Commonplace software program testing works simply high quality when you recognize your outputs. What occurs when success means “a minimum of 95% accuracy a minimum of 90% of the time”? Worse nonetheless, what occurs when success means “This group favored the evaluation it gave, so it’s high quality for stuff that appears like what got here in yesterday”? This presentation shares some experiences constructing FINRA’s first frameworks, steering, and case research on monitoring machine studying fashions in manufacturing. Upon constructing context round how machine studying fashions are evaluated, the discuss discusses expertise growing and making use of a few of the most promising approaches to monitoring mannequin predictions in addition to information high quality.
Video producer: https://www.associationforsoftwaretesting.org/