HOW DOES
HUMAN-LIKE KNOWLEDGE COME INTO BEING IN
ARTIFICIAL ASSOCIATIVE SYSTEMS?
AGH UNIVERSITY OF SCIENCE AND TECHNOLOGY
Faculty of Electrical Engineering, Automatics, Computer Science and Biomedical Engineering
Department of Automatics and Biomedical Engineering Unit of Biocybernetics
POLAND, 30-059 CRACOW, MICKIEWICZA AV. 30
Adrian Horzyk
horzyk@agh.edu.pl
Can we represent knowledge?
HUMAN-LIKE KNOWLEDGE
Knowledge allows to:
Remember facts, rules, objects or classes of them.
Consolidate various facts and rules after their similiarities.
Associate objects, facts, rules with contexts of their occurences.
Recall facts and rules using context and associations.
Generalize objects, facts and rules.
Be creative using learned classes of objects, facts and rules.
Various facts and rules can be associated and recalled thanks to:
Similarities of the data that define them.
Subsequences of the data that occur inside them.
Knowledge is active aggregation of data, facts and rules that can be recalled and generalized according to the context of their recalling.
Human-like knowledge can be represented only in reactive systems that can represent such not redundant aggregations.
WHAT IS NOT KNOWLEDGE ?
Knowledge:
Is not a set of facts, rules, objects or classes of them.
Is no kind of a computer memory or a database.
Does not remember everything precisely.
Cannot be collected alike data, facts and rules but it can be formed for given or collected data, facts and rules.
Cannot be easy transfered from one system to another alike data, databases, facts and rules etc. Only pieces of information, facts and
rules can be transfered into another system. Can be partially transfered through recalled facts and rules.
Is not limited to any set of facts, rules or objects because new, creative input contexts can lead to new facts, rules, notices, observations and remarks on the basis of the same knowledge.
Knowledge can be automatically formed only in special systems that allow to activelly associate and aggregate data, facts
and rules, and their various combinations and sequences.
NEURAL ASSOCIATIVE SYSTEMS
Neural associative systems allows to:
Represent various objects, facts and rules in a unified form of data combinations using neurons.
Create classes of represented objects after most representative features and their combinations.
Trigger neurons according to the context of other activated neurons or sense receptors.
Use the context of previously activated neurons according to the time that has elapsed from their activations.
Consolidate and combine various objects, facts and rules after their similiarities and subsequences.
Associate objects, facts, rules with contexts of their occurences.
Recall associated objects, facts, rules using new or previously used contexts, questions etc.
Generalize and even create new objects, facts and rules.
ARTIFICIAL
ASSOCIATIVE SYSTEMS
Artificial associative systems:
Model biological neural associative systems, nervous systems etc.
Define associative model of neurons (as-neurons) that are able to reproduce context and time dependencies of biological neurons.
Can be simulated, trained and adapted on today’s computers.
Can use various training data set
and even sets of training sequences.
Can reproduce training sequences or create new ones - be creative!
Can generalize at various levels:
Object
level Sequence level
Artificial Associative Systems and
Associative Artificial Intelligence (Polish)
Transformation of database tables into associative structures ASSORT creates the basis graph structure of associative systems.
Adrian Horzyk,horzyk@agh.edu.pl, AGH University of Science and Technology Adrian Horzyk,horzyk@agh.edu.pl, AGH University of Science and Technology
TABLE
Transformation of database tables into associative structures ASSORT creates the basis graph structure of associative systems.
Adrian Horzyk,horzyk@agh.edu.pl, AGH University of Science and Technology Adrian Horzyk,horzyk@agh.edu.pl, AGH University of Science and Technology
TABLE
Transformation of database tables into associative structures ASSORT creates the basis graph structure of associative systems.
Adrian Horzyk,horzyk@agh.edu.pl, AGH University of Science and Technology Adrian Horzyk,horzyk@agh.edu.pl, AGH University of Science and Technology
TABLE
Transformation of database tables into associative structures ASSORT creates the basis graph structure of associative systems.
Adrian Horzyk,horzyk@agh.edu.pl, AGH University of Science and Technology Adrian Horzyk,horzyk@agh.edu.pl, AGH University of Science and Technology
TABLE
Transformation of database tables into associative structures ASSORT creates the basis graph structure of associative systems.
