In late April, AI scientists around the world gathered in virtual attendance for the International Conference on Learning Representations (ICLR). The conference focuses on the advancement of artificial intelligence, statistics, and data science, as well as areas like computer vision, computational biology, speech recognition, text understanding, gaming, and robotics.
Amazon scholar and UC Berkeley professor, Michael I. Jordan, delivered a keynote talk on the complexities of making AI a field capable of designing planetary-scale systems that can help humans at scale.
“Abstracting the IT systems that currently exist in domains such as commerce, healthcare, or transportation, an AI system involves many humans, computers, data flows, and decisions,” explains Jordan. “An AI system at scale will involve an intricate network of many other machine learning decision-makers that need to work together to minimize the risk of data-aware decision-making.”
Below is Jordan’s keynote presentation, in which he discusses new results in bandits-meet-matching-markets, uncertainty-based Q learning, and anytime false discovery rate. These techniques will help mitigate the emerging challenges of decision making in machine learning, ensuring AI systems can make smarter decisions for the humans that rely on them.
The decision-making side of machine learning | ICLR 2021| Amazon Science
