Applications and Challenges of Deep Learning in Heliophysics Deep learning (DL) refers to a class of machine learning (ML) techniques that exploit many layers of nonlinear information processing that can approximate very complex mathematical functions/relationships. These function approximators have been found to outperform both classical ML techniques and humans in difficult tasks of data transformation and classification. The most exciting developments in DL are generally tied to their ability to fool, amaze, or simplify the lives of human consumers. However, DL applications to the physical sciences have more stringent quantitative requirements than the subjective human experience. In this talk we will discuss three applications of deep learning to heliophysics (instrument calibration, forecast of solar energetic particles, and detection of magnetic flux emergence), how they relate to commercial applications, and the critical role of uncertainty estimates in transforming them from a mere curiosity into a powerful scientific tool.