Parker Seegmiller

Innovation Fellow, PhD Candidate @ Dartmouth CS | Applied Science Intern @ Amazon

Hi! I’m Parker. I’m a fourth year PhD student in the Persist Lab at the Dartmouth College Department of Computer Science and an Innovation Program Fellow in the Dartmouth PhD Innovation Program. I’m also currently an Applied Science Intern at Amazon, working on the Artifical General Intelligence team.

My research at Dartmouth and Amazon has been all things large language models! I’m particularly interested in statistical tools for LLM evaluations, semantic textual similarity/relatedness, and data synthesis.

In addition to my research interests, I have a strong history in data science, sports (basketball/tennis/football) analytics, and teaching.

Interests

NLP
Statistics
ML/DL
Sports Analytics
Business Management
Skyrim

Education

PhD in Computer Science, 2021-
Dartmouth College
BSc in Statistics, 2017-2021
Brigham Young University

Publications

Statistical Depth for Ranking and Characterizing Transformer-Based Text Embeddings

A statistical tool for measuring and comparing distributions of transformer-based text embeddings.

Parker Seegmiller, Sarah Masud Preum

Statistical Depth for Ranking and Characterizing Transformer-Based Text Embeddings

Text Encoders Lack Knowledge: Leveraging Generative LLMs for Domain-Specific Semantic Textual Similarity

STS can be cast as a text generation problem while maintaining strong performance on multiple STS benchmarks.

Joseph Gatto, Omar Sharif, Parker Seegmiller, Philip Bohlman, Sarah Masud Preum

The Scope of In-Context Learning for the Extraction of Medical Temporal Constraints

Medical Temporal Constraints Large Language Models can Extraction Medical Temporal Constraints from Drug Usage Guidelines What Are Medical Temporal Constraints? Medications usually come with specific instructions about when and how to take them.

Parker Seegmiller, Joseph Gatto, Madhusudan Basak, Diane Cook, Hassan Ghasemzadeh, John Stankovic, Sarah Masud Preum

The Scope of In-Context Learning for the Extraction of Medical Temporal Constraints

HealthE: Recognizing Health Advice & Entities in Online Health Communities

Why We Need Better Health Text Datasets The rise of the internet has revolutionized how we access health information. A staggering 89% of patients in the U.S. now google their health symptoms before visiting their doctor.

Parker Seegmiller, Joseph Gatto, Garrett Johnston, Madhusudan Basak, Sarah Masud Preum

HealthE: Recognizing Health Advice & Entities in Online Health Communities

Identifying the Perceived Severity of Patient-Generated Telemedical Queries Regarding COVID: Developing and Evaluating a Transfer Learning-Based Solution

Triage of textual telemedical queries is a safety-critical task for medical service providers with limited remote health resources. The …

Joseph Gatto, Parker Seegmiller, Garrett Johnston, Sarah Masud Preum

See all publications

Blog

Costs of Calculating the Actual Causal Language Conditional Probabilities of English

Estimating the cost of a single actual conditional token probability distribution in causal language modeling in a given language

Parker Seegmiller

Last updated on Jan 3, 2024 8 min read blog

Costs of Calculating the Actual Causal Language Conditional Probabilities of English

Vanilla LLMs Are Not Fit for Medical Diagnosis

Large Language Models (LLMs) are unfit for the complex task of medical diagnosis.

Parker Seegmiller

Last updated on Sep 8, 2023 7 min read blog

Vanilla LLMs Are Not Fit for Medical Diagnosis

Experience

Applied Science Intern

Amazon

Aug 2024 – Present Sunnyvale, CA

Research: Data synthesis for LLM math reasoning

PhD Candidate

Dartmouth CS Dept

Sep 2021 – Present Hanover, New Hampshire

Research: language model evaluation, text embedding distributions, semantic information processing
Teaching: machine learning, discrete math, android programming