Neural Networks

Introduction to Neural Networks

Exercise 10.1

Intro to Neural Networks

Questions:

  1. Would you rather use TensorFlow or Keras to build your models?
  2. Tasks 1 & 2: Report your best hyper-parameter settings and their resulting performance on the testing dataset.

Save your answers in lab10_1.txt.

Training Neural Networks

Exercise 10.2

Improving Neural Net Performance

Questions:

  1. What does AdaGrad do to boost performance?
  2. Tasks 1–3: Report your best hyperparameter settings and their resulting performance.
  3. Optional Challenge: You can skip this exercise.

Save your answers in lab10_2.txt.

Multi-class neural networks

Exercise 10.3

Classifying Handwritten Digits with Neural Networks

Questions:

  1. Task 1: What does the confusion matrix show for this example?
  2. Task 2: How does the TensorFlow network architecture differ from the Keras example given in class? Report any improvements you can make over the baseline testset accuracy for this task.
  3. Task 3: What differences can you see between the visualizations for 10 steps and 1000 steps?

Save your answers in lab10_3.txt.

Keras — Convolutional Neural Networks

Build a Convolutional Neural Network (CNN) for the Cats & Dogs image datasets.

Exercise 10.4

ML Practicum: Image Classification — Do the first exercise included in this tutorial.

Questions:

  1. Exercise 1:
    1. What’s the size/shape of the cats/dogs datasets?
    2. How does the first CNN compare with the one we did in class.
    3. Can you see any interesting patterns in the intermediate representations?
  2. You can skip Exercises 2 & 3.

Save your answers in lab10_4.txt.

Checking in

We will grade your work according to the following criteria:

See the policies page for lab due-dates and times.