Programs

Computational Neuroscience

After decades of exponential growth in power, existing computer architectures still fail to match the ability of the mammalian brain to interpret, respond to, and learn from natural sensory inputs. Rapid progress in neuroscience suggests an alternative strategy for achieving brain-like behavior: identifying the computational primitives that underlie the processing in biological neural circuits. We develop high-performance neural simulation tools and to use them to find these primitives, which make the brain so much more powerful than the familiar von Neumann computer or artificial neural networks (ANNs).

Mapping Sustainability Science

Although Sustainability Science is by now widely discussed in the scientific community, and is beginning to be connected to the political agenda for economic and social development, it remains unclear to what extent its many facets are being integrated into a global perspective and whether researchers are utilizing it as a nexus to collaborate across traditional scientific and technological fields. We address these issues by delineating, analyzing, and mapping both basic research and technology on sustainability.

Segmenting Hyperspectral Scenes with Occlusions

We develop new models for segmentation hyperspectral scenes with occlusions and, particularly, the problem of target identification in these scenes. The approach is based on modeling the appearance of three-dimensional scenes with backgrounds, objects of interest, and possible occluders. We also consider novel invariant shape priors that can be used to guide segmentation and detection processes towards user specified objects of interest.