You can outsource your thinking but you cannot outsource your understanding. 🧠✨

That's the entire problem with ML education right now. 📉

PyTorch will let you train a model without knowing what a gradient is. ⚡️ Keras will let you stack layers without knowing what any of them compute. The code runs. The model trains. You have output. You have zero understanding. 🤷‍♂️

Simon J.D. Prince built a notebook collection that won't let you skip the hard part. 🛠

Shallow networks first. What does one layer actually compute? What do the decision regions look like? You see it geometrically before you write a single line. 📐👀

Optimization compared, not prescribed. Line Search vs SGD vs Adam on the same problem. You watch them diverge. You understand why Adam isn't always the answer. 📉📈

Backpropagation to Self-Attention to Graph Neural Networks as one continuous thread. Not isolated tutorials. A progression. 🔗🚀

Three lines of code can train a model. These notebooks make sure you understand the model you trained. 🧐

Here's the resource: udlbook.github.io/udlbook/ 🔗

https://t.me/MachineLearning9 🩵