Math & Stats Colloquium Series - Oct. 10, 2025 (Dr. Dave Campbell)

Speaker: Dr. Dave Campbell (Carleton University)

Title:Turning Text into Evidence: Assessing Random Seeds and Barley Yields in Statistical Language Processing Without Breaking the Carbon Budget

Abstract: 

Algorithmic data science for text has thrived on producing tools for categorizing, summarizing, correcting, and generating content.  Although many of those tools are based on probabilistic models, their utility has not proliferated as far into inferential data science owing to their lack of stability.  Models for text are rife with multimodal likelihoods and ridges, some of which are caused by equivalence through label switching or rotations of the latent space.  Consequently model output depends as much on the algorithm’s random seed as it doesn’t on sampling variability.  Questions about the impact of covariates on the topics of discourse do not have statistical validity when suffering from a lack of stability.  This talk provides an overview of text tools that might be useful for statistical inference.  For analysts to make use of text, it must first be converted into numeric or categorical data.  From there statistical tools can be developed.  The talk covers both of these two main topics, data transformation and inference for text.  

Using an example related to following the balance of discourse around natural resource supply and demand,  the talk will address whether analysts working on problems from economics should fit their own models or recycle a general purpose text model when balancing accuracy, carbon cost, and data privacy.  Then the talk highlights new statistical tools for testing hypotheses and producing covariate effects for assessing whether or not beers have regional flavour differences, perhaps owing to the local production of ingredients.

Bio: 

Dr. Dave Campbell is a Professor in the School of Mathematics and Statistics and the School of Computer Science at Carleton University and President of the Business and Industrial Statistics Section of the Statistical Society of Canada.  Academically, he runs a collaborative team researching inferential algorithms at the intersections of statistics with machine learning, computing, natural language processing, and applied mathematics to solve problems inspired by industry and government collaborations. 

Dave’s career path maintains a theme of Industrial collaborations. He spent 2021-2023 on leave from academia, leading the inferential Data Science team at the Bank of Canada overseeing projects relating to cybersecurity, forecasting banknote demand, understanding drivers of inflation, and ensuring data privacy. Before moving to Carleton University in 2019, Dave was a Professor at Simon Fraser University, where he led the creation one of Canada’s first Bachelors of Data Science degrees. He was the inaugural President of the Data Science and Analytics Section of the Statistical Society of Canada and was a co-organizer of the popular Vancouver Learn Data Science Meetup linking industry and academia.