Making Information Systems

imminentpoppedΤεχνίτη Νοημοσύνη και Ρομποτική

23 Φεβ 2014 (πριν από 3 χρόνια και 8 μήνες)

68 εμφανίσεις


Making Information Systems
Intelligent

Dr. Geoffrey P Malafsky

TECHi2

2

The Need


Information overload


Time compression


Uncertainty


Proactive decision making and actions

3

What is Intelligence


Turing test


Reasoning


Accuracy


Fusion and
transformation of
inputs


Sensor


Data


Learning

4

Time and Certainty

5

Intertwined Complex Information


Example from DARPA
Evidence Extraction & Link
Discovery


Today’s Situation: ~10k
messages/day from multiple
sources read by multiple
analysts and analyzed in
multiple manual non
-
integrated tools


Similar to Social Network
Analysis

6

Knowledge is Personal

“Set the soldering iron to 350 degrees”


information

from manual for general use


knowledge

from expert for specific
manufacturing process


“If the soldering iron is even 20 degrees
hotter or colder, the connection will fail
and the part will be returned and
eliminate all profit. Watch carefully for the
color of the solder”

7

Taxonomy Complexity

80. INTERDISCIPLINARY PHYSICS AND RELATED AREAS OF SCIENCE AND
TECHNOLOGY
81.
Materials science
81.05.

t
Specific materials: fabrication, treatment, testing and analysis










Superconducting materials, see 74.70 and 74.72










Magnetic materials, see 75.50










Optical materials, see 42.70










Dielectric, piezoelectric, and ferroelectric materials, see 77.80










Colloids, gels, and emulsions, see 82.70.D, G, K respectively










Biological materials, see 87.14
81.05.Bx
Metals, semimetals, and alloys
81.05.Cy
Elemental semiconductors
81.05.Dz
II–VI semiconductors
81.05.Ea
III–V semiconductors
81.05.Gc
Amorphous semiconductors
81.05.Hd
Other semiconductors
81.05.Je
Ceramics and refractories (including borides, carbides, hydrides, nitrides,
oxides, and silicides)
81.05.Kf
Glasses (including metallic glasses)
81.05.Lg
Polymers and plastics; rubber; synthetic and natural fibers; organometallic
and organic materials
81.05.Mh
Cermets, ceramic and refractory composites
81.05.Ni
Dispersion-, fiber-, and platelet-reinforced metal-based composites
81.05.Pj
Glass-based composites, vitroceramics
81.05.Qk
Reinforced polymers and polymer-based composites
81.05.Rm
Porous materials; granular materials
8

What We Need: IT Conversations

From James Hendler, Agents and the Semantic Web, IEEE Intel Sys, Mar/Apr 2001

9

Current Technology Performance

Aspects of Knowledge Discovery

State
-
of
-
Art

Beyond

State
-
of
-
Art

Far

Beyond

State
-
of
-
Art

Status

Knowledge

Representation

Data

Volume

Human
-
Computer

Interaction

Naïve

Discovery

Advanced

Discovery

Guided

Discovery

Complexity

Complex Relational
Information


Relations across time
and space for people,
places & things

Vast


>10
6

attributes, links,
nodes

Iterative

Incremental

Active Learning



Unspecified, evolving
problem

Simple Relational
Information


Relations among
people, places &
things

Substantial


10
3

-

10
4

attributes,
links, nodes

Interactive


User
-
specified
problem, with
suggested retargeting


Some prior knowledge

Propositional
Information


Simple attributes for
people, places &
things

Minimal


100s of attributes,
links, nodes

Negligible


User
-
specified
problem


No prior knowledge

10

Performance
Maturity
Augmented Cognition: Large
KB+Models+Human engineering
Intelligent Systems
Natural
Language+Ontology
Search/
classification
Human
Density
Current Performance

11

Systems Engineering: Matching
Functional Components

12

Coupling to the Human

13

DARPA Augmented Cognition

14

Multisensor Fusion

15

DARPA EELD: Knowledge Creation
Technologies

Knowledge
Acquisition

Facts

(Database)

Upper

Ontology

Core
Theories

Domain
-
Specific
Theories/Models

Evidence
Extraction

Knowledge
Engineering

Link
Discovery

AI/KR

Expert

Text

Documents

Patterns

(models)

(e.g. HPKB)

Domain

Expert

(e.g. RKF)

Labeled

Examples

P

P

P

P

P

P

P

P

P

P

Positive

Examples

Negative

Examples

N

N

N

N

N

N

N

N

N

N

Pattern
Learning

16

Semantic Web


Create a Web where
information can be
“understood” by
machines as well as
humans


Must convey machine
-
accessible semantics

17

Ontology Contains Context and
Relationships

-

Madache, Schnurr, Staab &
Studer, Representation Language
-
Neutral Modeling of Ontologies

18

Integrated Presentation

19

DRAFT OV
-
1