Deep Learning with Neural Networks
Build a real understanding of how neural nets work
About this course
From a single neuron to multi-layer networks and convolutions, this course gives you the foundations to read modern papers and build your own models.
What you'll learn
- Explain forward and backward propagation
- Pick activation functions intelligently
- Understand CNNs at a working level
- Reason about model capacity and depth
Curriculum
Train a Model You Can Brag About
Pick a problem and a dataset (image classification, text classification, or sequence prediction). Deliverables: 1. Notebook that trains a model from scratch (or fine-tunes a pre-trained one) to beat a baseline. 2. Training/validation loss curves + at least 2 sample predictions on held-out data. 3. A brief writeup (≤500 words) covering: architecture choice, what you tried that didn't work, final result vs baseline. 4. Bonus: a tiny demo (Gradio) where someone can paste/upload an input. Don't aim for state-of-the-art. Aim for *understood*.
- 1Baseline is named and beaten
- 2Loss curves show convergence (not divergence or stagnation)
- 3Writeup explains failed attempts — shows real iteration
- 4Architecture choice is justified, not random
- 5Code runs end-to-end on a fresh machine
What graduates shipped.
Once you ship, your project lives here — linked from your public verify page so recruiters can click through.