Download Introduction

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Artificial intelligence in video games wikipedia , lookup

Ecological interface design wikipedia , lookup

Computer Go wikipedia , lookup

Visual servoing wikipedia , lookup

Turing test wikipedia , lookup

Human–computer interaction wikipedia , lookup

Soar (cognitive architecture) wikipedia , lookup

AI winter wikipedia , lookup

Enactivism wikipedia , lookup

Visual Turing Test wikipedia , lookup

Computer vision wikipedia , lookup

Machine learning wikipedia , lookup

Agent-based model wikipedia , lookup

Neural modeling fields wikipedia , lookup

Concept learning wikipedia , lookup

Pattern recognition wikipedia , lookup

Intelligence explosion wikipedia , lookup

Knowledge representation and reasoning wikipedia , lookup

Ethics of artificial intelligence wikipedia , lookup

Existential risk from artificial general intelligence wikipedia , lookup

Cognitive model wikipedia , lookup

Embodied cognitive science wikipedia , lookup

Philosophy of artificial intelligence wikipedia , lookup

History of artificial intelligence wikipedia , lookup

Transcript
Knowledge Based Systems and
Artificial Intelligence
Introduction
Aleksandra Pizurica
Department Telecommunications and Information Processing (TELIN)
Image Processing and Interpretation (IPI) Group
Statistical Image Modeling Lab
Lecturer
 Prof. Aleksandra Pizurica
 Department Telecommunications en Information processing (TELIN)
• Research group Image Processing and Interpretation (IPI)
– Statistical Image Modeling Lab
 Research: statistical image modeling, multiresolution representations of
images, image and video restoration; understanding and modeling visual
perception and visual intelligence
 Teaching: Computer graphics, Knowledge based systems and artificial
intelligence
 Office: Sint-Pietersnieuwstraat 41, (Technicum), TELIN, 2.05
 Email: [email protected];
 Tel: 09 264 34 15
 Web: http://telin.ugent.be/~sanja
2
Study material
• Slides (available on Minerva)
• Book:
S. Russel and P. Norvig:
Artificial Intelligence – A Modern Approach
(denoted as [R&N] in the slides)
• Some other articles/chapters from books
 C. Bishop, Pattern Recognition and Machine Learning, Ch. 8: Graphical Models
3
Exam
• Written exam (2/3)
 Theory (closed book)
 Problem solving (open book)
• Projects/computer exercises (1/3)
4
What is AI?
Systems that think like humans Systems that think rationally
Systems that act like humans
Systems that act rationally
Four categories of AI definitions
5
Acting humanly: The Turing test
• The Turing test (Alan Turing, 1950) was designed to provide a
satisfactory operational definition of intelligence
• Suggested major components of AI:




