Title: Generalization Error from Approximate Probability Propagation Method
Speaker: Ms. Ayaka Sakata (Associate Professor, The Institute of Statistical Mathematics)
Abstract: Approximate probability propagation methods are widely used to efficiently approximate inference in large-scale stochastic models. The method simplifies computation by assuming a local tree structure and evaluating the marginalized distribution based on a graphical representation of the stochastic model. The procedure can be interpreted as implicitly including leave-one-out cross-validation error evaluation. In linear models, the expression of leave-one-out cross-validation error can be obtained analytically, but using an interpretation based on the approximate probability propagation method, this expression can be extended to a more general framework. In this presentation, the approximate probability propagation method and its statistical implications will be explained, and various estimator evaluation methods for generalization errors, including leave-one-out cross-validation errors, will be presented.
- Date
-
21st March, 2025（Fri,）18:00～20:00
- Venue
-
Held online
- Organizer
-
Co-organizer (HRAM The Japan Society for Industrial and Applied Mathematics, D-DRIVE National Network)
- Participation Fee
-
Free（Advance registration required）
https://www-mmds.sigmath.es.osaka-u.ac.jp/structure/activity/ai_data.php?id=89
- web
-
https://www-mmds.sigmath.es.osaka-u.ac.jp/structure/activity/ai_data.php?id=89
- Contact
-
Takashi Suzuki
suzuki@sigmath.es.osaka-u.ac.jp