Véronique Stéphan

Learn More
| We present a neural eld approach to distributed Q-learning in continuous state and action spaces that is based on action coding and selection in dynamic neural elds. It is, to the best of our knowledge, one of the rst attempts that combines the advantages of a topological action coding with a distributed action-value learning in one neural architecture.(More)
We present a rapidly learning neural control architecture for sensory-based navigation of a mobile robot and compare the learning dynamics and the navigation behavior in the context of diierent implemented network approaches and learning schemes. Our control architecture is a combination of i) alternative vector quantiza-tion techniques (Neural gas and(More)
The basic idea o f our anticipatory approach t o perception is t o avoid the common separation o f perception and generation o f behavior and t o fuse both aspects into a consistent neural process. Our approach is based on the prediction o f the consequences o f hypothetically executed actions. In this sense, perception o f space and shape is assumed t o be(More)
This contribution presents an architecture, which enables collision-avoiding robot navigation in indoor-environments based on monocular visual information. This is achieved by neural linkage of image features with depth information, which can be extracted from the stream of visual data. A first approach only segments camera images into obstacle and passable(More)
Dealing with large volumes of data, OLAP data cubes aggregated values are often spoiled by errors due to missing values in detailed data. This paper suggests to adjust aggregate answers, noticing that non-missing values constitute a biased sample of the true result of the query. Using basic random sampling theory, we show that two different problems can be(More)
  • 1