- Author:
- Samuel Stanton,
- Advisor:
- Wilson, Andrew G,
- Committee Members:
- Cho, Kyunghyun,
- Pinto, Lerrell,
- Fergus, Rob,
- Peherstorfer, Benjamin
The study of machine learning is concerned with the development and analysis of general-purpose programs which receive data as input, extract useful patterns and statistics, and automatically modify their output accordingly.The purpose of many programs is to make consistent, justifiable decisions in a timely manner based on as much relevant information as possible.In this work we will focus particularly on programs that operate online, meaning programs whose output is transformed into more information and fed back in as input to create a sequence of conditional decisions.Online programs are particularly useful for applications involving a great deal of complexity or uncertainty, since they can break down difficult planning problems into easier steps, and can collect additional information as needed.We will discuss techniques to efficiently update specific statistical models on infinite streams of data, coherent data collection strategies to optimize program outputs for arbitrary objectives, and means for turning imperfect models into reliable, trustworthy decisions based on provably valid predictions.In addition to fundamental contributions to the body of machine learning methods, we will also present exemplar applications including public health surveillance, control of mechanical systems, and experimental design for scientific discovery.
Index Terms
- Probabilistic Machine Learning for Online Decision-Making
Recommendations
Decision making process: typology, intelligence, and optimization
Decision making is concerned with evaluating and/or ranking possible alternatives of action. In this paper, we develop a model for the process of decision making. Understanding the decision process can provide insights into how humans make decisions, ...
Machine learning advice in managerial decision-making: The overlooked role of decision makers’ advice utilization
Highlights- We identify technical and social aspects that affect decision makers’ utilization of advice that results from ML analysis.
- Information quality and transparency of ML advice can foster advice utilization.
- A decision maker’s trust in ...
AbstractMachine learning (ML) analyses offer great potential to craft profound advice for augmenting managerial decision-making. Yet, even the most promising ML advice cannot improve decision-making if it is not utilized by decision makers. We therefore ...
Sustainable decision making: the role of decision support systems
Sustainable decision making stands for decision making which contributes to the transition to a sustainable society. It raises a number of challenging problems for which existing decision support systems (DSS) may not be equipped. The role of DSS in ...