[ИАД, осень 2025] Байесовский выбор моделей. Лекция 1: Введение

Item: [ИАД, осень 2025] Байесовский выбор моделей. Лекция 1: Введение
Rating: 4
Author: PostSingularity Institute

by Alexander Duenko

Formal & Physical Sciences Mathematics PBTProbability and statistics PBTBBayesian inference

🎙 Alexander Duenko 👥 8K 📅 September 11, 2025 ⏱ 88 min 👁 210 📄 lecture 🧭 2026-08-16

Available in: English (current) Français

Keywords

Bayes' theorempriorposteriorhypothesis testingmodel selection

Summary

This is the first lecture of a course on Bayesian model selection, taught by Alexander Duenko. The lecture begins with organizational details, including grading and assignments. The main content reviews Bayes’ theorem and its application to a problem of identifying whether a person is an oracle based on their success in guessing coin tosses. The lecturer emphasizes the role of the prior probability and how it can be influenced by prior experience and the experimental design. He illustrates that a zero prior prevents updating, and that the experimental setup (e.g., selecting the best from a group) can change the effective prior. The lecture then transitions to a review of hypothesis testing, discussing type I error and power, and hints at connections to Bayesian evidence. The lecture is interactive, with questions from students, and sets the stage for future topics including exponential families, Bayesian models for classification, regression, clustering, and dimensionality reduction.

151 words

Critical Evaluation

Value of the Information & Strength of the Argument

The lecture provides a solid introduction to Bayesian reasoning, using a clear and intuitive example to illustrate the importance of priors and the influence of experimental design. The argumentation is logical and well-structured, building from Bayes’ theorem to a discussion of hypothesis testing. The lecturer effectively demonstrates how different priors lead to different conclusions and how the experimental setup can alter the effective prior. The interactive Q&A enhances the value by addressing student misconceptions and clarifying key points. However, the lecture is introductory and does not present novel information or advanced techniques, limiting its value for an expert audience.

Scientific Rigor, Source Quality, Title Accuracy

The lecture is scientifically rigorous, with correct mathematical derivations and clear explanations. The content is based on standard Bayesian statistics and hypothesis testing, which are well-established. The title accurately reflects the content, as it is an introductory lecture on Bayesian model selection. No external sources are cited, but the lecture is self-contained and relies on fundamental principles. The instructor demonstrates expertise and provides a coherent pedagogical narrative.

181 words

Title / Content Match

The title accurately reflects the content: an introductory lecture on Bayesian model selection.

Quality & Reliability

8/10

The lecture is delivered by an academic instructor, presents foundational Bayesian concepts with clear mathematical derivations and examples, and includes interactive Q&A. The content is well-structured and pedagogically sound, though it is an introductory lecture without novel research.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Contribution & Novelties

The lecture provides a clear pedagogical introduction to Bayesian model selection, emphasizing the role of priors and experimental design. It connects Bayesian inference with hypothesis testing, setting the stage for more advanced topics. The example of the oracle and the pre-experiment illustrates subtle points about prior elicitation and the impact of selection bias.

Pour aller plus loin :

Bayesian inference — Overview of Bayesian methods.
Bayes’ theorem — Fundamental theorem.
Hypothesis testing — Standard statistical framework.

76 words

Radar Profile

The radar profile shows high scores in quality and reliability, moderate in quantity and technical level, indicating a well-delivered introductory lecture with solid content but limited depth.

Reliability 8/10