[ИАД, осень 2026] Интеллектуальный анализ данных 2

[ИАД, осень 2026] Интеллектуальный анализ данных 2

[IAD, autumn 2026] Data Mining 2

by Machine Learning – Intelligent Systems · View the channel on PSI

Formal & Physical Sciences AI — Models & Research UYQArtificial intelligence (AI)UYQMMachine learning
🎙 Machine Learning – Intelligent Systems 👥 8K 📅 October 7, 2026 ⏱ 123 min 👁 6 📄 tutorial 🧭 2026-10-07
Available in: English (current) Français

Keywords

causal inferenceill-posed problemsdirect probleminverse problemsymbolic regression

Summary

This lecture, part of the ‘Intelligent Data Analysis’ course, builds on a previous session to explore causal models as a generalization of standard machine learning models. The instructor begins with a motivational discussion on the limitations of current language models, which lack structural connections between objects, and introduces the concept of ‘world models’ as a future direction. He then presents a schematic framework for direct and inverse problems in model selection, emphasizing the ill-posed nature of parameter optimization, illustrated with the sine function example. The discussion shifts to feature selection and the role of structure (matrix A) and its probabilistic representation (Gamma). Practical laboratory work is detailed, involving the selection of an equation (e.g., from Arnold’s book or a modern paper), converting it into a graph representation, constructing an adjacency matrix, and validating the model on real-world data like the Lotka-Volterra predator-prey system. The lecture includes interactive Q&A segments, clarifying expectations for a two-week lab cycle and the importance of code maintainability.

162 words

Critical Evaluation

Value of the Information & Strength of the Argument

The information provided is valuable for students and practitioners seeking a deeper understanding of model selection beyond standard ML practice. The lecturer presents a coherent argument connecting causal models, ill-posed problems, and practical implementation, using clear examples such as the sine function with multiple local optima and the Lotka-Volterra model. The argumentation is logically structured, moving from motivation to theoretical framework and then to hands-on laboratory instructions, though it sometimes relies on personal anecdotes rather than systematic literature review.

Scientific Rigor, Source Quality, Title Accuracy

The rigor is acceptable for a lecture: the instructor references classical concepts like Hadamard’s well-posedness, Arnold’s models, and mentions Wikipedia as a resource, but does not provide formal citations. The title ‘Intelligent Data Analysis 2’ is generic but appropriate as a continuation of a course, and the content aligns with the expected advanced topics in data analysis. No external sources are systematically cited beyond the mention of Wikipedia and the authors of the notations (Oleg Bakhteev, Nikita Zhivotovsky), which limits verifiability.

175 words

Title / Content Match

The title 'Intelligent Data Analysis 2' accurately reflects the course content, which focuses on advanced topics in data analysis, particularly causal models and model selection.

Quality & Reliability

7/10

The lecture is technically rigorous, covering well-established concepts such as Hadamard's ill-posedness and the direct/inverse problem dichotomy, but relies on few formal citations and shows a somewhat informal style with limited external references.

Key Moments

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

Cited Sources

Contribution & Novelties

The lecture contributes a pedagogical framework linking causal models to the classical theory of model selection, emphasizing the ill-posed nature of parameter optimization and the importance of structure representation via adjacency matrices. It offers practical guidance for building and testing such models, which is a step toward unifying ML with causal reasoning.

Pour aller plus loin :

  • Causal inference (Wikipedia) — Related to the idea of causal models and their distinction from purely statistical models.
  • Symbolic regression (Wikipedia) — Pertains to the automated discovery of equations from data, which aligns with the lab work on selecting and representing equations.
  • Neural ordinary differential equations (Wikipedia) — A modern approach to modeling dynamic systems, relevant to the ODE examples mentioned in the lecture.

121 words

Radar Profile

The radar profile shows high technical content and good information quantity, but lower reliability due to limited formal sources. The lecture is theoretical and practical, with a strong emphasis on formal definitions and examples.

Reliability 7/10