[ИАД, весна 2026] Введение в машинное обучение. Лекция 9: Логические методы машинного обучения

Value of the Information & Strength of the Argument

The lecture provides substantial value by bridging historical foundations with modern relevance. The presenter effectively argues for the importance of interpretable models in the context of trustworthy AI, using concrete examples from medicine and credit scoring. The argumentation is solid, built on formal definitions and a clear logical progression. The use of Bongard’s problems as illustrative examples is particularly effective in conveying abstract concepts like feature extraction and rule selection. The presenter also addresses the fundamental issue of multiple equally valid rules on a given dataset, linking it to regularization and ill-posed problems. The discussion is well-structured and technically sound, though it remains at an introductory level, avoiding deep mathematical derivations.

Scientific Rigor, Source Quality, Title Accuracy

The lecture demonstrates scientific rigor through its historical accuracy and formal definitions. The presenter references key works, such as Pedro Domingos’ ‘The Master Algorithm’ and Bongard’s book ‘Pattern Recognition’, without providing explicit citations or URLs. The quality of sources is high, but the lack of direct references in the description limits verifiability. The title accurately reflects the content, and the lecture stays on-topic throughout. The presenter’s expertise is evident, and the content aligns with established knowledge in the field. However, the absence of peer-reviewed sources and the reliance on historical anecdotes slightly reduce the overall scientific rigor.

The title accurately reflects the content: a lecture on logical methods in machine learning, part of a series on machine learning schools.

8/10

The lecture is well-structured, historically grounded, and technically accurate, with clear formal definitions and references to established concepts (e.g., Bongard's tests, rule induction). The presenter demonstrates deep expertise, though the content is not peer-reviewed and represents a pedagogical perspective.

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

• 02:00Introduction to the lecture series and the five schools of machine learning, positioning logical methods as the fourth school.
• 05:00Discussion on interpretability and explainability in AI, introducing five related terms and the trade-off between interpretability and accuracy.
• 10:00Historical overview of Mikhail Bongard and his contributions to pattern recognition, including the Bongard problems.
• 15:00Examples of Bongard problems illustrating feature extraction, feature selection, and the challenge of learning from raw data.
• 20:00Formal definition of logical patterns (rules) as predicates with interpretability and informativeness criteria.
• 25:00Explanation of the four steps of rule induction: rule form, generation, evaluation, and combination into a classifier.
• 30:00Discussion on the use of conjunctions and syndromes in rule-based systems, with examples from medicine.
• 35:00Introduction to decision trees as another class of logical methods, to be covered in more detail later.

• The Master Algorithm — Referenced as the source for the five schools of machine learning.
• Bongard's book 'Pattern Recognition' — Mentioned as the source of the 100 Bongard problems and early insights into overfitting.

• The Master Algorithm — Pedro Domingos' book, which the lecturer references for the taxonomy of machine learning schools.
• Bongard's book 'Pattern Recognition' — Mikhail Bongard's book, which contains the 100 problems and early discussions on overfitting.