Communications of the ACM
AI Diversity Scoring System Could Help Root Out Algorithmic Bias
Databases used for training AI are often not very diverse.
Credit: fizkes/Shutterstock
Princeton University's Adji Bousso Dieng and Dan Friedman have developed a diversity score for artificial intelligence (AI) to help detect bias.
The researchers were inspired by how ecologists measure biodiversity to create the Vendi Score system, which quantifies the effective number of unique or dissimilar items within datasets or AI systems.
The system calculates the similarity between any two pairs of items in a set and configures the scores in a table of numbers, then computes the randomness of the table's similarity scores into a final diversity score.
The Vendi Score has demonstrated the ability to assess diversity in AI systems and datasets used for AI training.
Dieng suggests diversity scoring also could enhance the performance of AIs used to predict new molecules and materials.
From New Scientist
View Full Article
Abstracts Copyright © 2021 SmithBucklin, Washington, DC, USA
No entries found