Adrian Horzyk,horzyk@agh.edu.pl, AGH University of Science and Technology Adrian Horzyk,horzyk@agh.edu.pl, AGH University of Science and Technology
TABLE
Transformation of database tables into associative structures ASSORT creates the basis graph structure of associative systems.
Adrian Horzyk,horzyk@agh.edu.pl, AGH University of Science and Technology Adrian Horzyk,horzyk@agh.edu.pl, AGH University of Science and Technology
TABLE
Transformation of database tables into associative structures ASSORT creates the basis graph structure of associative systems.
Adrian Horzyk,horzyk@agh.edu.pl, AGH University of Science and Technology Adrian Horzyk,horzyk@agh.edu.pl, AGH University of Science and Technology
TABLE
Transformation of database tables into associative structures ASSORT creates the basis graph structure of associative systems.
Adrian Horzyk,horzyk@agh.edu.pl, AGH University of Science and Technology Adrian Horzyk,horzyk@agh.edu.pl, AGH University of Science and Technology
TABLE
Transformation of database tables into associative structures ASSORT creates the basis graph structure of associative systems.
Adrian Horzyk,horzyk@agh.edu.pl, AGH University of Science and Technology Adrian Horzyk,horzyk@agh.edu.pl, AGH University of Science and Technology
TABLE
Transformation of database tables into associative structures ASSORT creates the basis graph structure of associative systems.
Adrian Horzyk,horzyk@agh.edu.pl, AGH University of Science and Technology Adrian Horzyk,horzyk@agh.edu.pl, AGH University of Science and Technology
TABLE
Transformation of database tables into associative structures ASSORT creates the basis graph structure of associative systems.
Adrian Horzyk,horzyk@agh.edu.pl, AGH University of Science and Technology Adrian Horzyk,horzyk@agh.edu.pl, AGH University of Science and Technology
TABLE
ASSOCIATIVE NEURAL GRAPH CONSTRUCTION for training sequences: S1, S2, S3, S4, S5
ASSOCIATIVE NEURAL GRAPH CONSTRUCTION for training sequences: S1, S2, S3, S4, S5
ASSOCIATIVE NEURAL GRAPH CONSTRUCTION for training sequences: S1, S2, S3, S4, S5
ASSOCIATIVE NEURAL GRAPH CONSTRUCTION for training sequences: S1, S2, S3, S4, S5
ASSOCIATIVE NEURAL GRAPH CONSTRUCTION for training sequences: S1, S2, S3, S4, S5
ASSOCIATIVE NEURAL GRAPH CONSTRUCTION for training sequences: S1, S2, S3, S4, S5
ASSOCIATIVE NEURAL GRAPH CONSTRUCTION for training sequences: S1, S2, S3, S4, S5
ASSOCIATIVE NEURAL GRAPH EVALUATION The external excitation of neuron E4 triggers the following
activations of neurons: E4 E5 E2 E6
We got sequence S2 as the answer for the external excitement of neuron E4:
THE SIMPLE NEURAL STRUCTURE OF THE
CONSECUTIVE LINGUISTIC OBJECTS representing 7 sentences
Neural associative structure for the linguistic objects
Response to „What is knowledge?”
As-neurons are consecutively activated after training sequences and give the answers:
Knowledge is fundamental for intelligence.
Knowledge is not a set of facts and rules
ASSOCIATIVE MODEL OF NEURONS
Associative model of neurons AS-NEURON:
Works in time that is crucial for all associative processes in the network of connected as-neurons.
Models relaxation and refraction processes of biological neurons
Relaxation – continuous gradual returning to its resting state
Refraction – gradual returning to its resting state after activation
Optimizes its activity responces for input data combinations
chosing only the the most intensive and frequent subset of them.
Conditionally plastically changes its size, synaptic transmission and connections to other as-neurons.
Can represent many similar as well as quite different combinations of input stimuli (data).
CONCLUSION
Knowledge can be modelled using artificial associative systems.
Training sequences can be used to adapt artificial associative systems
Associative systems supply us with ability to generalize on various levels:
classes created for objects
sequences describing facts and rules
Associative systems can be creative according to
the context, which can recall new associations.
?
Questions? Remarks?
Google: Horzyk Adrian
horzyk@agh.edu.pl
Theory of neural
associative computations and knowledge engineering
in the associative systems
Artificial Associative Systems and
Associative Artificial Intelligence (Polish)