SIGGIAACS

1
SIGGI-AACS

• Smart Databases for smart users

2
etymology

• SIGGI is a neural agent name comes from
  sigmoid as in curve function and Sigmund as
  in Freudian
• A Archaeological
• A Auto-
• C classification
• S System

3
Freudian Interface

• SIGGI is a neural agent for our AI interface
• SIGGI is a surrogate for the archaeological
  expert
• SIGGI can learn with training
• SIGGIs ideas on classification change with new
  teachers and new information
• SIGGI will become an advisor, remain a surrogate,
  and perhaps, become a teacher some day

4
Smart databases are dynamic

• Design goals
• - create interoperable databases
• - allow variable views of scientific information
  to be supported, while ensuring electronic
  accessibility
• - allow the database to be self-correcting
• - develop standard tools that lead to building
  national infrastructure
• - smart databases will have analytical tools
  that use database information to make predictions

5
Data building

• Archaeological databases have
• - complex data types
• - data and data types and interpretations are
  constantly changing
• 2-fold data complexity
• - archaeologists need to organize the same data
  in different ways
• - archaeologists may differ on what are the
  requisite or important data

6
SIGGI AACS Project Objectives

• Establish a dynamic data model with smart user
  interfaces for large data sets
• Develop these models working closely with federal
  and state managers, tribal groups, researchers
  and public users
• Develop a smart neural agent for systematic
  classification of archaeological diagnostics
• Demonstrate that federal and state databases can
  be held securely
• Salvage old and fragmented databases into the new
  powerful and extensible data structure
• Store artifact data as live digital images
• Train archaeologists in use of the database
• Disseminate results in meetings and publications

7
Making Thoughtful Agents
8
How Does SIGGI Think?
9
Expert Knowledge

• Obtain information from domain expert
• Classify methods by interactions with domain
  experts
• Classify by types of information elicited from
  domain experts

10
SIGGI Prototype

• Domain Expert Lohse (1985)
• Rufus Wood Lake projectile point chronology
  used. Large collection with established
  provenience and radiocarbon dates.

11
Basic Training for SIGGI

• Lohse classification chosen because it was
  explicitly based on established types,
• had clean provenience information,
• a suite of radiocarbon dates,
• a clear analytical framework,
• and was statistically driven.

12
Rufus Wood Lake Projectile Point Chronology
Modal types - descriptive - historical
13
Lanceolate Forms Defined Metrically
14
Lanceolate Forms Defined Conceptually in
Dimensional Space
15
Lanceolate Forms Distinguished
16
Triangular Forms Defined Metrically
17
Triangular Forms Defined Conceptually in
Dimensional Space
18
BASIC TRAINING
19
EXPERT KNOWLEDGE BASED SYSTEMS ELEMENTS

• KNOWLEDGE DATABASE
• HUMAN DESIGNER/USER
• DIAGNOSTIC DESIGN
• INTELLIGENT BEHAVIORS
• lt CONTROLS gt
• - PERTINENT KNOWLEDGE
• - SEQUENTIAL DEDUCTIONS AND TRANSFORMS

20
EXPERT KNOWLEDGE BASED SYSTEMS KEYS

• KNOWLEDGE ACQUISITION
• KNOWLEDGE REPRESENTATION
• KNOWLEDGE UTILISATION

21
THE DATABASE
22
Database Structure
23
Recording Outlines Inputs
24
USER Access and Security
25
Reality Bytes We Cant Include Them All

• Sampling an inherent issue
• Ideal is unreachable impossible because the
  number of possible observations is infinite
• Must use precedent the Traditional
• Modern technology merely makes the issues of
  sampling and bias more evident

26
Content-based Image Retrieval

• Types of queries
• Level 1 retrieval of primitive features
  (color, texture, shape, spatial location)
• Level 2 retrieval by derived or logical
  features (by type or by object)
• Level 3 retrieval by abstract attributes
  (interpretation of forms)

27
Data Retrieval Different Questions
Type assignment of fragments
Type descriptions
28
USER INTERFACES
29
SIGGI USER SCREEN
30
LOADING IMAGES
31
IMAGE ANALYSIS
32
PROBABILITY ASSIGNMENT
33
To Dos

• Obtain larger samples of Plateau projectile
  points from known contexts
• Bring archaeological experts to train and
  interact with SIGGI
• Demonstrate research potential of AI neural
  networks in refining typological issues

34
Suggested References

• E.S. Lohse
• 1985 Rufus Woods Lake Projectile Point
  Chronology. In S.K. Campbell (ed.), Summary of
  Results, Chief Joseph Dam Cultural Resources
  Project, Washington, pp.317-364. Seattle Office
  of Public Archaeology.
• 1994 The Southeastern Idaho Prehistoric
  Sequence. Northwest Anthropological Research
  Notes 28(2)135-156.
• 1995 Northern Intermountain West Projectile
  Point Chronology. Tebiwa 253-51.
• 2003 Automated Classification of Stone
  Projectile Points in a Neural Network, with C.
  Schou, A. Strickland, D. Sammons and R. Schlader.
  Paper to be published in Proceedings of the 2003
  Computer Applications in Archaeology Conference,
  BAR International Series, W. Borner (ed.).
• 2003 Automated Classification a Neural Network
  Information Data Archives, with C. Schou, A.
  Strickland, D. Sammons, and R. Schlader. Paper
  presented at the Fifith World Archaeological
  Congress, Catholic University, Washington, D.C.
  In Managing Archaeological Resources Advancing
  Access to Digital Data Strategies for Preserving
  Archaeological Digital Records, M.S. Carroll, A.
  Simon and A. Prinke (co-organizers).
  http//imnh.isu.edu/wac5/