Deep Learning A-Z: Neural Networks & AI — Review

Our review

Deep Learning A-Z is the neural-network follow-on to Machine Learning A-Z, built in the same template-driven style that made the original so approachable. Across 23 hours, Eremenko and de Ponteves work through the major architectures — artificial, convolutional and recurrent neural networks, LSTMs, self-organising maps, Boltzmann machines and autoencoders — in both TensorFlow/Keras and PyTorch.

The course's strongest feature is that it demystifies architectures that sound forbidding. Each section pairs an intuition segment — what a CNN "sees," why an RNN suits sequences — with a working code template you run and then adapt to your own problem. That structure lets you build a convolutional image classifier or a recurrent sequence model without first earning a maths degree, which is exactly what most applied learners need to get unstuck.

This suits someone who already has machine-learning basics and wants to move into deep learning with breadth across the main network types, and learners who prefer building from a template over deriving from first principles. By the end you've touched every major architecture and can recognise which one fits a given problem.

The honest limitation, as with its sibling, is theoretical depth and production-readiness. It teaches intuition and code, not the mathematics of backpropagation or optimisation, and it stops at building models rather than deploying and maintaining them. Some of the covered techniques (self-organising maps, Boltzmann machines) are also more academically interesting than commonly used day to day.

The clearest prerequisite and companion is Machine Learning A-Z, which covers the supervised-learning groundwork this course assumes. For the maths and theory these applied courses skip, Andrew Ng's Machine Learning Specialization is the free-to-audit standard. And if you want a certification-oriented, TensorFlow-specific deep-learning path instead, the TensorFlow Developer Certificate course is the more focused alternative.

Worth knowing about the framework coverage: teaching both TensorFlow/Keras and PyTorch is unusual and useful, because it means you're not locked into one ecosystem and can read code from either camp. The trade-off is that neither is covered as deeply as a single-framework course would manage, so treat this as the breadth-first survey that helps you decide which framework to then specialise in rather than the final word on either.

Verdict: at $14.99 this is solid value for a broad, hands-on introduction to deep learning that gets you building real neural networks quickly. Have Python and ML basics first, treat it as applied rather than theoretical, and supplement with Ng's course for the underlying maths. Our editorial score: 9/10.