Evaluating the Ability of Machine Learning Models to Learn

Generalizations of Contextual and Temporal Tasks

University of California, Los Angeles

The ability of machine learning models to generalize tasks and learn concepts without contextual reliance is crucial for increases in speed, efficiency, and overall performance of intelligent systems. When humans learn a concept, they are able to isolate it from its context almost immediately and then generalize to many other situations where the concept might be found. The evaluation of several machine learning models’ ability to learn certain tennis swings within a specific context and generalize to another context is presented. Failure and success cases are discussed as well and compared to human performance in similar scenarios. The original code as well as data used can be found here.

Watch video on YouTube

Click here to open paper

© 2015-2017 Titus Lungu. All rights reserved.

Photo credit: © iStockphotos