[ИАД, осень 2026] Байесовское мультимоделирование I. Лекция 2

Value of the Information & Strength of the Argument

The lecture offers valuable pedagogical content by systematically revising fundamental concepts and connecting them to Bayesian inference. The argumentation is solid: mathematical derivations for MLE properties and KL divergence are presented clearly, and examples like the coin toss illustrate the intuition. The inclusion of less common topics such as circular distributions broadens the perspective. However, the lecture is more of a review than an original contribution; its value lies in synthesis and clarity rather than novelty.

Scientific Rigor, Source Quality, Title Accuracy

Scientific rigor is moderate; the instructor references textbooks (e.g., ‘How to Lie with Statistics’) and internal courses but provides no formal citations. The title matches the content, as it is a foundational lecture for a Bayesian multimodeling course. The material is standard and aligns with mainstream statistics, so reliability is high despite informal sourcing. The adéquation titre/contenu is adequate; the lecture is a faithful introduction to Bayesian concepts.

The title indicates a Bayesian multimodeling lecture; content matches as it covers core Bayesian concepts and distributions, though it is more of a review/revision lecture.

7/10

The lecture provides a solid overview of foundational distributions and Bayesian inference, with clear derivations and examples. However, sources are informal (mentions of textbooks and courses) and no formal citations are given.

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

• 00:30Introduction: Bayesian problem formulation, goal to recover data generation process.
• 05:00Coin toss example: Bernoulli distribution and frequency-based estimation.
• 12:00Log-likelihood, score function, Fisher information, and MLE properties.
• 20:00Connection between MLE and KL divergence, law of large numbers.
• 28:00Central tendency measures: mean, median, mode, and skewness/kurtosis.
• 36:00Empirical PDF and histogram estimation, kernel density smoothing.
• 42:00Beta distribution as conjugate prior for Bernoulli.
• 48:00Softmax parametrization, multinomial, and Dirichlet distribution.
• 58:00Normal and Student's t distributions, heavy tails and limit.
• 1:08:00Circular distributions: von Mises and Kent, motivation and examples.

• How to Lie with Statistics — Mentioned as an example of using different central tendency measures.
• Course by Andrey Gribovoy on applied statistics — Referenced for terminology on score function and other statistical concepts.