Abstract:
|
Deep neural networks are a powerful method for automatically learning distributed representations at multiple levels of abstraction. Over the past decade, they have dramatically pushed forward the state-of-the-art in any domains including neuroimaging data analysis. Reinforcement Learning is a family of approaches for developing systems that learn optimal behaviour through interaction with an environment. In recent years, reinforcement learning has seen success as an essential component of Deep Reinforcement Learning, which has helped AI researchers achieve previously unheard of results in games like Go and in the development of autonomous vehicles. It also gains great attention in neuroimaging data analysis. We will discuss currently popular neuroimaging applications of deep learning and reinforcement learning, analyze their strength and point out issues for potential improvements. We will further talk about the big picture of how deep learning and reinforcement learning can help shape neuroimaging data analysis in the near future academically and industrially.
|