Thorsten Suttorp

Learn More
Randomized direct search algorithms for continuous domains, such as evolution strategies, are basic tools in machine learning. They are especially needed when the gradient of an objective function (e.g., loss, energy, or reward function) cannot be computed or estimated efficiently. Application areas include supervised and reinforcement learning as well as(More)
This paper presents an architecture for real-time vanishing point estimation for driver assistance applications. It consists of a data-driven estimation and a model-based filtering module. The data-driven estimation algorithm is based on line-segments that are assumed to be calculated in an independent preprocessing stage. Model-based filtering is achieved(More)
First, the covariance matrix adaptation (CMA) with rank-one update is introduced into the (1+1)-evolution strategy. An improved implementation of the 1/5-th success rule is proposed for step size adaptation, which replaces cumulative path length control. Second, an incremental Cholesky update for the covariance matrix is developed replacing the(More)
Designing supervised learning systems is in general a multi-objective optimization problem. It requires finding appropriate trade-offs between several objectives, for example between model complexity and accuracy or sensitivity and specificity. We consider the adaptation of kernel and regularization parameters of support vector machines (SVMs) by means of(More)
The multi-objective covariance matrix adaptation evolution strategy (MO-CMA-ES) combines a mutation operator that adapts its search distribution to the underlying optimization problem with multicriteria selection. Here, a generational and two steady-state selection schemes for the MO-CMA-ES are compared. Further, a recently proposed method for(More)
Computer vision for object detection often relies on complex classifiers and large feature sets to achieve high detection rates. But when real-time constraints have to be met, for example in driver assistance systems, fast classifiers are required. Here we consider the design of a computationally efficient system for pedestrian detection. We propose an(More)
This paper presents a framework for learning of system parameters for vision-based lane detection systems. Learning is achieved by ground-truth data based optimization of a performance measure evaluated on video sequences. Different options for evaluating the performance of lane detection systems are discussed, and in order to allow for a linear(More)
  • 1