6: Tutorial stuff

Question 1

What does nCr expand to?

Answer 1

nCr = n! / r! * (n - r)!

Question 2

What is a full joint distribution?

Answer 2

???

Question 3

How do you find missing probabilities in full joint distributions?

Answer 3

???

Question 4

What is a Bayesian network?

Answer 4

A dag map of probabilistic connections between probabilities such that an arrow points from a probability that is conditionally dependant on another to the othe

Question 5

How do you build a Bayesian network?

Answer 5

???

Question 6

How do you draw a Bayesian network?

Answer 6

???

Question 7

What is a knowledge base (in logic)?

Answer 7

???

Question 8

What are entailments?

Answer 8

???

Question 9

When can you infer entailments?

Answer 9

???

Question 10

What is a most general unifier?

Answer 10

???

Question 11

What is a unifier?

Answer 11

???

Question 12

What is unification?

Answer 12

???

Question 13

What is substitution?

Answer 13

???

Question 14

How does unification work?

Answer 14

???

Question 15

What is the factoring rule?

Answer 15

???

Question 16

How does the factoring rule work?

Answer 16

???

Question 17

What is an 8-puzzle?

Answer 17

???

Question 18

What is NY taxi distance?

Answer 18

min non-diagonal node by node distance between two points on a graph (vertical and horizontal moves only)

Question 19

What is Manhattan distance?

Answer 19

min non-diagonal node by node distance between two points on a graph (vertical and horizontal moves only)

Question 20

What is backprop?

Answer 20

???

Question 21

How does backprop work?

Answer 21

???

Question 22

How does backprop relate to steepest gradient descent?

Answer 22

???

Question 23

How do you find the error-weight gradient for a unit behind a current hidden one, given its output (input into the current one)?

Answer 23

???

Question 24

How do gradients come to explode or vanish when the chain rule is applied to multiple hidden layers?

Answer 24

???

Question 25

What is the connection between validation and early stopping?

Answer 25

early stopping done to mainimise error in training/validation data to improve validation process

Question 26

What are hyperplanes?

Answer 26

???

Question 27

Where do hyperplanes rotate to from their initial setting when training is applied through back-propagation?

Answer 27

???

Question 28

What are convolutional neural networks?

Answer 28

Divide and conquer ish neural networks that convolute data to find patterns before recombining adjacent data into groups. often used in computer vision.

Question 29

How have convolutional neural networks helped recent breakthroughs in machine learning?

Answer 29

allowed advances in computer vision

Question 30

What is the difference in capability between a recurrent neural network and a feedforward network? Where does this difference arise from?

Answer 30

recurrent facilitates context and memory because of neurons being able to connect to themselves