Knowledge representation (store what it hears or knows)
Automated reasoning (use the stored info to draw conclusions)
Machine learning (adapt to new circumstances; detect and extrapolate patterns)
Language processing (e.g., able to communicate in English or another language)
• Extension - total Turing test includes video to test perceptual abilities
6
Thinking humanly: Cognitive Science
• If the goal is to have a program that thinks like a human, we need to
determine first how humans think
 Introspection (catch our own thoughts as they go by)
 Psychological experiments (observing a person in action)
 Brain imaging (observing the brain in action)
• Cognitive science aims at constructing testable theories of human
mind using experimental psychology and computer models
 Scientific theories of internal activities in the brain at different levels of abstraction
 Validation of these theories can be
• Predicting and testing behavior of human subjects (top-down)
• Direct identification from neurological data (bottom-up) –cognitive neuroscience
• AI, cognitive science and cognitive neuroscience are separate fields that
fertilize each other
7
Thinking rationally: Laws of Thought
• Aristotle (384 – 322 BC): codifying the “right thinking” (irrefutable reasoning)
 His syllogisms provided patterns for argument structures that always yielded correct
conclusions when given the right premises.
• E.g., “Socrates is a man; all men are mortal; therefore, Socrates is mortal”
 Studying these lows of thought (supposedly governing the operation of mind) initiated
the field of logic.
• Direct line through mathematics and philosophy to modern AI
 Logicians in the 19th century: a precise notation for statements about all kinds of
objects and relations among them
 By 1965 programs existed for solving “in principle” any problem in logistic notation
• Logicist tradition in AI is still present. Problems:
 Not easy to take informal knowledge and state it in the formal terms of logistic
notation, especially when the knowledge is not entirely certain
 Exhausting the computational resources (solvable “in principle” ≠ solvable in practice)
8
Acting rationally
• Rational behavior: doing the right thing
• The right thing: that which is expected to maximize goal achievement, given
the available information
• Doesn't necessarily involve thinking (e.g., blinking reflex) but thinking should
be in the service of rational action
 Making correct inferences (reason logically) is sometimes part of acting rationally
 But, correct inference is not all of rationality
• The approach of acting rationally has two important advantages over
the other listed categories of AI definitions:
 More general (e.g. all the skills needed for Turing test are needed to act rationally;
also more general than “lows of thought”, which are only one of possible
mechanisms for acting rationally)
 Much better suited for formal mathematical description than approaches based on
human behavior and human thought
9
Rational agents
• Agent: an entity that perceives and acts (from Latin agere, to do)
• Rational agent is one that acts so as to achieve the best outcome, or when
there is uncertainty, the best expected outcome
• This course is about designing rational agents
• Abstractly, an agent is a function from percept histories to actions:
• For any given class of environments and tasks, we seek the agent (or class of
agents) with the best performance
• In practice, computational limitations make perfect rationality unachievable
 design best program for given machine resources
10
The foundations of AI
11
History of AI
12
State of the art in AI
• Robotic vehicles
 Driverless robotic cars e.g. in 2005 DARPA Grand Challenge (driving on a
rough terrain) and in 2007 Urban Challenge (obeying all traffic regulations
while negotiating with other traffic and obstacles)
• Speech recognition
• Autonomous planning and scheduling
 E.g. NASA’s remote agent program (1999)
“It's one small step in the history of space flight. But it was one giant leap for
computer-kind, with a state of the art artificial intelligence system being given
primary command of a spacecraft”
http://ti.arc.nasa.gov/tech/asr/planning-and-scheduling/remote-agent/
• Game playing
 DEEP BLUE defeated world champion Garry Kasparov (1997)
13
State of the art in AI
• Spam fighting
 Algorithms learn to classify messages as spam
• Logistics planning
• Computer aided diagnosis
 E.g. in radiology the computer output is already routinely used as a
"second opinion" in assisting radiologists' image interpretations
• Robotics
 E.g. robotic vacuum cleaners, robotic lawn mowers,… but also in more
delicate domains, like robotic surgery
• Machine translation
• …
14
Contents of this course
1. Introduction to artificial intelligence (R&N, ch. 1,2)
 Foundations and history of AI
 Intelligent agents
2. Problem solving
 Solving problems by searching (R&N, extracts from ch. 1-5)
 Constraint satisfaction problems (R&N, ch. 6)
3. Knowledge, reasoning and planning
 Planning (R&N, ch. 10-11)
 Knowledge representation (R&N, ch. 12)
15
Contents of this course, contd.
4. Uncertain knowledge and reasoning
 Uncertainty and probabilistic reasoning (R&N, ch. 13 en 14)
 Reasoning over time and hidden Markov models (R&N, ch. 15)
 Graphical models and inference (Bishop*, ch. 8)
 Simple and complex decisions (R&N, ch. 16 en 17)
5. Learning (R&N, ch. 18)
 Learnig from examples
 Artificial neural networks
6. Perception
 Image perception and visual intelligence
*[Bishop]: C. Bishop, Pattern Recognition and Machine Learning
16
Intelligent Agents
(R&N, ch. 2)
17
Problem Solving
(R&N, ch. 3)
18
Uncertainty and probabilistic reasoning
(R&N, ch. 14, 15)
• Why uncertainty?
 Too complex, non-deterministic, partially observable environment
• Probabilistic reasoning
 Knowledge representation taking into account uncertainty
• Basic principles of Bayesian networks
A arrows:
causal
nodes: random influence
variables
B
C
D
Bus too late
Train too late
Too late at work
Joost too late
Late for the meeting
19
Statistical learning and Hidden Markov models
(R&N, ch. 15, 20)
• Bayesian learning methods
 Maximum Likelihood (ML); Maximum A Posteriori (MAP); Naive Bayes
• Dynamic Bayesian Networks
 Kalman filters, Hidden Markov Models (HMM)
• Markov Random Fields
• Inference
 Expectation-Maximization (EM), particle filters, MCMC samplers
20
Graphical models and inference
(C. Bishop: Pattern Recognition and Machine Learning, hoofdstuk 8)
Belief propagation
Factor graph
Bericht
Bericht
http://ai.cs.washington.edu/people/nath
21
Simple and complex decisions
(R&N, ch. 16, 17)
• Decision networks, influence diagrams
 Combining utility theory and probability theory
 Acting based on what we want and what we know
• Sequential decision problems
• Multiple agents: Game theory
22
Learning from examples
(R&N, ch. 18)
Decision tree
Input: a vector of attributes
Output: a single decision
Example: waiting for a free table in a raetaurant [R&N, Fig. 18.2]
23
Learning from examples
(R&N, ch. 18)
Artificial neural networks
R&N, Fig. 18.20 (b)
R&N, Fig. 1.2
24
Modeling visual perception
• Mathematical models of visual
perception
• Important for
 Understanding the human visual
system
 Building real “intelligent” visual
procesors
T. Poggio et al
http://www.csail.mit.edu/
25