[ИАД, осень 2025] Методы глубокого обучения. Занятие 1: Introduction, MLP, Backpropagation

Value of the Information & Strength of the Argument

The lecture provides a solid introduction to deep learning, with clear explanations of key concepts and mathematical foundations. The argumentation is logical, starting from logistic regression and building up to the universal approximation theorem and the need for non-linear activation functions. The instructor effectively uses examples and visual aids to illustrate points. The value is high for beginners, offering a comprehensive overview of the field’s history and core principles.

Scientific Rigor, Source Quality, Title Accuracy

The scientific rigor is adequate for an introductory lecture. The instructor references well-known models and papers (e.g., AlexNet, AlphaGo, GPT-4) but does not provide formal citations. The title accurately reflects the content, and the lecture is well-structured. The description does not include additional sources, so the sources cited are based on the video content. The lecture is suitable for a university course, and the instructor’s industry experience adds practical perspective.

The title accurately reflects the content: an introductory lecture on deep learning methods, covering MLP and backpropagation.

8/10

The lecture is delivered by a PhD student with industry experience, covering foundational concepts with mathematical derivations and references to well-known models. The content is accurate and well-structured, though it lacks formal citations and peer-reviewed sources.

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

• 02:00Introduction: AI vs ML vs Deep Learning
• 05:00Examples of deep learning models: AlexNet, AlphaGo, StyleGAN, DALL-E, AlphaFold, GPT-4
• 10:00Reasons for deep learning boom: big data, GPUs, frameworks
• 15:00Course structure and grading
• 20:00Logistic regression and its limitations
• 25:00Universal approximation theorem and sigmoid approximation
• 30:00Multilayer perceptron and activation functions
• 35:00Discussion of activation functions: sigmoid, tanh, ReLU, Leaky ReLU, ELU

• AlexNet — Mentioned as a landmark model for image classification.
• AlphaGo — Mentioned as a model that defeated a top Go player.
• StyleGAN — Mentioned as a generative model for images.
• DALL-E — Mentioned as a text-to-image generation model.
• AlphaFold — Mentioned as a model for protein structure prediction.
• GPT-4 — Mentioned as a multimodal model that sparked the AI boom.

• Deep Learning Book — Standard reference for deep learning concepts.
• CS231n Lecture Notes — Stanford course on CNNs, relevant to the lecture's content.