Events and Information

Friday, September 25, 2020 at 12pm

Designing Virtual Humans to Facilitate Learning: What Really Matters? by Noah Schroeder, PH.D.

Wright State University

Abstract: 
For the past 20 years, researchers have incorporated virtual characters, including human-like representations known as virtual humans, into learning environments aiming to facilitate learning. Recently, the field has been revitalized as technologies incorporating virtual humans have become more common in everyday life. Researchers now often examine how to most effectively design virtual humans for different contexts. One of these lines of research focuses on how virtual humans are perceived by learners. Theorists suggest that how we perceive these characters can influence the learning process, but it is not yet known which perceptions matter. Our recent experimental work has shown that the existing measures commonly used by researchers to evaluate the construct of "agent persona" are poor predictors of learning. This talk will focus on what virtual humans are, why they can be effective learning tools, and the findings of recent studies around the design of these systems.

Friday, October 16, 2020 at noon

Explainable Random Forests by Lucas Mentch, ph.d.

University of Pittsburgh

Abstract: 

Random forests remain among the most popular off-the-shelf supervised machine learning tools with a well-established track record of predictive accuracy in both regression and classification settings.  Despite their empirical success, a full and satisfying explanation for their success has yet to be put forth. In this talk, we will show that the additional randomness injected into individual trees serves as a form of implicit regularization, making random forests an ideal model in low signal-to-noise ratio (SNR) settings. From a model-complexity perspective, this means that the mtry parameter in random forests serves much the same purpose as the shrinkage penalty in explicit regularization procedures like the lasso.  Realizing this, we demonstrate that alternative forms of randomness can provide similarly beneficial stabilization.  In particular, we show that augmenting the feature space with additional features consisting of only random noise can substantially improve the predictive accuracy of the model.  This surprising fact has been largely overlooked in both the statistics and ML communities, but has crucial implications for thinking about how to objectively define and measure variable importance.  Numerous demonstrations on both real and synthetic data are provided.