Inductive Analysis of Texts with Embeddings

Prof. Dustin Stoltz, Lehigh University

Word or text embeddings are a central component in modern language models, including those powering generative AI. Embeddings represent word meanings as positions in space, where words that are closer together are used in similar contexts or evoke similar concepts -- even if those words never actually co-occur. We navigate the meaning space created by embeddings directly using basic arithmetic, and in doing so, explore how meaning changes overtime or how meaning differs between different collections of texts. 

Dustin Stoltz is an Assistant Professor of Sociology and Cognitive Science at Lehigh University.  He studies a variety of topics in cultural and economic sociology and specializes in computational methods.  Five copies of his recently published book, Mapping Texts: Computational Text Analysis for the Social Sciences (coauthored with Marshall Taylor), will be raffled off to workshop registrants.  

Register here.

Lunch will be provided for registered attendees.

Hosted by the Center for Social Science Scholarship and cosponsored by the Departments of English; Sociology, Anthropology, & Public Health; Modern Languages, Linguistics, & Intercultural Communication; the Division of Information Technology; the Center for Scalable Data and Computational Science; and CGC-SCIPE.

Dr. Reetam Majumder from the University of Arkansas will lead a three-hour short course on AI for statistical analysis from 2:00 to 5:00 pm on Friday, October 31, 2025, in room 104 of UMBC's Math & Psychology building. The course will provide a hands-on introduction to deep learning models using the Keras package in R.

Register here by October 24.

Deep learning is the cornerstone of all current AI models. While they have historically been used extensively in supervised and unsupervised learning problems, deep learning is also seeing increased adoption for statistical inference associated with complex/intractable models, massive datasets, or both. The short course will provide a hands-on introduction to deep learning models using the keras package in R.

The course aims to familiarize statisticians with the entire pipeline for analyzing data, including software setup, data preprocessing, and some common neural network architectures used in predictive modeling for different data types like tables, images, and time series. Participants will also be introduced to deep learning approaches for density estimation and parameter estimation in statistical problems, and an overview of the current state-of-the-science. 

The hands-on sections will be demonstrated on Google Colab, and participants will be able to follow along on their own personal devices. Course materials, including code, data, slides, and resources for installing the Keras and Torch deep learning libraries on R, will be made available on GitHub.