Semantic Web - University of Huddersfield

Download Report

Transcript Semantic Web - University of Huddersfield

Artificial
Intelligence
CHA2555
Lee McCluskey
Email [email protected]
CW3/10
Resources on:
http://scom.hud.ac.uk/scomtlm/cha2555/
CHA2555 - changed
Change from last year.
The module specification has been updated :
last year you had 2 perspectives (Term 1
Symbolic AI, Term 2: Subsymbolic with 2
different lecturers) This year we are
integrating the course with 1 lecturer (me)
Effectively this will mean less emphasis on
Neural Networks ..
Overview
Resources:

http://scom.hud.ac.uk/scomtlm/cha2555/
The course contains a combination of theory and practical in the area of (mostly
symbolic) artificial intelligence
“My brain is a learning neural network” (Terminator 2)
No, its more likely to be symbolic AI …. ;-)
Overview First Term
Practical
 Prolog – an AI programming language
 Automated Planning Programs
 Games Programs
Theory
 Knowledge Representation, Logic, Search,
Heuristics, Automated Reasoning
 Planning Algorithms and Representation
 2 person games algorithms
Overview Second Term
Tentative
 Knowledge Engineering
 Machine Learning
 Language Understanding
With applications such as Games, Semantic Web
and UAVs …
Assessment
Practical Coursework given out Term 1, hand in
Term 2 - 40% of assessment
Exam is 3 hours, and 60% of assessment
You have to do 4 Questions out of 6
c.1 out of 2 for semester 1
c.3 out of 4 for semester 2
Artificial Intelligence – its about
three aspects
1. Intelligent abilities
2. Applications embedding intelligent
abilities
3. Techniques for
implementing 1. in 2.
In this course we will study 3.
Artificial Intelligence –
Intelligent Abilities







