Communications of the ACM
How Accurate Is Your AI?
Credit: Kyoto University
A researcher at Kyoto University in Japan has developed a new technique that evaluates an artificial intelligence program's ability based on the nature of its input data.
In typical AI development, a performance evaluation is trusted if there is an equal number of positive and negative results, and data biased toward either value means the current system of evaluation will distort the system's ability. "The novelty of this technique is that it doesn't depend on any one type of AI technology, such as deep learning," says Kyoto's J.B. Brown. "It can help develop new evaluation metrics by looking at how a metric interplays with the balance in predicted data. We can then tell if the resulting metrics could be biased."
Brown's work breaks down the AI utilization and analyzes the nature of the statistics used for reporting an AI's ability, while also producing a probability of the performance level, given evaluation data.
From Kyoto University
View Full Article
Abstracts Copyright © 2018 Information Inc., Bethesda, Maryland, USA
No entries found