Keynote Speakers

More information regarding our invited speakers will be announced in this page as it becomes available.

William Edward Hahn (Florida Atlantic University)

Title: Compressed Inference: Machine Learning on Random Projections

In this talk we will discuss the theoretical foundation that make compressed sensing possible and how these ideas can be extended to generate novel machine learning architectures. In many signal processing tasks the end goal is classification and reconstruction is unnecessary. Here we demonstrate how off the shelf machine learning algorithms can learn from randomly compressed input vectors. Random projections are universal with respect to sparsifying basis which can simplify network design. Randomly compressed input vectors are holographic in the sense that information in the original vector is evenly distributed throughout the new vector. This suggests a new role for untrainable layers in deep learning architectures.

Lisa Hellerstein (New York University)

Title: Sequential Ordering of Binary Tests in Stochastic Environments

There are many tasks which involve performing a sequence of tests or probes, or asking a sequence of queries. Examples include performing medical tests to diagnose a patient's disease, asking questions when interviewing a job candidate, performing probes to determine s-t connectivity in a network, and searching a sequence of locations for a hidden object. The cost or time spent in ``testing'' can vary greatly depending on the order in which tests are performed. In this talk I will survey work on algorithms for sequential test ordering problems in a stochastic environment, where test outcomes are binary and the goal is to minimize expected testing cost. The talk will cover techniques used in developing exact and approximation algorithms, and include connections to work on submodularity, search games, and Boolean functions. Open problems will be highlighted.

Aaron Roth (University of Pennsylvania)

Title: The Ethical Algorithm

Many recent mainstream media articles and popular books have raised alarms over anti-social algorithmic behavior, especially when driven by machine learning. The concerns include leaks of sensitive personal data by predictive models, algorithmic discrimination as a side-effect of machine learning, and inscrutable decisions made by complex models. While standard and legitimate responses to these phenomena include calls for better laws and regulations, researchers in machine learning, statistics and related areas are also working on designing better-behaved algorithms. An explosion of recent research in areas such as differential privacy, algorithmic fairness and algorithmic game theory is forging a new science of socially aware algorithm design. I will survey these developments and give a taste for what this work is like via technical anecdotes from our own recent research. This talk is based on our book “The Ethical Algorithm”, co-authored with Michael Kearns.

Shlomo Zilberstein (University of Massachusetts, Amherst)

Title: Challenges and New Directions in Planning for Long-Term Autonomy

AI is experiencing a golden age. Scientific breakthroughs and game-changing technologies are rapidly altering the way we live, work, communicate, and entertain. In this talk, I examine the challenges involved in deploying autonomous systems such as self-driving cars that can operate in unstructured environments over extended periods of time. These challenges will be explored through insights from my research on automated planning, autonomous driving, and human-in-the-loop AI. Many obstacles remain. Can AI deliver on the promise of reliable autonomous systems or will we repeat the old pattern of research leaps followed by unmet expectations? To what extent will autonomous systems need to rely on human assistance? How will they obtain assistance when needed and learn to reduce the reliance on humans over time? The ability to sustain progress and responsibly deploy AI technology depends on AI-human collaboration mechanisms and a far better understanding of how humans use commonsense to handle ordinary situations that they encounter every day.