Intelligent Systems: Machine Learning seminars

Intelligent Systems: Machine Learning seminars

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 14, 2026 ⏱ 90 min 👁 264 📄 tutorial 🧭 2026-10-14
Available in: English (current) Français

Keywords

Bayesian predictionModel averagingMarginal likelihoodLinear logistic regressionPosterior probability

Summary

This seminar reviews and extends Bayesian methods for model selection and prediction in machine learning. The lecturer begins with organizational announcements: assignment 2 is released, practical tasks require team formation, and tests are scheduled. The main lecture recaps the concept of posterior distribution over model parameters, emphasizing its role beyond point estimates. It derives the Bayesian predictive distribution for a single model by integrating out the parameter vector, showing that classical point estimates are only valid under zero uncertainty. Extending to multiple models, the lecture presents the Bayesian model averaging formula, where predictions from each model are weighted by the posterior model probabilities. These are computed using the model evidence (marginal likelihood) and prior model probabilities. The discussion includes the distinction between mixture-of-models and mixture-of-experts, and conditions under which the average reduces to a single model. A concrete example illustrates model comparison using interview data with PhD indicators, computing evidences for four competing models. The lecture concludes with an illustrative calculation of a combined prediction for a new candidate, demonstrating the weighted mixture of model-specific forecasts.

176 words

Critical Evaluation

Value of the Information & Strength of the Argument

The video offers high pedagogical value, systematically building from a single model to multiple models, with clear derivations and intuitive explanations. The argumentation is rigorous, relying on probability theory and Bayesian principles. The worked example with interview data effectively consolidates the theoretical points, showing how evidence and posterior model probabilities are computed in practice. The lecturer encourages questions and addresses them throughout, enhancing clarity.

Scientific Rigor, Source Quality, Title Accuracy

No external sources are cited in the video or description, so the scientific rigor rests on the internal consistency of the derivations and the established framework of Bayesian statistics. The title directly matches the content, which is a specialized seminar on machine learning. The presentation is technically sound and well structured, though the lack of references might limit verifiability for self-learners.

140 words

Title / Content Match

The title accurately reflects the content of an advanced machine learning seminar focused on Bayesian model selection.

Quality & Reliability

8/10

The lecture provides rigorous mathematical derivations of Bayesian concepts, with a clear pedagogical structure. The content aligns with established statistical theory, although no external sources are explicitly cited in the video or description.

Key Moments

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

Contribution & Novelties

The lecture provides a clear, rigorous exposition of Bayesian model averaging and evidence-based model comparison, bridging theoretical derivations with a practical example. It highlights the conditions under which model averaging collapses to a single model, offering insight into model uncertainty. The pedagogical approach of re-deriving formulas from first principles is particularly valuable for advanced students.

Pour aller plus loin :

  • Bayesian model averaging — Key concept for combining predictions from multiple models.
  • Marginal likelihood — The evidence term central to model comparison.
  • Posterior predictive distribution — The foundation of Bayesian prediction as applied in the lecture.

96 words

Radar Profile

The radar profile shows high scores across information quantity, quality, technical level, and reliability, indicating a dense and well-structured academic lecture. The balance between derivations and examples makes it suitable for an audience with a solid background in probability and statistics.

Reliability 8/10