Recent research from Bar-Ilan University asks the basic question: “Can deep learning systems achieve much above-average confidence for a large portion of their inputs while maintaining an overall average confidence?” By answering the question, we have reached a turning point in the field of artificial intelligence (AI).
The results of this study answer this question with an unequivocal “YES” and demonstrate a significant increase in AI’s ability to recognize and accommodate different levels of confidence in classification tasks. . By leveraging deep architecture's trust level insights, the research team opens new avenues for real-world applications ranging from self-driving cars to medicine.
A team of researchers led by Professor Ido Kanter from the Department of Physics at Bar-Ilan University and the Gonda (Goldschmied) Multidisciplinary Brain Research Center published their findings in the following paper. Physical A.
Understanding the trust level of AI systems allows you to develop applications that prioritize safety and reliability. For example, in the context of self-driving cars, if the identification of road signs is very reliable, the system can make decisions autonomously. However, in scenarios where the trust level is low, the system prompts human intervention to ensure careful and informed decision-making.
Ella Koresh, Bar-Ilan University Graduate Student
Improving the reliability of AI systems has far-reaching implications, from AI-based writing and image recognition to critical decision-making processes in healthcare and self-driving cars. This research raises the bar for AI performance and safety by enabling AI systems to make better and more reliable decisions in the face of ambiguity.
Highly reliable AI applications in self-driving cars
Highly reliable AI applications in self-driving cars. Video credit: Bar-Ilan University
Reference magazines:
Meir, Y. other. (2024) Advanced trust methods in deep learning. Physical A. doi:10.1016/j.physa.2024.129758
Source: https://www.biu.ac.il/en
