Matthew Schlegel

Papers

Total Citations

H-Index

About

Matthew Schlegel is a researcher whose work lies at the intersection of reinforcement learning and continual learning, with a particular focus on developing algorithms that can adapt to non-stationary environments. His most-cited paper, "Meta-Descent for Online, Continual Prediction" (2019, 11 citations), addresses a fundamental challenge in online learning: how to automatically tune step-sizes for vector-valued updates in non-stationary prediction problems. Schlegel demonstrates that vanilla stochastic gradient descent can be significantly improved by scaling updates with appropriately chosen step-size vectors, and he investigates meta-descent methods—including AdaGrad—for this purpose. This work is notable for its practical implications in continual learning settings where models must adapt to shifting data distributions without forgetting past knowledge. Schlegel’s contributions are particularly relevant to researchers working on lifelong learning systems, online prediction, and adaptive optimization, offering insights into how algorithms can maintain performance over extended periods of interaction with changing environments.