From pixels to predictions — how machines learn to see
A story-driven, from-first-principles walk through convolutional neural networks — starting with why plain MLPs fail on images and ending with modern architectures, training pipelines, and the interview questions that test it all.
28 chapters across 7 parts — the full book for this course.
No previous year questions have been added yet.
4 chapters · 44 min
Why images need a different kind of network
What happens when you feed raw pixels into a plain MLP
Refreshing perceptrons and MLPs before going deeper
The vocabulary you need before the math starts
5 chapters · 82 min
The standard pipeline from pixels to predictions
Convolution, padding, stride, and the formulas behind them
Tracing an image through every layer, step by step
How gradients flow backward through convolution and pooling
Convolution, pooling, flatten, and dense layers up close
6 chapters · 76 min
Why ReLU won, and what its alternatives fix
Teaching a vision model what "wrong" looks like
Choosing how a network descends its loss landscape
The dials that decide whether training succeeds or stalls
Wiring every piece into one end-to-end workflow
Keeping a CNN from memorizing its training images
3 chapters · 41 min
4 chapters · 56 min
From LeNet to EfficientNet, in one lineage
Where CNNs shine, and where they quietly break
Where CNNs actually run in production today
CNNs versus MLPs, RNNs, and Transformers
4 chapters · 47 min
2 chapters · 14 min
Why images need a different kind of network