Download presentation
Presentation is loading. Please wait.
Published byMaryann Harris Modified over 8 years ago
1
CSC 110 - Intro. to Computing Lecture 22: Artificial Intelligence
2
Knowledge Representation As video hinted at, AI has big problem representing knowledge Must also be precise enough to capture information But must be able to handle ambiguity Consider the following sentence: Buffalo buffalo buffalo Buffalo buffalo
3
Semantic Network One method of representing knowledge Focuses on relationship between items Typically shown as a directed multi-graph Vertices (points) represent the data items Each data item should be a simple noun Edges (lines) are relationships between items Each edge is usually a verb Relationships go in one direction (shown as arrow) Items can have multiple relationships
4
Semantic Network Example Nosipho is a student. Nosiphostudent instance of
5
Semantic Network Example Students are people. Nosiphostudent Students are people. person is a instance of
6
Semantic Network Example Students go to school. Nosiphostudent Students go to school. person is a school go to instance of
7
Semantic Network Example Nosiphostudent person is a school go to Each person has a gender. gender has a instance of
8
Semantic Network Example Nosiphostudent person is a school go to Nosipho is female gender has a female instance of
9
Semantic Network Example Nosiphostudent person is a school go to Hillary Clinton is also female gender has a female Hillary Clinton instance of
10
Semantic Network Example Nosiphostudent person is a school go to Females are people. gender has a female Hillary Clinton type of instance of
11
Semantic Network Example Nosiphostudent person is a school go to gender has a female Hillary Clinton instance of Dog is not a Nosipho is a person is not a type of
12
instance of Semantic Network Example Nosiphostudent person is a school go to gender has a female Hillary Clinton instance of Nosipho is a person Dog is not a type of
13
instance of Semantic Network Example Nosiphostudent person is a school go to gender has a female Hillary Clinton instance of Nosipho is a person Dog is not a type of
14
Semantic Network Example Nosiphostudent person is a school go to gender has a female Hillary Clinton instance of Hillary Clinton is not a dog Dog is not a type of
15
Semantic Network Example Nosiphostudent person is a school go to gender has a female Hillary Clinton instance of Hillary Clinton is not a dog Dog is not a type of
16
Semantic Network Example Nosiphostudent person is a school go to gender has a female Hillary Clinton instance of Dog is not a type of
17
Semantic Network Example Nosiphostudent person is a school go to gender has a female Hillary Clinton instance of Dog is not a What is Nosipho’s gender? is not a type of
18
Semantic Network Example Nosiphostudent person is a school go to gender has a female Hillary Clinton instance of Dog is not a What is Nosipho’s gender? is not a type of
19
Semantic Networks Good for problems asking relationships between data Bad for playing checkers or chess Checkers has 10 9 different states to express Difficult, but doable Chess has 10 43 different possible states Lot of relationships to express Big jump from knowing how horsey moves to winning a chess match
20
Search Trees Consider game of Nim Players take turns place 1 – 3 marks on board Whomever marks last slot, wins “Best” move can be hard Need to know what your opponent will do Then depends upon how you move and your opponents response to the move, and so on
21
Search Trees Following a line from one state to another is like making a move Shows all possible games of Nim
22
Search Tree Limits Search trees really used to play games Provides easy method of solving problem But evaluating entire tree is difficult Checkers: 10 9 different possible game states Chess: 10 43 different game states Evaluating entire tree could require years… Need to re-evaluate tree for each move. Cannot even store entire tree on a computer!
23
Tree Searches Depth-first: Could I possibly win/lose if I make this move? Breadth-first: Which move will put me in the best position in 3 rounds?
24
Search Trees
25
Computer vs. Human Humans are big and slow Have 100 billion neurons Neurons have many inputs Neurons “run” at 100 Hz Computers are small and fast Have 100 million gates Gates combine 1 or 2 inputs Gates run at 1,000,000,000 Hz
26
Neural Networks Simulate how a biological neural network learns How it ultimately makes decisions often does not make sense to people Each artificial neuron assigns numerical weight to each input “Learns” by adjusting these weights
27
Artificial Neuron Operation Effective weight of an input is the input value multiplied by input’s weight If sum of effective weights is above threshold neuron outputs 1 When sum of effective weights is less than the threshold, neuron outputs 0
28
Artificial Neuron Example Consider the following neuron: Value to compare against threshold: ValueWeightEffective Weight 10.3 1-3 02 14
29
Artificial Neural Network Train neural network on variety of input values When output is correct, share reward (increase weight of active inputs) When output is wrong, “punish” network (decrease weight of inputs) Apply occasional “reverse brain damage”
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.