Week 1 Flashcards
1
Q
what is generalisation
A
a general category of algorithm whereby:
- pairs of data (xi, yi) are provided to the algo and for i in range 1, …, N
- the algo creates a mapping from each x to each y, thereby it is able to generate a y given and x
this holds even if it has not seen the x before
2
Q
Pro and con of supervised learning
A
Pro: well understood and performance is easy to measure
Con: labelling can be laborious
3
Q
Pro and con of unsupervised learning
A
Pro: easy to implement at scale
Con: harder to understand and can only be used in specific scenarios (e.g: clustering)
4
Q
Average loss
A
(Empirical risk)
5
Q
Binary cross entropy loss
A
6
Q
Neural network (notation)
A
7
Q
Activation function
A
8
Q
Deep NN
A
NN with multiple hidden layers
9
Q
Squared lost
A
