₹0 for students · Certificate · Intermediate

Dive into Deep Learning

18 lessons: neural networks from tensors and autograd to CNNs, RNNs and transformers

Start free in the app

18 lessons · ~16 hours · 30-question final exam · pass mark 70%

About the course

What you’ll learn

The deep-learning textbook used at hundreds of universities, as a guided course. Learn tensors and automatic differentiation, linear and softmax regression, multilayer perceptrons, regularisation, convolutional networks and ResNets, recurrent networks and LSTMs, attention and transformers, optimisation algorithms, transfer learning, word embeddings and BERT, and GANs. Each lesson has a study guide, links to the runnable chapters, and a quiz. Pass every quiz and the final exam to earn a verifiable certificate. Some comfort with Python and basic maths helps.

Deep learningPyTorchAutogradCNNsResNetRNNsLSTMTransformersAdamTransfer learningBERT

Part of the career track: Machine Learning Engineer

How the certificate works

  1. Read each lesson’s study guide, then the full lesson and any notebooks or exercises.
  2. Pass every lesson quiz with 70% or more.
  3. Pass the final exam: 30 questions in 45 minutes.
  4. Get a certificate with a public verification link, and add it to LinkedIn.

Syllabus

18 lessons in 5 parts

Foundations

1
Introduction to Deep Learning
40 min · Core ingredients: data, model, objective (loss) and optimisation algorithm.
2
Tensors, Linear Algebra and Automatic Differentiation
60 min · Tensors are n-dimensional arrays with fast element-wise ops, reshaping, slicing and reductions.

Neural network basics

3
Linear Regression as a Neural Network
50 min · Linear regression is a one-layer network: ŷ = Xw + b with squared loss.
4
Softmax Regression for Classification
50 min · Labels as one-hot vectors; one output (logit) per class.
5
Multilayer Perceptrons and Backpropagation
55 min · MLPs add hidden layers with non-linear activations to model non-linear relationships.
6
Generalisation, Dropout and Practical Training
50 min · Deep nets can memorise but still generalise; validate carefully.

Convolutional networks

7
Convolutional Neural Networks
60 min · CNNs exploit locality and translation invariance in images.
8
Modern CNNs: AlexNet, VGG, ResNet and Batch Normalisation
60 min · AlexNet (2012): deeper, ReLU, dropout, augmentation, GPUs.

Sequence models and attention

9
Recurrent Neural Networks
55 min · Sequences need models that respect order and variable length.
10
LSTMs, GRUs and Sequence-to-Sequence Models
60 min · LSTMs use a memory cell with input, forget and output gates.
11
Attention Mechanisms
55 min · Attention lets a model focus on relevant parts of the input at each step.
12
The Transformer Architecture
60 min · Transformers use attention without recurrence.

Training and applications

13
Optimisation Algorithms
55 min · Optimisation in deep learning serves generalisation; loss surfaces are non-convex with saddle points.
14
Computational Performance and GPUs
40 min · GPUs run the same operation on many numbers in parallel; memory movement is often the bottleneck.
15
Computer Vision: Augmentation and Fine-Tuning
50 min · Augmentation (flips, crops, colour changes) improves robustness; training only.
16
NLP Pretraining: word2vec and BERT
55 min · Pretraining learns reusable language representations from unlabelled text.
17
Generative Adversarial Networks
40 min · Generative models create new data resembling the training data.
18
Hyperparameter Optimisation and Practical Deep Learning
40 min · Hyperparameters strongly affect results; HPO automates the search.

This course is based on "Dive into Deep Learning" by Aston Zhang, Zachary C. Lipton, Mu Li and Alexander J. Smola (https://d2l.ai/), whose text is licensed under CC BY-SA 4.0. Vikosha's study guides and quiz questions for this course adapt that material and are shared under the same licence (CC BY-SA 4.0). Certificates are issued by Vikosha and are not issued or endorsed by the authors or Amazon. Licence: Text CC BY-SA 4.0; sample code under a modified MIT licence. By Aston Zhang, Zachary C. Lipton, Mu Li and Alexander J. Smola. View the original curriculum