Keegan Harris

Email: keeganh [at] cs.cmu.edu
Twitter: @keegan_w_harris
Google Scholar
CV

I am a Machine Learning Ph.D. student in the School of Computer Science at Carnegie Mellon University, where I am advised by Nina Balcan and Steven Wu, and supported by the NDSEG Fellowship. I am also the head editor of the ML@CMU blog. Fill out this form if you have a CMU affiliation and are interested in writing a blog post about your research!

In Summer 2023 and Summer 2024 I was a research intern with the Economics and Computation group at Microsoft Research, where I was fortunate to be hosted by Alex Slivkins. I obtained a Master of Science degree in Machine Learning from Carnegie Mellon in 2022. Before coming to CMU, I graduated from Penn State University with Bachelor of Science degrees in Computer Science and Physics.

I am broadly interested in topics at the intersection of machine learning, algorithmic game theory, and econometrics. My goal is to develop principled algorithms for modern data-driven decision-making in online markets. To this end, my research has been focused on three (often interleaved) areas:

  1. Learning and decision-making using panel data: In settings such as e-commerce, content recommendation systems, and clinical trials, one often observes repeated, noisy, measurements from a collection of individuals over a period of time. Such settings are ubiquitous in today's digital world, and are often referred to as panel data settings. I am interested in leveraging the temporal structure and similarity between individuals typically present in panel data to design better algorithms for decision-making in real-world settings. The leap from counterfactual inference to decision making results in additional challenges which must be taken into consideration. To this end, my research has focused on learning from panel data which has been adaptively collected and learning & decision-making in the presence of strategic individuals.
    Representative Publications:
  2. Learning in non-stationary environments: While classical machine learning focuses on learning a single model for a single task using data from a stationary distribution, the real world is often non-stationary and requires reasoning about many different-but-related tasks, often with just a small amount of data from each. For example, in online advertising auctions, the advertiser’s value for different keywords adapts based on current marketing trends. In online marketplaces, the price consumers are willing to pay for different products varies based on the current economic outlook. My research in this area has been focused on developing algorithms for decision-making with provable performance guarantees, which are capable of operating under partial feedback and in the presence of other strategic agents.
    Representative Publications:
  3. Algorithmic decision-making under incentives: When algorithmic assessment tools are used in high-stakes domains such as lending, education, or employment, decision-subjects have an incentive to modify their input to the algorithm in order to receive a more desirable outcome. As a result, machine learning systems deployed in these settings need to take such strategic interactions into consideration in order to make reliable predictions (and decisions). To this end, I have investigated the effects of partial feedback, incomplete information, causal learning, and repeated interactions in the domain of algorithmic decision-making under incentives.
    Representative Publications:

News