Terms and Concepts

Question 1

Tri-level hypothesis

Answer 1

When studying an intelligent system, we should investigate three levels:

• computational
• procedural/algorithmic
• implementational

Question 2

CRUM

Answer 2

Computational-Representational Understanding of Mind

• > Human cognition in terms of representational structures and computational procedures that operate on those structures
• > Evaluated on:
  1. Representational Power
  2. Computational Power
  3. Psychological Plausibility
  4. Neurological Plausibility
  5. Practical Applicability

Question 3

Expert Systems

Answer 3

• A system incorporating the knowledge of experts

- Rule-based with preprogrammed facts and procedures -> No machine learning

Question 4

BVSR theory (Campbell)

Answer 4

• > Blind Variation and Selective Retention
  1. Unsighted generation of ideas
  2. Elaborating whether the idea was useful

Question 5

Primary vs Secondary Thinking

Answer 5

Primary:

• analogical, free associative
• makes discovery of new combinations more likely
• flat associative curve, defocused attention

Secondary:

• abstract, logical, goal-oriented
• steep associative curve, focused attention

-> Creative People can alternate more between the two

Question 6

Associative Hierarchy (Steep vs Flat)

Answer 6

Steep associative hierarchy:

• Focused Attention
• Lateral inhibition
• Following one train of thought

Flat associative hierarchy:

• Defocused attention
• more lateral activation
• more widespread activation of nodes/ideas

Question 7

Klondike problems of creativity

Answer 7

1) Rarity - Payoff is rare in the conceptual space
2) Isolation - Areas of payoff are often independent from another in the problem space
3) Oasis - One might stay for long in one area of payoff because they are hard to leave
4) Plateau - The direction where the greater payoff lies might not be clear

Question 8

Fault Tolerance

Answer 8

A neural network might still give you the desired output even if the input was imperfect

Question 9

Delta Rule

Answer 9

Supervised learning rule that calculates the change in weight that needs to be done after a learning trial in order to get to the desired activity of the nodes.

Question 10

Auto Associator

Answer 10

Special type of pattern associator which aims to reproduce the input as the output and has recurrent connections.

• Knows which connections were active before and can predict sequences
• Often used in models of episodic memory like the hippocampus

Question 11

4-Stage Model of Information Processing

Answer 11

1. Sensory Processing - Acquisition and registration of multiple information sources
2. Perception/Working Memory - Conscious percept and manipulation of info in WM
3. Decision Making
4. Response Selection

Question 12

4 Stages of Automation

Answer 12

1. Information Automation
2. Analysis Automation
3. Decision Automation
4. Action Automation

Each stage can be on a high or a low level.

Question 13

Competitive Learning

Answer 13

The nodes in a network compete for the right response to a given input. The output with the largest activity receives the weight change.

• Form of unsupervised learning
• Based on Hebbian Learning

Question 14

When is a learning algorithm called “supervised”?

Answer 14

When it has to be fed (labeled) input and constraints regarding the output

Question 15

Drive-reinforcement theory was a solution to a problem of differential Hebbian learning. What was this problem?

Answer 15

Differential learning did not take time into account

Question 16

What kind of learning do pattern associators use?

Answer 16

Hebbian

Question 17

How can attractors and phase transitions be related to connectionist models?

Answer 17

When a neural network settles, weight changes decrease, acting like an attractor.