Topic area: Modeling
This talk will discuss the current generation of AI methods, and how they differ from previous generation methods. In particular I will discuss algorithms which are based on discriminative/generative models and computational architectures consisting of a hierarchy of concepts known as deep learning.
Creating artifacts exhibiting intelligence has been of interest since the days of Greek mythology. While computer-based artificial intelligence (AI) has been a topic of research for nearly 75 years, their achievements were modest until about a decade ago. This pace has picked up dramatically over the last few years. Researchers in AI now predict that computers will be able to perform tasks that were once considered the prerogative of human beings. They include tasks such as translating languages, writing high school essays, driving trucks, working in retail and even writing a best-selling book or work as a surgeon. Although some of these predictions are predicted to happen over several decades, what are the current principles and algorithms of AI that allow researchers to make such bold predictions? In this talk I will differentiate between the current generation of AI methods and previous generation methods. They will include: knowledge-based systems, simple machine learning methods and deep learning methods. Specific topics such as the use of a hierarchy of concepts, discriminative vs. generative models and disentangling factors of variation will be discussed.