Final

Question 1

Machine learning

Answer 1

Machine learning is the study of algorithms that:
improve their performance P
at some task T
with experience E

Put data and output into computer to come out with a program

Examples of tasks best solved by program learning: recognizing patterns, generating patterns, recognizing anomalies, prediction/recommendation

Question 2

Machine learning problems

Answer 2

Categorization: Discrete supervised learning
Clustering: Discrete unsupervised learning
Regression: Continuous supervised learning (function that predicts y given x)
Dimensionality reduction: Continuous unsupervised learning

Question 3

Test-set method

Answer 3

To avoid the problem of overfitting (fitting noise, too precise)
Randomly choose 30% of the data to be in a test set, remainder is a training set, perform your regression on the training set, estimate your future performance with the test set
It imposes a penalty for unnecessary complexity (best method is quadratic)

Question 4

KNN

Answer 4

K-nearest neighbour
Assign label to item based on the class of the 1 or 3 or whatever nearest neighbour, an item is classified by a plurality vote of its neighbours

Question 5

Risk attitudes

Answer 5

§ You’re risk-adverse if you have two alternatives with the same EV, and you choose the alternative with less variation in outcomes. (gains)
§ You’re risk-neutral if you have two alternatives with different variations in outcomes and the same EV and you’re indifferent between those
alternatives.
§ You’re risk-seeking if two alternatives have the same expected monetary value and you choose the one with the highest variability. (losses)

Question 6

Heuristics

Answer 6

§ Availability
The easier it is to consider instances of class Y, the more frequent we think it is
§ Representativeness
The more object X is similar to class Y, the more likely we think X belongs to Y
§ Anchoring
Initial estimated values affect the final estimates, even after considerable adjustments

Question 7

Gambler’s fallacy

Answer 7

Assuming that a departure
for what occurs on average will be corrected in the short run.
In other words: because an event has not
happened recently, it has become “overdue”
and is more likely to occur

Question 8

Design principles for software

Answer 8

User familiarity
* The interface should be based on user-
oriented terms and concepts rather than
computer concepts.
à e.g. letters, documents, folders etc.
rather directories, file identifiers, etc.
Consistency
* Information, Commands, and Menus
should have the same format
Minimal surprise
* If an action operates in a known way,
the user should be able to predict the
operation of comparable commands

Question 9

Modality

Answer 9

An human-machine interface is modal with
respect to a given gesture when (1) the current state of the interface is not the user’s locus of attention and
(2) the interface will execute one among several different responses to the gesture, depending on the system’s current state

Question 10

Affordances

Answer 10

• The perceived and actual
  fundamental properties of the
  object that suggest how it
  could be used
• As electronic devices grow in
  complexity and functionality,
  understanding an object’s
  functions from its appearance
  grows harder
• Just by looking the user should know: State of the system + Possible actions