Some Challenges and Solutions in Data-Driven AI

摘要

Owing to the availability of massive amounts of multimodal data and the reemergence of neural network-based methods, data-driven artificial intelligence (AI) appears to be on the verge of becoming the dominant technology of the future. While this development has basically shaken many established fields such as computer vision, defense, medicine, natural language processing, robotics, and smart transportation, data-driven AI has also raised many concerns about robustness, vulnerability to adversarial attacks, and bias. In this chapter, we discuss the challenges posed by these issues and present a summary of recent research efforts that address them.