Download Intelligenza artificiale

Intelligenza artificiale
Introduzione al corso
Stefano De Luca
11 Ottobre 2010
Obiettivi del corso
 Introdurre le tematiche principali dell’intelligenza artificiale
(IA)
 Focus sul semantic web, ontologie e logica
 Introduzione al linguaggio PROLOG
 Introduzione a tecniche di IA non logiche, come reti neurali,
classificatori e clustering
 Alla fine del corso, gli studenti saranno capaci di progettare
un semplice agente intelligente, scegliendo la tecnica più
consona
Programma
 Introduzione all’IA
 Cos’è un agente; introduzione al semantic web
 Ricerche e pianificazioni
 Logica, semantic web e PROLOG
 Tecniche non simboliche:
 Reti neurali
 Algoritmi genetici
 Clustering. KNN e reti di Kohonen
 Classificazione. SVM
Chi insegna
 Docente: Stefano De Luca
 Si occupa di intelligenza artificiale da più di 20 anni
 Lavora in un’azienda specializzata in intelligenza artificiale, evodevo
 Responsabile di numerosi progetti che hanno usato ed usano tecniche di IA
 Pubblicazioni internazionali in particolare su agenti e semantic web
 Nel review board di riviste internazionali di IA ( International Journal of
Agent Technologies and Systems )
 Reviewer per conferenze internazionali di IA (AAMAS) e per pubblicazione
di libri internazionali
 Temi di ricerca: analisi dei testi, semantic web e semantic GIS, uso dell’IA
per sistemi di investigazione
 Ricevimento: alla fine della lezione, su appuntamento
 eMail: [email protected], [email protected]
Orientamento
 Il corso sarà orientato alla presentazione di idee più che di
elementi specifici di programmazione
 Alcune parti del corso verranno introdotte tramite il
linguaggio di programmazione Prolog
 Prequisiti al corso: non ci sono prerequisiti specifici
Libri di testo & letture
 Tutto il materiale del corso verrà distribuito sulla pagina del
corso stesso
 Verranno fornite delle dispense e degli articoli su cui studiare
(sempre in formato elettronico)
 Durante il corso, verranno proposte delle letture su temi
attuali dell’IA; gli articoli verranno forniti in formato
elettronico. È molto gradita la lettura di questi articoli, per
poter poi avviare una discussione in classe (ultima parte della
lezione di venerdì)
Modalità di esame
 L’esame consta di una parte scritta obbligatoria e una parte orale.
 Lo scritto è basato su esercizi e sulla descrizione della




progettazione di un agente intelligente
È previsto un esonero a metà corso; l’esonero sarà preceduto da
una esercitazione
Per superare l’esame, è necessario presentare un progetto su uno
dei temi che verrà fornito. Il progetto (per 2..4 persone) è relativo
alla messa in pratica di una delle tematiche trattate.
Chi vuole, può presentare una tesina su un argomento concordato.
Per i non informatici: il progetto non è obbligatorio, eventualmente
va sostituito dalla tesina.
Software necessari
 Per la creazione di ontologie:
 Protégé, http://protege.stanford.edu/
 Per il Prolog:
 SWI Prolog, http://www.swi-prolog.org/
 Sono entrambi open source e free, e disponibili su più sistemi
operativi.
Cos’è l’intelligenza artificiale
La “vecchia” IA
AI
10
18/10/2010
Ora: Smart Music System
 The Bose uMusic system uses
artificial intelligence to learn
the listening habits and
preferences of its users. Load
your CDs into the digital music
delivery system, it can hold
thousands of songs, and it will
learn your listening preferences
and prioritize your music
collection.
11
AI
18/10/2010
Pandora.com
AI
12
18/10/2010
Control systems
AI
13
18/10/2010
Google Instanst Search
Foundations of AI
 Different fields have contributed to AI in the form of
ideas,viewpoints and techniques.
 Philosophy: Logic, reasoning, mind as a physical system, foundations of learning, language and