Sensing eg Seeing, hearing, recognising
Understanding eg language understanding
Communicating eg language generation
Having beliefs, desires, intentions
Reasoning and Problem Solving
Planning and Acting to achieve goals
Learning
Example Application Areas
image processing
Bar Code
Bar code
ANPR
–
Example Application Areas
Chatbots, Language
Translators..
Bar code
E.g.GOOGLE TRANSLATE
My son has grown another foot =>
Mon fils a grandi un autre pied.
Example Application Areas
UAVs
Mars Rover ->
Bar code
Mission Control…
Move from X to Y
Pickup Rock
Perform Experiment
etc
Example Application Areas
Narrative Generation
Initial state: (at shylock venice-rialto), ...
.....
4: (RECEIVE-LOAN-REQUEST SHYLOCK BASSANIO VENICE-RIALTO)
5: (MAKE-BUSINESS-DECISION SHYLOCK BASSANIO VENICE-RIALTO)
6: (RESPOND-TO-LOAN-REQUEST SHYLOCK BASSANIO VENICE-RIALTO)
7: (RECEIVE-DINNER-INVITATION SHYLOCK BASSANIO VENICE-RIALTO)
8: (REFUSE-DINNER-INVITATION SHYLOCK BASSANIO VENICE-RIALTO)
9: (RECEIVE-LOAN-REQUEST SHYLOCK ANTONIO VENICE-RIALTO)
10: (EXPRESS-ANGER-AT-PERSECUTION SHYLOCK ANTONIO VENICE-RIALTO)
11: (ASK-ABOUT-LENDING-WITH-INTEREST SHYLOCK ANTONIO VENICE-RIALTO)
12: (RESPOND-TO-LOAN-REQUEST SHYLOCK ANTONIO VENICE-RIALTO)
13: (LEND-MONEY-AS-FAVOUR SHYLOCK ANTONIO VENICE-RIALTO)
C2: (sealed-bond-over-loan shylock antonio)
....
19: (SHOW-DESPAIR-AT-ELOPEMENT SHYLOCK SHYLOCK-RESIDENCE)
C1: (shown-despair-at-elopement shylock)
...,
29: (ASK-FOR-JUSTICE SHYLOCK DUKE COURTROOM)
30: (SPEAK-OF-JUSTICE SHYLOCK ANTONIO DUKE COURTROOM)
31: (SPEAK-OF-PERSECUTION SHYLOCK ANTONIO COURTROOM)
32: (RECEIVE-MERCY-REQUEST SHYLOCK ANTONIO COURTROOM)
33: (SHOW-MERCY SHYLOCK ANTONIO COURTROOM)
34: (RECEIVE-VERDICT-MERCY SHYLOCK ANTONIO COURTROOM)
C3: (received-verdict-of-court shylock)
[extract from a presentation
by Dr Julie Porteous, Univ of
Teeside ]
20/07/2015
...,
38: (SHOW-SADNESS-OVER-FAMILY SHYLOCK SHYLOCK-RESIDENCE), ...
40: (END-OF-PLAY SHYLOCK)
Goal: (end-of-play)
University of Huddersfield
Example Application Areas
Robotics
Still huge challenges, but “low level”
autonomous behaviour is now
becoming well established
(example – NASA’s latest
robonauts)
Robotic Football ;-)
20/07/2015
University of Huddersfield
Picture from www.carbonated.tv
Techniques
Artificial Neural Networks
A network of “simple” processing units that can be trained to
simulate complex processing eg recognition
Hidden Layers
Each link has an adjustable
weight
Each node takes inputs and
produces an output
INPUT
NODES
OUPUT
NODES
A FEED-FORWARD ANN
Techniques
Artificial Neural Networks ..
are really “sub-symbolic” techniques – like evolutionary
computing (genetic algorithms) or swarm intelligence
(connectionist approaches..)
Their main advantage is their “robustness” or lack of
brittleness and their potential to scale-up.
ANNs are techniques within the area of Soft
Computing which is primarily aimed at solving
complex problems with techniques that allow for
uncertainty, imprecision, approximation ..
Techniques in
Symbolic AI...
In essence …
Use Symbols to represent objects in the world;
Use Logic to represent assertions about objects;
Use automated inference to simulate reasoning
with assertions;
Use heuristics to overcome complexity problems
Fundamental Assumption of
Symbolic AI No 1:

To simulate intelligent behaviour you need
VERY HIGH LEVEL DATA STRUCTURES
EXPLICITLY REPRESENTING KNOWLEDGE
HIGH LEVEL
Special Logics – Modal, Temporal etc
First Order Logic – relations, properties, V, &, =>, not, variables, quantifiers, terms
Description Logic – classes, membership, properties, disjunction
Objects – state, inheritance, aggregation, polymorphism
Sets, maps, relations, RDBs
pointers, arrays, records
Numbers, characters
Bits, bytes
LOW LEVEL
Machine Oriented
Fundamental Assumption of
Symbolic AI No 2:

To simulate intelligent behaviour you need
ALGORITHMS THAT REASON WITH
(REPRESENTATIONS OF) KNOWLEDGE
These algorithms are often
 “SEARCH” - based and
 “HEURISTIC”
Symbolic AI Platforms
To investigate symbolic AI we need a HIGH LEVEL
PLATFORM to do so.
We choose the programming language PROLOG to
do so:
 It has very high level data structures
 It is “easy” to implement reasoning / search
algorithms
Practical this week – self –
study: introduction to Prolog
Prolog is a very high level, logical, declarative
language useful for experimenting and prototyping
AI algorithms.
Prolog programs are lists of Rules and Facts.
Practical: Work through the file “notes” as directed on
the website
http://scom.hud.ac.uk/scomtlm/cha2555/
Summary
The course is (mainly) about Symbolic approaches
to AI
Fundamental to symbolic AI is the use of

High level logic-based data structures

Algorithms which reason with logic-based data

In symbolic AI, symbols represent entities in the
outside world

We will use Prolog as a Platform for Symbolic AI