7 September, 12:00, WeH 1327 Learning to Recognize Objects Sebastian Thrun In this latest IPC(*) in a legendary series, I will chat about an application of machine learning to vision. Suppose a learner that learns to recognize objects from color camera images. The central question I seek to answer is whether the learner will generalize better, if it has previously learned to recognize other, related objects. Conceivably, certain aspects (such as invariances wrt translation, rotation, and illumination) are generally important for object recognition, and knowledge about these aspects will improve generalization. Can the learner learn the domain-specific invariances? I will chat about a new approach, which employs articifial neural networks to represent invariances. Invariance knowledge is used to guide generalization using the EBNN algorithm, which has been presented in oh so many IPCs. There will be free pizza. * Incredibly Preliminary Chat