Statistical Foundations of Generative Modeling

Description

Generative models have rapidly become a central tool in modern AI and data science. At a high level, a generative model learns an underlying probability distribution from data and can sample from it to create synthetic yet realistic outputs: text, images, financial scenarios, molecules, patient records, climate simulations, and more. Recent advances such as variational autoencoders, generative adversarial networks, normalizing flows, diffusion models, and flow matching have delivered striking empirical performance. At the same time, many of the most pressing questions remain fundamentally statistical: What distribution is being learned, and under what assumptions is it identifiable? Which distributional features are easy or hard to capture (e.g., modes with complex geometry, rare events, and tail behavior)? How can we quantify uncertainty, control bias, and ensure calibration, especially in high-stakes settings where downstream decisions depend on faithful modeling of extremes?

These challenges become even sharper in domain-specific contexts. Open-ended text generation lacks a single “correct” output, making objective evaluation of quality, coherence, diversity, and fluency a critical open problem. In finance and other dependent-data regimes, correlations and selection effects in training data raise questions about generalization and downstream validity. Across application areas, statistics plays a key role in designing reliable metrics to evaluate and compare generative models, and in understanding the properties of common fine-tuning and alignment procedures. More broadly, determining when synthetic data is “good enough” for inference, prediction, or decision-making remains an open question, as do opportunities to use generative models for tasks such as anomaly and changepoint detection.

This workshop brings together statisticians, machine learning researchers, and practitioners from domains including language modeling, finance, biomedicine, and the natural sciences to develop a shared language and research agenda. The goal is to connect modern generative modeling techniques to classical statistical principles, while advancing theory, methodology, and practices that enable reliable deployment in real-world scientific and societal applications.

Registration Fee

A non-refundable registration fee will be payable by credit card or debit card for any participants invited to attend this workshop in-person. In-person participants agree to pay the non-refundable fee by the deadline given by IMSI. Failure to pay the fee by the deadline may mean that the invitation to attend in-person is revoked.

Current fees:

• $25 for students
• $50 for non-students

Registration

IMSI is committed to making all of our programs and events inclusive and accessible. Contact [email protected] to request disability-related accommodations.