Current AIs are very accurate but inflexible at image recognition. Exactly why this is remains a mystery. Researchers have developed a method called ‘Raw Zero-Shot’ to assess how neural networks handle elements unknown to them. The results have the potential to help researchers identify the common features that make neural networks ‘non-robust,’ and develop methods to make AIs more reliable.
Breaking AIs to make them better: Researchers investigate ways to make AIs more robust by studying patterns in their answers when faced with the unknown
