VFlash: Fast, Accurate Visual Planning via Flash Primitive Decoding
u_Unrolled
D4: Deep Learning
Phase 3: Deep Learning & Neural Networks/DOMAIN 4

Domain 4: Deep Learning & Neural Networks

IntermediateUnrolled Pro

Neural networks from scratch: backprop and autograd, debugging training, CNNs and ResNets, the major architectures, and engineering with PyTorch.

Estimated study time: about 110 hours
Access: Unrolled Pro
Unrolled Pro Domain

Unlock all 6 lessons with Unrolled Pro

Full production code, quantitative capacity math, failure recovery playbooks, and interview coaching.

Get Pro from ₹167/mo (~$2/mo)14-day refund guarantee

Lessons

6 lessons
01

Neural Network Fundamentals — Layers That Learn Their Own Features

FREE 15m

Every LLM is a very deep neural network. Its building blocks — linear layers, activations, normalization, residual connections, careful initialization — are all on this page.

✓ Runnable code✓ Exercise with solutionStart Lesson
02

Autograd From Scratch, and Debugging Training — Gradient Checks, Loss Curves and the Usual Bugs

Pro 15m

PyTorch's autograd feels like magic until something goes wrong.

✓ Runnable code✓ Exercise with solutionStart Lesson
03

Major Architectures — Matching the Network to the Shape of the Data

Pro 15m

An MLP treats its input as a flat list and ignores structure. But images are grids (nearby pixels matter), events are sequences (order matters), knowledge is a graph (connections matter).

✓ Runnable code✓ Exercise with solutionStart Lesson
04

CNNs From Scratch and ResNets — Convolution, Pooling, Residuals and Modern CNN Design

Pro 15m

Writing a convolution's forward *and backward* pass once makes everything about CNNs concrete: why weight sharing gives translation equivariance and few parameters, how receptive fields grow, what the backward pass costs.

✓ Runnable code✓ Exercise with solutionStart Lesson
05

PyTorch Engineering — From Tensors to Fast, Correct Training Loops

Pro 20m

The difference between a model that trains in a day and one that takes a week — or silently trains wrong — is usually engineering: unnecessary copies, a broken autograd graph, a data loader starving the GPU, memory blown on activations, or a checkpoint that can't be loaded safely.

✓ Runnable code✓ Exercise with solutionStart Lesson
06

Bridge: From Learning Numbers to Learning Language

Pro 10m

Atlas can train deep networks efficiently — but its input is still numbers we prepared. Students type text.

✓ Runnable codeStart Lesson
Domain 4: Deep Learning · Unrolled