Millicent Li
I'm a first-year PhD student at Northeastern University where I'm advised by Byron Wallace and work closely with David Bau. Though I'm still pinning down my exact research interests, broadly, I'm interested in the intersection of natural language processing and human-computer interaction, particularly in domains related to language modeling, intepretability, and the interaction between humans and AI tools. I'm honored to be supported by a (Northeastern) Khoury PhD Fellowship and an NSF GRFP.
Before my PhD, I spent a year as an AI Resident with Fundamental AI Research (FAIR) at Meta, working with Marjan Ghazvininejad and Mike Lewis and collaborating with Asli Celikyilmaz and Mona Diab on topics related to prompting and pretrained language modeling. I've also interned at Microsoft Research where I worked with Tristan Naumann on building biomedical NLP benchmarks.
In undergrad, I studied computer science at the University of Washington. There, I collaborated with both the DUB and UW NLP research groups, where I worked with Shwetak Patel on ubiquitous computing and Noah Smith on natural language processing, respectively.
News
Our paper, Summarizing, Simplifying, and Synthesizing Medical Evidence using GPT-3 (with Varying Success), was accepted to ACL!
I was awarded a 2022 NSF Graduate Research Fellowship. Northeastern wrote an article about it here.
Started as an AI Resident with Fundamental AI Research (FAIR) at Meta in Seattle, working on natural language processing and human-computer interaction research for a year.
Started my internship at Microsoft Research working with Tristan Naumann on the intersection of natural language processing and healthcare!
Excited to announce that I’ll be starting my PhD in the Khoury College of Computer Sciences at Northeastern University in Boston, fall of 2022. Thanks to everyone who has supported me on this journey thus far!
I was awarded an Honorable Mention for the 2021 NSF Graduate Research Fellowship competition.
Publications
2023
-
Summarizing, Simplifying, and Synthesizing Medical Evidence using GPT-3 (with Varying Success)Annual Meeting of the Association for Computational Linguistics (ACL), 2023HTMLLarge language models, particularly GPT-3, are able to produce high quality summaries of general domain news articles in few- and zero-shot settings. However, it is unclear if such models are similarly capable in more specialized, high-stakes domains such as biomedicine. In this paper, we enlist domain experts (individuals with medical training) to evaluate summaries of biomedical articles generated by GPT-3, given zero supervision. We consider both single- and multi-document settings. In the former, GPT-3 is tasked with generating regular and plain-language summaries of articles describing randomized controlled trials; in the latter, we assess the degree to which GPT-3 is able to \emph{synthesize} evidence reported across a collection of articles. We design an annotation scheme for evaluating model outputs, with an emphasis on assessing the factual accuracy of generated summaries. We find that while GPT-3 is able to summarize and simplify single biomedical articles faithfully, it struggles to provide accurate aggregations of findings over multiple documents. We release all data and annotations used in this work.
2022
-
A Review on Language Models as Knowledge BasesarXiv* denotes equal contribution^ denotes equal supervisionHTML Project WebsiteRecently, there has been a surge of interest in the NLP community on the use of pretrained Language Models (LMs) as Knowledge Bases (KBs). Researchers have shown that LMs trained on a sufficiently large (web) corpus will encode a significant amount of knowledge implicitly in its parameters. The resulting LM can be probed for different kinds of knowledge and thus acting as a KB. This has a major advantage over traditional KBs in that this method requires no human supervision. In this paper, we present a set of aspects that we deem a LM should have to fully act as a KB, and review the recent literature with respect to those aspects.
2020
-
Multi-Channel Facial Photoplethysmography SensingInternational Conference of the IEEE Engineering in Medicine Biology Society (EMBC 2020)HTMLMotivated by the need for continuous cardiovascular monitoring, we present a system for performing photoplethysmography sensing at multiple facial locations. As a proof-of-concept, our system incorporates an optical sensor array into a wearable face mask form factor for application in a surgical hemodynamic monitoring use case. Here we demonstrate that our design can accurately detect pulse timing by validating estimated heart rate against ground truth electrocardiogram recordings. In an experiment across 10 experimental subjects, our system achieves an error standard deviation of 2.84 beats per minute. This system shows promise for performing non-invasive, continuous pulse waveform recording from multiple locations on the face.