Researchers at Stanford Propose a Unified Regression-based Machine Learning Framework for Sequence Models with Associative Memory