AI
rationality.
Mathematics: Formal representation and proof algorithms, computation, (un)decidability,
(in)tractability, probability.
Psychology: adaptation, phenomena of perception and motor control.
Economics: formal theory of rational decisions, game theory.
Linguistics: knowledge represetatio, grammar.
Neuroscience: physical substrate for mental activities.
Control theory: homeostatic systems, stability, optimal agent design.
15
18/10/2010
Basi dell’IA
Robotica
Logica
Filosofia
IA
Sistemi
basati sulla
natura
ALife
Economia
Scienze
Sociali
Test di Turing
 When does a system behave intelligently?
 Turing (1950) Computing Machinery and Intelligence
 Operational test of intelligence: imitation game
 Test still relevant now, yet might be the wrong question.
 Requires the collaboration of major components of AI: knowledge, reasoning,
language understanding, learning, …
AI
17
18/10/2010
A brief history
 What happened after WWII?
 1943: Warren Mc Culloch and Walter Pitts: a model of artificial boolean
neurons to perform computations.
 First steps toward connectionist computation and learning (Hebbian learning).
 Marvin Minsky and Dann Edmonds (1951) constructed the first neural network computer
 1950: Alan Turing’s “Computing Machinery and Intelligence”
 First complete vision of AI.
 Idea of Genetic Algorithms
AI
18
18/10/2010
A brief history (2)
 The birth of (the term) AI (1956)
 Darmouth Workshop bringing together top minds on automata theory, neural nets
and the study of intelligence.
 Allen Newell and Herbert Simon: The logic theorist (first nonnumerical thinking program used for
theorem proving)
 For the next 20 years the field was dominated by these participants.
 Great expectations (1952-1969)
 Newell and Simon introduced the General Problem Solver.
 Imitation of human problem-solving
 Arthur Samuel (1952-)investigated game playing (checkers ) with great success.
 John McCarthy(1958-) :
 Inventor of Lisp (second-oldest high-level language)
 Logic oriented, Advice Taker (separation between knowledge and reasoning)
AI
19
18/10/2010
A brief history (3)
 The birth of AI (1956)
 Great expectations continued ..
 Marvin Minsky (1958 -)
 Introduction of microworlds that appear to require intelligence to solve: e.g. blocks-world.
 Anti-logic orientation, society of the mind.
 Collapse in AI research (1966 - 1973)
 Progress was slower than expected.
 Unrealistic predictions.
 Some systems lacked scalability.
 Combinatorial explosion in search.
 Fundamental limitations on techniques and representations.
 Minsky and Papert (1969) Perceptrons.
AI
20
18/10/2010
A brief history (4)
 AI revival through knowledge-based systems
(1969-1970)
 General-purpose vs. domain specific
 E.g. the DENDRAL project (Buchanan et al. 1969)

First successful knowledge intensive system.
 Expert systems
 MYCIN to diagnose blood infections (Feigenbaum et al.)

Introduction of uncertainty in reasoning.
 Increase in knowledge representation research.
 Logic, frames, semantic nets, …
AI
21
18/10/2010
A brief history (5)
 AI becomes an industry (1980 - present)
 R1 at DEC (McDermott, 1982)
 Fifth generation project in Japan (1981)
 American response …
 Puts an end to the AI winter.
 Connectionist revival (1986 - present)
 Parallel distributed processing (RumelHart and McClelland, 1986); back-
propagation
AI
22
18/10/2010
A brief history (6)
 AI becomes a science (1987 - present)
 Neats vs. scruffies.
 In speech recognition: hidden markov models
 In neural networks
 In uncertain reasoning and expert systems: Bayesian network formalism
 …
 The emergence of intelligent agents (1995 - present)
 The whole agent problem:
“How does an agent act/behave embedded in real environments with continuous sensory
inputs”
AI
23
18/10/2010
The Hype Cycle of Emerging Technologies (Gartner
2005)
AI
24
18/10/2010
Major research areas (Applications)
 Natural Language Understanding
 Image, Speech and pattern recognition
 Uncertainty Modeling
 Expert systems
 Virtual Reality
 …..
AI
25
18/10/2010
Symbolic Programming
 Program as Representation of world
 Symbol as basic element of representation
 atom, property, relationship




