12:30 – Reception, lunch

13:15 – Opening talk by invited speaker Niels Taatgen (researcher RUG)

14:15 – Break

14:30 – Willem van Willigen (VU): “Adaptive Sensor Systems: Design & Applications”

In this talk, we give a gentle introduction to the field of adaptive sensor systems. We illustrate the eld by giving example applications of these systems in several domains: safety, security and mobility. In the safety domain, we describe a system for dynamic evacuation routing; for security, we describe a UAV surveillance scenario; for mobility, we describe safety in cooperative driving scenarios on highways.

15:00 – BioMAV (RU)

The BioMAV research aimed to contribute to research on biologically inspired micro air vehicles in two ways. First it explores a novel repertoire of behavioral modules which can be controlled through finite state machines (FSM). Secondly, elementary movement detectors are combined with edge-detection for object recognition and tracking. Using these methods the team managed to command third place in the IMAV 2011 Indoor Pylon Challenge.

15:30 – Break

15:45 – Coert van Gemeren (UU): “Concept generation for part-based visual object detection”

Introduction. Applying ideas from the theory of conceptual spaces by Gärdenfors, we enhance the meaningfulness of the discriminatively trained part based deformable object model proposed by Felzenszwalb et al. Semantics is added to the model under the assumption that large data sets of images representing a single object class contain shape invariant features, which can be found by measuring the co-occurrence of the parts in a single deformable object model. Parts are fused in the models if the similarity measure of a pair of features exceeds a certain threshold. Because the creation of parts in Felzenszwalb’s model is unsupervised, there is no meaning to the features of the part model. In our implementation semantics is added under the assumption that while the positions of features that are similar in appearance, but have different locations in the object model may vary, the positions that these features occupy in a conceptual space defined by the appearance (ie. the shape quality) of the features can be clustered.
Methods. We use a combination of a Histogram of oriented gradients (HOG) feature detector and a Support vector machine (SVM) to iteratively generate discriminatively trained object models that represent a given class of objects. The model is comprised of a root layer which represents the general shape of an object class. Layered on top of the root layer are several parts that are anchored to the root layer in different positions. The layout of the parts on top of the root layer is deformable to a certain extent to create more robust detection results. The amount of deformation allowed to fit a given object to the model is governed by a deformation penalty.
Results. The models generate state of the art performance in the Pascal VOC2011 challenge. We show that it is possible to fuse similar parts in the object model and that it is possible to find part similarities among different object models. Using these results we generate concepts for the most dominant features occurring in different object models that fit the description of a natural concept, as proposed by Gärdenfors.

16:15 – Mike Farjam (RU): “Greed, Envy, Jealousy. A tool for more efficient resouce management”

Highly social animals like humans developed features such as greed, envy, and jealousy through evolution. Assuming that the concept of envy has already been learned, experiments are performed in an artificial life environment. They show the benefits of envy for a multiagent system and how principles underlying envy can make agents more effective with respect to resource management.
Furthermore they show under which circumstances (such as the population size or the possibility to punish greed) jealousy turns into a useful feature in a multiagent system. Concepts like population size or availability of resources are translated back into real world phenomena to show possible applications of artificial envy. Simulations show that the benefits in resource-management outweigh the costs of having an envy system.

16:45 – End