Michael Eickenberg (Berkeley) Bio: Michael Eickenberg is currently in the Gallant lab at UC Berkeley, he works with single-neuron recordings from visual area V4 and real-time fMRI experiments, as well as trying to understand what deep learning does. Michael describes himself as an applied mathematician with scientific Python skills, who became a scientific Python specialist with applied mathematics skills.

Title: Machine Learning Representations for Imaging Neuroscience and Quantum Chemistry

Abstract: Machine learning can be applied to scientific data streams with surprising utility, such as gains with respect to traditional methods in processing speed or accuracy. In some cases they also directly yield scientific insight. One example is the correspondence of object-recognition convolutional network layer representations to the visual processing hierarchy in the human brain. Another is the recent developments in quantum chemistry, where DFT-level accuracy for estimation of chemical properties in small molecules has been attained and surpassed by machine-learning approaches trained on data bases of previously computed properties. I will present my work in both fields, which involves simple regression methods applied to carefully chosen data representations.