Top News Of The Week
A examine by researchers at the universities of California, Berkeley and Texas at Austin highlights how off-label use of datasets can inject bias into synthetic intelligence algorithms. UC Berkeley’s Michael Lustig said the researchers traced their failure to duplicate the results of a medical imaging examine to a preprocessed dataset used to coach the algorithm. The group processed uncooked images using two common data-processing pipelines that impression many open-access magnetic resonance imaging databases—commercial scanner software and knowledge storage with JPEG compression. It skilled three image reconstruction algorithms on these datasets, then quantified the accuracy of the reconstructed images versus the…Continue Reading
