Machine Learning: Enabling and Enabled by Advances in Storage and Memory Systems

Machine learning is becoming an important tool for analyzing and enabling storage/memory systems. At the same time, it also needs new innovations in storage/memory systems to store its increasingly large models reliably, and to run its increasingly costly models efficiently. This paper reviews some recent topics on the interactions between machine learning and storage/memory systems. They range from supervised learning to unsupervised and generative learning, from magnetic recording to 2D and 3D flash memories, from analog error-correcting codes for compute-in-memory to binary codes for protecting trained deep learning models, and from storage channel modeling, bit-error rate prediction, signal detection to symbolic regression. The continuation of research in the area can lead to deeper synergy between AI and storage/memory for more scientific and engineering discoveries.