Distributed analytics in fog computing is one of the best ways to solve latency, bandwidth limitations, and security protection in smart factories. Many proofs of concept (PoC) have been performed using edge analytics such as machine learning. However, the effective use of distributed analytics in fog computing is a remaining issue. This session shows several industrial use-cases featuring deep, machine, and reinforcement learning, and it addresses the minimum viable usage enabling fog computing. For deep learning, error correction by the 3Ms (men, machines, and materials) based on security by holistic sensing has been introduced for analyzing multiple image data on production lines. A dynamic smart sensing system with data extraction for machine learning is described, using environmental information, such as temperature, humidity, and vibration. In addition, clustering of machine states by machine learning and parameter optimization for the launch of infrastructure by reinforcement learning are discussed. A part of this work was funded by NEDO, the New Energy and Industrial Technology Development Organization, in Japan.