Symbolic Expression as method of combination
LISP for Symbolic programming
PROLOG for logic programming
Object-Oriented Concept
Knowledge Representation
 What kind of Knowledge needed for Problem solving ?
 Structure of knowledge ?
 declarative vs procedural
 Representation techniques ?
 explicit vs (implicit + inference)
 logic, frame, object-oriented, semantic net, script
 Knowledge acquisition and update
Search Theory
 An Optimization method
 Analyze alternative cases and select one
 Cope with Exponential complexity, NP classes
 Try likely one first (Heuristic Search)
 Utilize local information (Hill Climbing Method)
 Optimal solution vs good solution
 Genetic Algorithm, Simulated Annealing
 Stochastic search
Automated Reasoning
 Qualitative Reasoning
 Utilization of qualitative knowledge such as
 Non-monotonic Reasoning
 Ostrich flys ?
 Plausible Reasoning
 Information fusion under uncertainty
 Case-based Reasoning
 Utilization of Experience
Machine Learning
 Performance improvement by experience




How much of knowledge required to start learning ?
Method of acquiring new knowledge and merging it to existing knowledge-base
Role of teacher
Role of examples and experience
 Parameter Adjustment
 Inductive learning
 Computational Learning Theory
 Quality of generalization capability in terms of Training data
 Used in Practice such as Data Mining
Data Mining
Knowledge extraction for decision making
Data
AI
31
Information
/ knowledge
Decision
Making
18/10/2010
Neural Network
 Computational model of Neurons
 Power comes from Connection of simple processing element -
connectionism
X1
X2
.
.
.
Xn
w1
w2
wn
S
F(X1, X2, …, Xn)
Neural Network
 learning = link weigh adjustment
 Error-back-propagation : supervised learning
 Any Functional Mapping is learnable
 Strong at Sensory Data Processing
 Symbolic Grounding
 Old Horse on the race again
 Massive parallelism, graceful degradation
AI
33
18/10/2010
Neural Network Classifier
Job(1/0)
good
age
medium
Salary
bad
#mouth
Debt
Input
layer
AI
34
Hidden
layer
Output
layer
18/10/2010
Genetic Algorithms
 Computational model of life evolution
 Stochastic optimization technique
 Initial chromosome creation
 New generations are made (cross over, mutation)
 survival of the fittest
 Base of artificial life research
 study evolution of life, by simulation
AI Success Story (Planning)
 MARVEL (Schwuttke, 1992)
 Real-time Space shuttle Mission planning
 Berth assignment (KAL, 1997)
 Unmanned Vehicle
 Ground and air
 Pathfinder Rover, 1996
 Asimo – a walking robot
36
AI
18/10/2010
Autonomous Land Vehicle
(DARPA’s GrandChallenge contest)
AI
37
18/10/2010
AI Success Story : Medical expert systems
Programs listed by Special Field
 Antibiotics & Infectious










AI
Diseases
Cancer
Chest pain
Dentistry
Dermatology
Drugs &
Toxicology
Emergency
Epilepsy
Family Practice
Genetics
Geriatrics
38
Gynecology
Imaging Analysis
Internal Medicine
Intensive Care
Laboratory Systems
Orthopedics
Pediatrics
Pulmonology & Ventilation
Surgery & Post-Operative
Care
Trauma Management
18/10/2010
Pattern Recognition Applications
 Handwriting and document recognition
 forms, postal mail, historic documents
 PDA pen recognition
 Signature, biometrics (finger, face, iris, etc.)
 Gesture, facial expression
 As a Human computer intertraction
 EEG, EKG, X-ray
 Trafic monitoring, Remote Sensing
 Smart Weapon – guided missile, target homing
AI
39
18/10/2010
Handwriting Understanding
AI
40
18/10/2010
전자 펜으로 수식 입력
수식 인식
Decision Support Systems (DSS)
41
AI
18/10/2010
Intelligent Transportation Systems
AI
42
18/10/2010
Future of AI
 Making AI Easy to use
 Easy-to-use Expert system building tools
 Web auto translation system
 Recognition-based Interface Packages
 Integrated Paradigm
 Symbolic Processing + Neural Processing
 AI in everywhere, AI in nowhere
 AI embedded in all products
 Ubiquitous Computing, Pervasive Computing
AI
43
18/10/2010