Deep Learning: final assessment
This covers the whole course. Take it after the last module.
Final quiz
Practical assessment: Train a network from scratch
Without using hints or libraries, write a program that trains a two-layer neural network on a small classification dataset you create, and reports what it learned.
- The forward pass, backpropagation and update are written by you, and pass a numerical gradient check
- The network uses ReLU hidden units, a softmax output and cross-entropy loss
- Training uses mini-batches and either momentum or Adam
- The program reports training and validation accuracy every few epochs and stops early when validation stops improving
- The final test accuracy and a few misclassified examples are printed
Final project
A digit classifier from scratch: Build and train a neural network in plain Python that recognizes small pixel images of the digits 0 to 9, and measure how well it works.
How it is evaluated
- Correctness: the gradient check passes and the network learns the task
- Understanding: you can explain every line of the training loop
- Experiment: the effect of at least two settings, such as learning rate and hidden size, is measured
- Clarity: data, model and training code are separate and readable
To complete this course
- Score at least 70% on the final quiz
- Solve the beginner and intermediate exercises in every lesson
- Finish the final project and check it against the evaluation criteria