Artificial Intelligence Flashcards
Programs
Implement algorithms
–written in specific programming languages
Algorithm
Sequence of instructions, effective procedure
–independent of programming language
ANN
Artificial Neural Network
Parts of a ANN (neural net)
Nodes, links
Nodes
- activation value
- thresholds (nodes can fire if input exceeds threshold)
- units in a network
- like cell body of a neuron
Links
- connections b/w nodes
- weights
- like synapses of neuron
Weights
Number associated with that link
- can be positive or negative
- numerical value b/w -1 and 1
- closer to |1| = heavier weight
Basis Function
Receives inputs, calculates summation (like cell body of neuron)
Sj = (sum) aiwji
Inputs
Each connected node’s activation value multiplies by connection weight of link b/w the two
Identity function
When activation same as input
Simple network (feed forward)
Flow of activation = forward only
Hidden layer
Only communicate with units INSIDE network (don’t get external input or send output outside)
Backpropogation
- initial response compared to target response
- target provided by “teacher”
- actual result subtracts from target result
- -error signal (target - actual)
- -feeds back to network
Definition of a robot (4 things)
- mechanical creature, can function w/ no operator
- can adapt to changing environment
- can function even if some parts break
- can move and change things
hard stuff
Can robot do things like have conversation, interact in human like way?
Top to bottom: start w/ language, interaction
Easy stuff
Movement in an environment
Bottom to top: start w/ interaction w/ environment
The uncanny valley
(Hard stuff)
- as we get closer to a robot being more human like…things get creepier
- -more human like, more positive response
- -gross
- -positive
Mobots
Mobile robots
- modeled on organisms (e.g. Cockroach)
- adapt to uneven terrain
Early robotic systems based on 3 primitive functions
- Sense
- takes data from sensors - Plan
- takes sense data, creates task to execute - Act
- sends commands to manipulator
Reactive paradigm
Get rid of planning, just sense and act
Subsumption architecture
Planning is subsumed (absorbed) by sense/act interaction
Series of layers for different behaviors
Layer 1: use sonar to detect obstacles, move away from them
Layer 2: wander around