Title: Goal-Oriented Learning
Abstract: Despite the wide-spread applications of machine learning across many different areas, most of the current learning algorithm designs do not have the capability to align with the final goal of the application. This discrepancy between the optimized objective in learning algorithms and the original goal of the applications leads to inferior results of current machine learning applications in many real-world tasks. In this talk, I will highlight the sources of the problem that causes this discrepancy and present an alternative framework to overcome the problem. I will then discuss several applications of the framework for designing goal-oriented learning algorithms across different areas of machine learning. Finally, I will discuss the future potential of the paradigm to transform the culture machine learning application as well as exciting future research directions.
Bio: Rizal Fathony is a postdoctoral fellow in the School of Computer Science at Carnegie Mellon University hosted by Prof. Zico Kolter. He received his Ph.D. in Computer Science from the University of Illinois at Chicago in 2019 under the supervision of Prof. Brian Ziebart. He was awarded a Fulbright scholarship from the U.S. Department of State in 2012 -2014 for his Master's study. His research focuses on designing machine learning algorithms that align with the final goal of application deployment. His research interests span across different areas in machine learning theory and applications. He has published his research in top-tier machine learning conferences such as NeurIPS, ICML, AISTATS, and AAAI.