Skip to content
#AI

CS230: Deep Learning by Stanford (AI column)

#AI #Software #ML #Engineering #Architecture

Started looking. lectures from the CS230 course Deep neural networks from Stanford 2025 A year and I like it so far. There are two teachers. Andrew Ng, founder of DeepLearning.AI and co-founder of Coursera, former chief scientist of Baidu, founder of Google Brain Kian Katanforoosh – CEO and Founder of Workera (platform for assessing AI skills)Co-founder of deeplearning.ai The course is interesting structure - students watch lectures online (from Deep Learning Specialization Coursera)and face-to-face classes are devoted to practical discussions, discussions and work on projects (You can actually see this in the lectures on YouTube that follow the first introductory session.).

The structure of the course is about this

1Basics of neural networks and Deep Learning Building a neural network from scratch in Python (no frameworks!) Understanding the internal device, not just using TensorFlow/PyTorch

2Hyperparameter Tuning and Optimization Ng frankly admits: 'Every PhD student I've had has been sitting up to 2-3 At night by selecting hyperparameters.” Quick setup skills can save you working hours - the difference between debugging in 3 morning 7 morning.

3Strategies for building ML-projects Critical for engineers: How to make system decisions in complex projects (facial recognition, security systems) Disciplined development process: the difference between completion in days vs months When you do not need to collect more data or buy a GPU (business-mistake) A real example from the lecture: Ng talks about a meeting with the CTO of a large company that bought a bunch of GPUs and gave a nephew-student the budget to "do AI." Without strategy is a waste of money. ​ 4️⃣ Convolutional Networks (Computer Vision) Specialized models for working with images and video.

5Sequence Models and Transformers

  • Working with time series and text Transformer architecture - the basis of the GenAI revolution

If we talk about the interesting moments of the first lecture, they are

Why Deep Learning Is Dominant Ng explains through simple visualization: Traditional ML: Performance plateaus as data increases Deep Learning: Continues to improve with scaling data and models Scaling Laws: Performance is predictable – you can pre-calculate how many GPUs to buy to achieve target accuracy ​ When to use Deep Learning vs GenAI Prompt LLM works well for:

  • Textual problems
  • Prototyping. ($20-100month acceptable) Deep Learning is required when: Prompts for a month don't work. AI billing is getting too expensive (user growth)

Work with audio, images, video, structured data Fine-tuning for specific tasks Many of our accounts from LLM providers have been really exciting. I don’t want to give you the numbers, but that’s a lot more than we wanted to pay.” ​ The Hierarchy of AI Skills CS Fundamentals (basis) -» Machine Learning (datawork) -» Deep Learning ← (course CS230 here) -» Generative AI (Transformers, LLMs) Ng stresses that understanding CS foundations is critical even when using AI code assistants (Cursor, Copilot).

In general, I have watched a couple of lectures and can say that the course is interesting, and the teachers explain well. I think I'll sign up for specialization.

#Software #ML #AI #Engineering #Architecture