Module 01 Knowledge Check
5 questions. Pass mark 4 out of 5. Answer every question before checking the answer key below, then retry after reading the feedback.
1. A linear model scores 0.5 on XOR while one hidden layer scores 1.0. What does that show?
- A. The linear model needed more training steps
- B. XOR is not linearly separable, so no single boundary can solve it
- C. The hidden layer memorised the four examples
- D. XOR requires a GPU
2. Training loss on randomly generated targets falls towards zero as hidden width grows. Why?
- A. The model found a hidden pattern
- B. The learning rate was too high
- C. With more parameters than training rows, the model can memorise them
- D. Random targets are easier to fit than real ones
3. Why does this course set a seed before every reported run?
- A. It makes training faster
- B. It improves accuracy
- C. It makes the run reproducible by someone else
- D. It prevents overfitting
4. Which four objects appear in essentially every training script in this course?
- A. Tensor, gradient, loss, accuracy
- B. Dataset, DataLoader, model, optimiser
- C. Encoder, decoder, head, scheduler
- D. Train, validation, test, holdout
5. What makes a reported number evidence rather than a claim?
- A. A high accuracy
- B. A large training set
- C. The seed, the population it covers, and its stated limitation
- D. Being produced by a deep model
---
Answer Key and Explanations
Check these only after attempting every question.
1. B — XOR is not linearly separable, so no single boundary can solve it
No straight line separates XOR's four points. The hidden layer builds an intermediate representation first, then separates that.
2. C — With more parameters than training rows, the model can memorise them
There is no pattern to find. A model with more parameters than examples stores them, which is why training loss alone is not evidence of anything.
3. C — It makes the run reproducible by someone else
Initialisation, dropout and shuffling all draw from a random generator. Seeding fixes its starting point so a reviewer can rebuild your exact result.
4. B — Dataset, DataLoader, model, optimiser
A Dataset holds examples, a DataLoader batches them, a model maps inputs to outputs, and an optimiser updates parameters.
5. C — The seed, the population it covers, and its stated limitation
Without those three a reviewer cannot reproduce, interpret, or bound the result.
Practical Check
Apply this module to your own work: complete the module activity for *Deep Learning Mindset, Boundaries, and Setup*, then write one sentence naming what your result shows and one naming what it does not.
Strong Answer Pattern
A strong answer names the task, the evidence used, the check performed, and the remaining limitation. It avoids "proved", "guaranteed", or "always" unless the evidence genuinely supports it.