#019 ML Евгений Разинков. Дорожная карта погружения в машинное обучение ч.2

Value of the Information & Strength of the Argument

The value of the information is high: the guest provides a coherent, experience-based roadmap for self-studying ML, with concrete justifications for the order of topics and the emphasis on classical CV before deep learning. He argues strongly that many ideas from classical CV, such as integral images, cascades, non-maximum suppression, and Hough transforms, are reused in modern deep learning architectures, and understanding them at a simple level aids later comprehension. He also addresses common concerns, like whether backpropagation needs to be derived manually, arguing it yields deep intuition and troubleshooting ability. The argumentation is solid, relying on professional experience and pedagogical reasoning rather than mere opinion. The discussion is structured and the points are interconnected, making a compelling case for a foundational approach.

Scientific Rigor, Source Quality, Title Accuracy

The content demonstrates strong scientific rigor within the framework of an informal podcast. The guest, as a PhD and industry lead, shows deep knowledge of the field. While no explicit academic sources are cited in the conversation, the references to algorithms and papers (e.g., Viola-Jones, ORB-SLAM, ResNet, single-shot detectors) are accurate and well-known. The title accurately reflects the content, being a continuation of a planned roadmap. The description provides links to the guest’s educational resources and previous episodes, adding credibility. The discussion is technically deep but remains accessible, and the lack of formal citations is typical for a practitioner-focused podcast.

The title accurately reflects the content: a continuation of a two-part discussion on a machine learning learning roadmap.

8/10

The podcast features an experienced ML practitioner and academic, offering a well-structured learning path with rationale for sequence and selection of algorithms, plus practical insights. However, the discussion is informal and lacks formal citations in the conversation itself.

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

• 01:00Recap of previous episode and introduction of roadmap continuation
• 03:00Importance of linear regression for understanding underfitting, regularization, and future reuse in deep learning
• 05:00Transition to logistic regression as a classification algorithm, introducing cross-entropy as a loss function
• 07:00Why decision trees and Random Forests are essential, including interpretability and their role in classical CV
• 10:00The fork: classical computer vision vs deep learning, with recommendation to study classical CV first
• 12:30Detailed examples from classical CV: Viola-Jones cascade, integral images, and their recurrence in modern detectors
• 15:00Hough transform and non-maximum suppression as timeless tools, with applications in ORB-SLAM and pose estimation
• 20:00Practical reasons for classical CV: low-power devices, small data scenarios, and commercial niche
• 25:00Deep learning course recommended as third step, starting from MLP to ResNet, with emphasis on backpropagation derivations

• Evgeny Razinkov's personal website — Link to the guest's personal site with information about him and his activities.
• Video lectures by Evgeny Razinkov — YouTube channel with lecture videos on machine learning.
• VK public page of Evgeny Razinkov — VK community with AI materials and announcements.
• Anchor podcast episodes page — All episodes of the Machine Learning Podcast.
• Episode #007 — Previous podcast episode with Evgeny Razinkov about managing commercial development in ML.
• Episode #018 — Part 1 of the roadmap discussion.
• Episode #019 on Anchor — This episode on Anchor.
• Apple Podcasts page — Podcast availability on Apple Podcasts.
• Google Podcasts feed — Podcast feed on Google Podcasts.
• Podcast in Yandex Music — Podcast availability on Yandex Music.
• Podcast's VK public — VK community of the podcast.
• Telegram channel about becoming ML specialist — Telegram channel dedicated to learning ML.
• Telegram channel of Evgeny Razinkov — Telegram channel with AI announcements and materials.

• Machine Learning course by Andrew Ng — A foundational ML course that teaches regression, classification, and neural networks in a structured manner, aligning with the guest's suggested starting point.
• Deep Learning textbook by Ian Goodfellow — Comprehensive resource covering deep learning topics, including backpropagation and convolutional networks.