Main talk (25 minutes + 5 minutes for Q&A)
Speaker: Salman Saeed Khan salman.saeed.khan91@gmail.com @
Title: Bayesian Modelling for Data-Scarce Problems: Integrating Priors and Generative Inference with STAN
Abstract: Probabilistic modelling offers a principled framework for decision-making under uncertainty, particularly in domains where data is limited, noisy, or expensive to acquire. This presentation explores how Bayesian methods—implemented via STAN and PyMC—facilitate the integration of domain expertise through informative priors, thereby enhancing model robustness, interpretability, and generalisability in data-scarce settings.
In contrast to classical approaches that often falter in the face of sparse data, Bayesian models explicitly quantify uncertainty in both parameters and predictions, supporting more reliable and transparent inference. These models also act as generative engines, capable of simulating plausible data and exploring counterfactual scenarios—an essential capability in high-stakes contexts such as healthcare (e.g., predicting drug efficacy with limited trials) and industrial systems (e.g., detecting rare faults).
Through real-world case studies, the presentation will demonstrate practical workflows for model specification, prior elicitation, and posterior diagnostics. The session offers actionable insights for practitioners and researchers aiming to leverage Bayesian methods in constrained data environments.
Recording consent: Yes
Publishing slides consent: Yes
Availability:
Special requirements: no
Submitted 04/04/2025 12:13:20 via PyData London - Submit a Talk
Main talk (25 minutes + 5 minutes for Q&A)
Speaker: Salman Saeed Khan salman.saeed.khan91@gmail.com @
Title: Bayesian Modelling for Data-Scarce Problems: Integrating Priors and Generative Inference with STAN
Abstract: Probabilistic modelling offers a principled framework for decision-making under uncertainty, particularly in domains where data is limited, noisy, or expensive to acquire. This presentation explores how Bayesian methods—implemented via STAN and PyMC—facilitate the integration of domain expertise through informative priors, thereby enhancing model robustness, interpretability, and generalisability in data-scarce settings.
In contrast to classical approaches that often falter in the face of sparse data, Bayesian models explicitly quantify uncertainty in both parameters and predictions, supporting more reliable and transparent inference. These models also act as generative engines, capable of simulating plausible data and exploring counterfactual scenarios—an essential capability in high-stakes contexts such as healthcare (e.g., predicting drug efficacy with limited trials) and industrial systems (e.g., detecting rare faults).
Through real-world case studies, the presentation will demonstrate practical workflows for model specification, prior elicitation, and posterior diagnostics. The session offers actionable insights for practitioners and researchers aiming to leverage Bayesian methods in constrained data environments.
Recording consent: Yes
Publishing slides consent: Yes
Availability:
Special requirements: no
Submitted 04/04/2025 12:13:20 via PyData London - Submit a Talk