Yield learning and optimization are critical for advanced IC design and manufacturing. Recent advance in machine learning has brought a lot of new opportunities in improving the performance and efficiency of IC yield learning and optimization. This paper surveys some recent results of using various machine learning/deep learning techniques for such purpose, including performance modeling under uncertainty, lithography modeling with transfer/active learning, lithography hotspot detection, and IC mask optimization. The state-of-the-art methods are explained, and challenges/opportunities are discussed.