Using Multiple Models to Understand Data

In our first experiment, we show that using A human's ability to diagnose errors, gather data, multiple models to identify potential label noise can provide and generate features in order to build better a threefold reduction in the number of spurious examples a models is largely untapped. We hypothesize that practitioner examines. In our second experiment, we show analyzing results from multiple models can help that analyses of multiple models can identify examples that people diagnose errors by understanding are significantly more likely to respond to additional relationships among data, features, and algorithms.