机器学习中的神经网络Neural Networks for Machine Learning：Lecture 1 Quiz

Lecture 1 QuizHelp Center

Warning: The hard deadline has passed. You can attempt it, but you will not get credit for it. You are welcome to try it as a learning exercise.

Question 1

We often don't know how much data we will need in order for a learning system to generalize well from training data to test data on a given task. 
True or false: when choosing how much data to give to a learning system in order to make it generalize well, we need to make sure that we don't give it too much data.

False

True

Question 2

Data can change over time, in particular we might observe different input/output relationships. In order to account for this we can adapt our learning system to the new data by, for example, training on new examples. 

If the relationship between inputs and outputs for old examples has not changed, how can we prevent a neural network from forgetting about the old data?

Ignore the issue and hope that everything will be OK.

Train on a mix of old and new data.

Prevent the system from changing the weights too much.

Train two networks, one for old data and one for new data.

Question 3

Which of the following are good reasons for why we are interested in unsupervised learning?

It lets us avoid supervised learning entirely.

It allows academic researchers to publish more papers.

It can be used to learn features that may help with supervised tasks.

It allows us to learn from vast amounts of unlabelled data.

Question 4

Which of the following tasks are neural networks good at?

logical reasoning

Recognizing fragments of words in a pre-processed sound wave.

Recognizing badly written characters.

Storing lists of names and birth dates.

Question 5

Which number is biggest?

The number of milleseconds in a human lifetime.

The number of bits of Random Access Memory (usually just called memory) in a modern laptop.

The number of synapes in a human brain.

The Greek national debt in euros

Question 6

Which of the following facts provides support for the theory that the local neural circuits in most parts of the cortex all use the same general purpose learning algorithm?

If the visual input is sent to the auditory cortex of a newborn ferret, the "auditory" cells learn to do vision.

The fine-scale anatomy of the cortex looks pretty much the same all over.

If part of the cortex is removed early in life, the function that it would have served often gets relocated to another part of cortex.

Brain scans show that different functions (like object recognition and language understanding) are located in different parts of the cortex.