Towards the Learnable Technologies in Complex Systems

1
Towards the Learnable Technologies in Complex
Systems

• ( research Activities
• of
• Computational Intelligence Group
• at CIT TU Kosice )

2
AI program at TU Kosice

• Department of Cybernetics and AI
• Faculty of EE Informatics
• TU Kosice, 40-th anniversary 2009

• Bc Course in Intelligent Systems
• MSc. Course in Artificial Intelligence
• PhD Course in Artificial Intelligence

3
What is Artificial (machine) intelligence ?
4
What is intelligence ?

• It is very complex notion but .....
• Intelligence is ability to learn from
  experience
• What is Artificial Intelligence number of
  tools of AI

5
Basic features of Intelligent systems and
Intelligent technologies
6
What are the main features of Intelligent
Systems ???
knowledge representation archivation
learning
reasoning - problem solving
Intelligent Systems
Intelligent Technologies
7
Why Intelligent ???

• Intelligence Knowledge
• (Biologically inspired systems, brain-like
  systems)
• Knowledge in data (neural networks)
• Knowledge in experience (fuzzy logic)
• Knowledge in state space heuristic search,
  chaos (evolutionary computing, evolution)

8
What type of tasks could be under consideration
using intelligent technology ?
9
What type of problems ?

• Classification from data and human experience
• Modelling from data or human experience
• Prediction forecast
• Optimalization finding the optimal values
• Human-machine interface (human-centered)

10
Can you call your product intelligent ?? Why not
???Will it have impact on demand and sales
???????
11
Basic principles of neural technology
12
What kind of neural networks we do have
Feedforward NN
Recurrent NN
output
Input
output
layer
neurons
hidden
layer
Synaptic weights
13
Basic approaches in neural technology

• Supervised training by examples
• so you are getting neural network a tool for
  classification, modelling , prediction and etc.
• you have to have a data
• (input-output examples)
• Unsupervised training
• so you are able to neural network for clustering,
  dimentionality reduction, compression etc.
• (only input data)

14
What type of problems especially with NN

• classification
• neural control more nelinearity
• prediction problems based on history
• signal tranformation
• clustering in hyper-dimensional space
  (diagnostic applications)
• many other

15
Basic principles of fuzzy technology
16
What is a fuzzy system ??

• Based on fuzzy logic fuzzy sets

It is good for expressing verbal values (small
people, mid-size people, tall people)
1
tall
small
mid
150
140
168
175
Height of people
17
Why is fuzzy set important ?

• You are able to describe a experience or
  behavior of the system in the form of
• IF .............. THEN ............... rules
• e.g.
• IF a car is big AND car is expensive THEN car
  is fast

Preposition
Consequens
18
Fuzzy system (controller) basic tool
De-fuzzification
Crisp output
Rule - Base (made by expert)
inference
Experience From the Expert
fuzzification
Crisp input
19
Where is good fuzzy logic ??

• Modeling e.g.
• experience in washing

Washing machine
In case when you are not able to get model and
you Are able to describe behavioral model by
fuzzy rule
Behavioral Model of the system
In case if you want to incorparate experience of
the expert In the system e.g. predictions,
decisions etc.
20
Application domains

• Transportation (cars, trains, traffic
  management...)
• Computing with words
• Internet information retrieval
• Fuzzy measures
• image processing, databases
• Control easy to design (if you have an expert)
• Felling sensors Keise problem description by
  fuzzy, etc

21
Basic principles of evolutionary technology
22
What are the basic tools in evolutionary
computation ?

• Genetic algorithms - optimalization tool
• Genetic programming
• system for data analysis with aim to provide
  analytical expression
• Based on biological inspiration and Darwin theory
  of evolution

23
Basic tools in Evolutionary computation

• Chromozomes encoding the problem
• Fittness function Evaluation function
• Operator mutation , selection, cross operator
  ...
• Some chromozomes survive, some are destroyed
• To find in heuristic way the best values

24
Evolution evolutionary solutions

• Interactive Evolutionary approach
• So e.g. you envolve design of the event
• Make few iterations
• Stop human will influence the evolution
• Envolving towards

25
Evolutionary programming
GP
Analytical expression
Data
Function approximation abilities
26
Where to use Evolutionary Technology

• Optimisation problems in general
• Planning and scheduling (TSP problem)
• GP for data-mining with aim of analytical
  expression
• Broad range of engineering, business and other
  applications

3Ghz ???????
Problem time consuming process
27
Multi Agent Systems as concept

• Not a tool !!!

28
Short Introduction

• What is agent? What is Multiagent system?

