Regularization in directable environments with application to Tetris -- Talk by J Malte Lichtenberg (University of Bath) Event

Event details

Learning statistical models from small data sets is difficult in the absence of specific domain knowledge. We present a regularized linear regression model called shrinking toward equal weights (or STEW), which benefits from a generic and prevalent form of prior knowledge: feature directions. By adaptively shrinking weights toward each other, STEW tries to find the happy medium between the ordinary least squares solution and an equal-weighting model. We provide theoretical results on the equal-weighting solution that explain how STEW can productively trade-off bias and variance, and how knowledge about feature directions helps to reduce its bias even further. Across a wide range of learning problems, including Tetris, STEW outperformed existing regularized linear models, including ridge regression and the Lasso, when feature directions were known. The model proved to be robust to unreliable (or absent) feature directions, outperforming alternative models under diverse conditions. Our results in Tetris were obtained by using a novel approach to learning in sequential decision environments based on multinomial logistic regression.