Phil Roth presented the results of a malware classification model bakeoff between several machine learning algorithms. The models evaluated were k-nearest neighbors, logistic regression, support vector machines, naive Bayes, random forests, gradient boosted decision trees, and deep learning. Based on the performance, size, and query time metrics, gradient boosted decision trees had the best overall results. However, the presenter noted that deep learning approaches deserve more research due to their potential to learn directly from file content. The conclusions were that gradient boosted decision trees could be deployed to endpoints while larger deep learning models could be used in the cloud after further development.