• Agent fully autonomous entity with ability to
  learn from environment, to communicate and
  cooperate with other agents METAGENT
• Common architectures for agents
• Reactive architec. (mostly on CI tools)
• Deliberative architec. (mostly on Symbolic
  tools)

29
What is Multiagent System?

• MAS system consisting of more (usually various
  types) agents solving common problem
• No one of the agents involved in the system is
  able to solve problem its self
• Agents have to communicate and cooperate to solve
  the problem
• The role of META-AGENT

30
Multi-robotic SystemsBenchmark

• Motivation
• too complex tasks for single robot
• more flexible, scalable, fault-tolerant
• study of group behavior (social sciences,
  cognitive sciences, life sciences)
• developed for real dynamic environments with high
  level of uncertainty

31
What is the basic feature of these technologies
to have them useful ???
Basic feature is
Universal aproximation theorem
Universal unknown function approximators
32
Some more tools of intelligent technologies
All tools related to Machine(Artificial
Intelligence)

• NN, FS, EC
• Expert systems
• Logic programming tools
• Tools of chaos theory
• Tools of Artificial Life
• Tools of Multi-Agent technologies
• Etc .......

33
When to use Intelligent technologies ?

• Answer is simple

Only if the application of IT will make a
product more advance and successful on the
market (money) Nobody cares what technology
but new technology must be better It is
general belief that Intell. Technology is able
to do it
34
So, If Machine Intelligence will help to
Humanoid to be more useful we should use MI
if not DO NOT use it !!!
35
Towards the World Incremental Learnable
Knowledge Intelligence - (WILKI)

• we do offer it as an working idea for a proposal
  or for a part of proposal to be implemented into
  the research proposal as a part of the project
• we do offer our framework as a startpoint for
  further development in building domain oriented
  Intelligent System
• we do believe that incremental learning and MAS
  (soft understanding) is an important way of
  building intelligent System with cognition and
  big knowledge base able to use it when ever you
  need
• we do offer an extensive knowledge in building
  recognition systems with incremental learning and
  ability to share information (neuro, fuzzy,
  evo-expertise)
• impact IF successful a project Engineering
  principles and starting to build a WILKI
  computer server and worldwide Agents - so
  similar to central computer server in I, Robot
  movie with large knowledge and ability to handle
  number of agents (e.g. robots).

36
Basic general idea
Server
Methods
Knowledge
Database
Central Server
Client1
Client2
Client3
Clientn
Agent1
Agent2
Agent3
Agentn
37
Research Challenges for the project

• To develop (further) and validate the scientific
  foundations, engineering principles and
  approaches required to build systems with the
  above described capabilities
• Many problems
• distribution of intelligence between server and
  clients it will be living procedure
• achieve that IF one Client gets knowledge the
  others HAVE IT !!!!
• gradual, incremental, learning with
  zero-forgetting factor or if needed forgetting
  remember forgetting
• extraction domain oriented knowledge from general
  methods
• Knowledge and information fusion and integration
  problems
• HCI of Agents using Kansei approaches
• Inclusion of Expert knowledge into Central server
  (could be a grid) for Agents use
• the systems should be build in software integrity
  we do propose C language and platforms e.g.
  PDAs or Smartphone, Robots, .. Flexibility and
  domain orientation, lately the domains can be
  linked.

38

• To build a strong basis for research on ways of
  reaching the long term goals
• Integration of various methods on Agent and
  Central server sides
• Define a MIQ for a system how to measure a
  quality of Intelligent System
• Agents could be various types (different software
  Interface in Agent) e.g. Aibo, Spykee
• Link to application world e.g. E-health
  assistant, monitoring assistant

Long term goal living and learnable networked
assistants for various human activities very
much of inspiration of movie I, Robot
39
Further Work

• Development of integrated interface for MAS (and
  MRS) simulations
• Development of methods for pattern recognition
  and scene analysis
• Development of methods for hybridization of
  reactive deliberative architectures of agents

40
NAO humanoid robots
41
(No Transcript)
42
(No Transcript)
43
ASIMO the ROBOT
22.8.2003
arrived with the Japanese delegation and,
although not human, is receiving more attention
than Premier Junichiro Koizumi sorry ?
the smallest member of the delegation a mere
120 centimeters tall is named ASIMO and is a
third-generation humanoid robot made by Honda
44
NAO new challenge in Humanoid Robotics

• It would be very interesting to create NAO family
  at TU Koice
• ? ? ?

45
Thank you for your attention !!!

• Peter.Sincak_at_tuke.sk
• TU Kosice, Slovakia , EU
• http//www.ai-cit.sk

Peter John
Sophia

The best Intelligent Systems made by myself ?)