projection. In the second part of this paper, the Convolutional Neural Network surrogate model was used by an Evolutionary Algorithm to optimize the shape of the stimulus waveform in terms energy efficiency. The resulting waveforms resemble a positive Gaussian-like peak, preceded by an elongated negative phase. When comparing the energy of the waveforms generated by the Evolutionary Algorithm with the commonly used square wave, energy decreases of 8% - 45% were observed for different pulse durations. These results were validated with the original auditory nerve fiber model, which demonstrates that our proposed surrogate model can be used as its accurate and efficient replacement.

Per-run Algorithm Selection with Warm-starting using Trajectory-based Features

Per-instance algorithm selection seeks to recommend, for a given problem instance and a given performance criterion, one or several suitable algorithms that are expected to perform well for the particular setting. The selection is classically done offline, using openly available information about the problem instance or features that are extracted from the instance during a dedicated feature extraction step. This ignores valuable information that the algorithms accumulate during the optimization process. In this work, we propose an alternative, online algorithm selection scheme which we coin per-run algorithm selection. In our approach, we start the optimization with a default algorithm, and, after a certain number of iterations, extract instance features from the observed trajectory of this initial optimizer to determine whether to switch to another optimizer. We test this approach using the CMA-ES as the default solver, and a portfolio of six different optimizers as potential algorithms to switch to. In contrast to other recent work on online per-run algorithm selection, we warm-start the second optimizer using information accumulated during the first optimization phase. We show that our approach outperforms static per-instance algorithm selection. We also compare two different feature extraction principles, based on exploratory landscape analysis and time series analysis of the internal state variables of the CMA-ES, respectively. We show that a combination of both feature sets provides the most accurate recommendations for our test cases, taken from the BBOB function suite from the COCO platform and the YABBOB suite from the Nevergrad platform.

Trajectory-based Algorithm Selection with Warm-starting

Landscape-aware algorithm selection approaches have so far mostly been relying on landscape feature extraction as a preprocessing step, independent of the execution of optimization algorithms in the portfolio. This introduces a significant overhead in computational cost for many practical applications, as features are extracted and computed via sampling and evaluating the problem instance at hand, similarly to what the optimization algorithm would perform anyway within its search trajectory. As suggested in Jankovic et al. (EvoAPPs 2021), trajectory-based algorithm selection circumvents the problem of costly feature extraction by computing landscape features from points that a solver sampled and evaluated during the optimization process. Features computed in this manner are used to train algorithm performance regression models, upon which a per-run algorithm selector is then built. In this work, we apply the trajectory-based approach onto a portfolio of five algorithms. We study the quality and accuracy of performance regression and algorithm selection models in the scenario of predicting different algorithm performances after a fixed budget of function evaluations. We rely on landscape features of the problem instance computed using one portion of the aforementioned budget of the same function evaluations. Moreover, we consider the possibility of switching between the solvers once, which requires them to be warm-started, i.e. when we switch, the second solver continues the optimization process already being initialized appropriately by making use of the information collected by the first solver. In this new context, we show promising performance of the trajectory-based per-run algorithm selection with warm-starting.

IOHexperimenter: Benchmarking Platform for Iterative Optimization Heuristics

We present IOHexperimenter, the experimentation module of the IOHprofiler project, which aims at providing an easy-to-use and highly customizable toolbox for benchmarking iterative optimization heuristics such as evolutionary and genetic algorithms, local search algorithms, Bayesian optimization techniques, etc. IOHexperimenter can be used as a stand-alone tool or as part of a benchmarking pipeline that uses other components of IOHprofiler such as IOHanalyzer, the module for interactive performance analysis and visualization. IOHexperimenter provides an efficient interface between optimization problems and their solvers while allowing for granular logging of the optimization process. These logs are fully compatible with existing tools for interactive data analysis, which significantly speeds up the deployment of a benchmarking pipeline. The main components of IOHexperimenter are the environment to build customized problem suites and the various logging options that allow users to steer the granularity of the data records.

Explorative Data Analysis of Time Series based AlgorithmFeatures of CMA-ES Variants

In this study, we analyze behaviours of the well-known CMA-ES by extracting the time-series features on its dynamic strategy parameters. An extensive experiment was conducted on twelve CMA-ES variants and 24 test problems taken from the BBOB (Black-Box Optimization Bench-marking) testbed, where we used two different cutoff times to stop those variants. We utilized the tsfresh package for extracting the features and performed the feature selection procedure using the Boruta algorithm, resulting in 32 features to distinguish either CMA-ES variants or the problems. After measuring the number of predefined targets reached by those variants, we contrive to predict those measured values on each test problem using the feature. From our analysis, we saw that the features can classify the CMA-ES variants, or the function groups decently, and show a potential for predicting the performance of those variants. We conducted a hierarchical clustering analysis on the test problems and noticed a drastic change in the clustering outcome when comparing the longer cutoff time to the shorter one, indicating a huge change in search behaviour of the algorithm. In general, we found that with longer time series, the predictive power of the time series features increase.