{"id":521642,"date":"2022-05-02T14:16:57","date_gmt":"2022-05-02T18:16:57","guid":{"rendered":"https:\/\/www.rochester.edu\/newscenter\/?p=521642"},"modified":"2022-05-02T15:40:47","modified_gmt":"2022-05-02T19:40:47","slug":"how-can-we-be-sure-machine-learning-is-accurate-521642","status":"publish","type":"post","link":"https:\/\/www.rochester.edu\/newscenter\/how-can-we-be-sure-machine-learning-is-accurate-521642\/","title":{"rendered":"How can we be sure machine learning is accurate?"},"content":{"rendered":"

Rochester researchers use \u2018counterfactuals\u2019 to verify predictions of drug safety.<\/h2>\n

Scientists rely increasingly on models trained with machine learning to provide solutions to complex problems. But how do we know the solutions are trustworthy when the complex algorithms the models use are not easily interrogated or able to explain their decisions to humans?<\/p>\n

That trust is especially crucial in drug discovery, for example, where machine learning is used to sort through millions of potentially toxic compounds to determine which might be safe candidates for pharmaceutical drugs.<\/p>\n

\u201cThere have been some high-profile accidents in computer science where a model could predict things quite well, but the predictions weren\u2019t based on anything meaningful,\u201d says Andrew White associate professor of chemical engineering at the 天美传媒<\/a>, in an interview with Chemistry World<\/em><\/a>.<\/p>\n

White and his lab have developed a new \u201ccounterfactual\u201d method, described in Chemical Science<\/em><\/a>, that can be used with any molecular structure-based machine learning model to better understand how the model arrived at a conclusion.<\/p>\n

Counterfactuals can tell researchers \u201cthe smallest change to the features that would alter the prediction,\u201d says lead author Geemi Wellawatte, a PhD student in White\u2019s lab. \u201cIn other words, a counterfactual is an example as close to the original, but with a different outcome.\u201d<\/p>\n

Counterfactuals can help researchers quickly pinpoint why a model made a prediction, and whether it is valid.<\/p>\n

The paper identifies three examples of how the new method, called MMACE (Molecular Model Agonistic Counterfactual Explanations), can be used to explain why:<\/p>\n