We want understand how neural systems perform visual perception. At the interface of computer vision and neuroscience, we try to understand both, how the human visual system works and how to teach computers to make sense of images. We use an interdisciplinary approach that combines methods from machine learning and computer vision with behavioral studies and neuronal population recordings in the brain. Our work is driven by the idea that we can advance artificial intelligence by understanding how biological systems implement intelligent behavior.
We build upon recent advances in Deep Learning to build predictive models and understand the nonlinear computations of the early visual system.
We work on one-shot learning in the context of object detection and segmentation.
We developed an algorithm based on deep learning that allows us to redraw a photograph in the style of an arbitrary painting with remarkable quality.
We developed a photorealistic parametric texture model and use it to probe the limits of human perception.
We investigate the role of top-down signals in perceptual inference tasks. In particular, we are interested in how sensory signals (bottom-up) and prior beliefs (top-down) interact in the primary visual cortex of behaving monkeys.
We study how populations of neurons in the primary visual cortex of monkeys and mice encode visual information. In particular, we ask how neuronal variability affects population coding.