Search

We will discuss Mixed-Integer Evolution Strategies (MIES) and their application to the optimization of control parameters of a semi-automatic image analysis system for Intravascular Ultrasound (IVUS) images. IVUS is a technique used to obtain real-time high-resolution tomographic images from the inside of coronary vessels and other arteries. The IVUS image feature detectors used in the analysis system are expert-designed and the default parameters are calibrated manually so far. The new approach, based on MIES, can automatically find good parameterizations for sets of images, which provide in better results than manually tuned parameters. From the algorithmic point of view, the difficulty is designing a blackbox optimization strategy that can deal with nonlinear functions and different types of parameters, including integer, nominal discrete and continuous variables. MIES turns out to be well suited for this task. The results presented in this contribution will summarize and extend recent studies on benchmark functions and the IVUS image analysis optimization problem. [ABSTRACT FROM AUTHOR]

Mixed-Integer Evolution Strategies (MIES) are a natural extension of standard Evolution Strategies (ES) for addressing optimization of various types of variables – continuous, ordinal integer, and nominal discrete – at the same time. Like most Evolutionary Algorithms (EAs), they experience problems in obtaining the global optimum in highly multimodal search landscapes. Niching methods, the extension of EAs to multimodal domains, are designed to treat this issue. In this study we present a dynamic niching technique for Mixed-Integer Evolution Strategies, based upon an existing ES niching approach, which was developed recently and successfully applied to continuous landscapes. The new approach is based on the heterogeneous distance measure that addresses search space similarity in a way consistent with the mutation operators of the MIES. We apply the proposed Dynamic Niching MIES framework to a test-bed of artificial landscapes and show the improvement on the global convergence in comparison to the standard MIES algorithm. [ABSTRACT FROM AUTHOR]

This paper discusses a study towards dynamic fitness based partitioning in IntraVascular UltraSound (IVUS) image analysis. Mixed-Integer Evolution Strategies (MI-ES) have recently been successfully used to optimize control parameters of a multi-agent image interpretation system for IVUS images lumen detection. However, because of complex interpretation contexts, it is impossible to find one single solution which works well on each possible image of each possible patient. Therefore it would be wise to let MI-ES find a set of solutions based on an optimal partition of IVUS images. Here a methodology is presented which does dynamic fitness based partitioning of the data during the MI-ES parameter optimization procedure. As a first step we applied this method to a challenging artificial test case which demonstrates the feasibility of our approach. [ABSTRACT FROM AUTHOR]

Reinforcement learning (RL) agents can benefit from adaptive exploration/exploitation behavior, especially in dynamic environments. We focus on regulating this exploration/exploitation behavior by controlling the action-selection mechanism of RL. Inspired by psychological studies which show that affect influences human decision making, we use artificial affect to influence an agent's action-selection. Two existing affective strategies are implemented and, in addition, a new hybrid method that combines both. These strategies are tested on ‘maze tasks' in which a RL agent has to find food (rewarded location) in a maze. We use Soar-RL, the new RL-enabled version of Soar, as a model environment. One task tests the ability to quickly adapt to an environmental change, while the other tests the ability to escape a local optimum in order to find the global optimum. We show that artificial affect-controlled action-selection in some cases helps agents to faster adapt to changes in the environment. [ABSTRACT FROM AUTHOR]

This paper discusses Mixed-Integer Evolution Strategies and their application to an automatic image analysis system for IntraVascular UltraSound (IVUS) images. Mixed-Integer Evolution Strategies can optimize different types of decision variables, including continuous, nominal discrete, and ordinal discrete values. The algorithm is first applied to a set of test problems with scalable ruggedness and dimensionality. The algorithm is then applied to the optimization of an IVUS image analysis system. The performance of this system depends on a large number of parameters that - so far - need to be chosen manually by a human expert. It will be shown that a mixed-integer evolution strategy algorithm can significantly improve these parameters compared to the manual settings by the human expert. [ABSTRACT FROM AUTHOR]

NK landscapes (NKL) are stochastically generated pseudo-boolean functions with N bits (genes) and K interactions between genes. By means of the parameter K ruggedness as well as the epistasis can be controlled. NKL are particularly useful to understand the dynamics of evolutionary search. We extend NKL from the traditional binary case to a mixed variable case with continuous, nominal discrete, and integer variables. The resulting test function generator is a suitable test model for mixed-integer evolutionary algorithms (MI-EA) - i. e. instantiations of evolution algorithms that can deal with the aforementioned variable types. We provide a comprehensive introduction to mixed-integer NKL and characteristics of the model (global/local optima, computation, etc.). Finally, a first study of the performance of mixed-integer evolution strategies on this problem family is provided, the results of which underpin its applicability for optimization algorithm design. [ABSTRACT FROM AUTHOR]