1 / 31

Graphical Representations and Expertise: Archaeological Types

Graphical Representations and Expertise: Archaeological Types. Knowledge Elicitation Expert Systems and Autoclassification. Neural networks. Neural networks (which are purely fictional and arbitrary) strive for models of learning.

olathe
Download Presentation

Graphical Representations and Expertise: Archaeological Types

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Graphical Representations and Expertise: Archaeological Types Knowledge Elicitation Expert Systems and Autoclassification

  2. Neural networks • Neural networks (which are purely fictional and arbitrary) strive for models of learning. • Learning is understood as the process that brings forth a learned state. • A learned state contains information about environmental regularities. • Networks of learned states require explicit representations. http://www.jaist.ac.jp/~g-kampis/DBF/Dynamic_Models_of_the_Mind.html

  3. Representations • Are ways of transmitting information • Are indicative of expertise • Level of experience • Knowledge domain mastery • Application of task-specific solution strategies

  4. What are representations good for? • A representation is important only insofar as it has a causal effect, or in other words, if it is active. • Representations in neural networks and dynamical systems are anything but active. • Representations mostly act as filters of perception and lack own causal power. http://www.jaist.ac.jp/~g-kampis/DBF/Dynamic_Models_of_the_Mind.html

  5. Representations should work • A representation should ‘work’ in the sense that it should be able to determine what future representations are possible • This constitutes “knowledge” (cf. Chomsky against behaviorism) • New representations should be consequences (effects) of the semantic properties of existing ones (cf. Fodor inferential role semantics etc) • “Active” representations should be possible • dynamic models: establish learning through rules or associations; feedback to the learning process is the dynamic

  6. Central characteristics • Computational offloading: extent to which differential representations reduce problem-solving time • Re-representation: how different external representations with the same abstract structure reduce problem-solving time • Graphical constraining: how graphical elements in representations constrain the kinds of inferences that are drawn Scaife and Rogers (1996)

  7. Meanings of representation • Representation as process vs. representation as product • As process … transformations and preservations that occur in deriving the representation from what is represented • As product … structural characterizations of the representation, as image-like, mental model or proposition

  8. Archaeological types as “geons” • Geon: the fundamental local features of objects • Action basis of transferring expert information • Idealization of object components and their spatial organization is important in recognition of complex objects or shapes Interrelations of geometric shapes http://www.pigeon.psy.tufts.edu/avc/kirkpatrick/threegeon.htm

  9. Geons and classification • Biederman’s recognition-by-components theory – Biederman identified 36 different "geons", i.e., basic shapes (rectangles, circles, wedges, etc.) that can describe all objects. • We determine what we are looking at by matching of shapes for the best possible fit among our internal representations. • Biederman also found it necessary to explain what features help us to determine what an object is despite perceptual differences, and he referred to them as the five invariant properties of edges (curvature, parallel, cotermination, symmetry, and co-linearity). • Dickinsen et al. 1994 present an active object recognition strategy which combines the use of an attention mechanism for focusing the search for a 3-D object in a 2-D image, with a viewpoint control strategy for disambiguating recovered object features. • This requires a probabilistic search through a hierarchy of predicted feature observations, taking objects into a set of regions classified according to the shapes of their bounding contours. http://www.psybox.com/web_dictionary/Biederman.html http://citeseer.ist.psu.edu/dickinson94active.html

  10. Structure extraction: feature selection

  11. Visualization as polygons: ideal geometric forms Polygon: a closed figure formed by non-parallel lines. Polygons are important image relations: human vision completes curves to form enclosed regions. The coterminations are non-accidental.

  12. Archaeological geons: types • Two basic geometric shapes recognized: • Triangular • Lanceolate Lohse (1985)

  13. Mirroring recognized types: Triangular Finer geometric distinctions may be drawn

  14. Recognizing wholes from parts = triangular – tip, midsection, base = side-notched triangular – tip, midsection, base = notched side-notched base

  15. Lexicons and protocols • Formal definitions must be created • E.g., “B3. Blade Angle(degrees). The outside angle between the blade-haft juncture and the margin of the blade (axis A and line segment aA). If the blade margin is markedly convex, the blade angle follows the lower section near the blade-haft juncture. Otherwise, it follows the general trend of the blade margin. • Formal rules for analysis follow closely thereafter

  16. What are we measuring? Schema applied • Curvature • Parallel dimensions • Cotermination of diagnostic features • Symmetry • Co-linearity Lohse 1985

  17. Adding measurements to geometric forms: distinctions Lohse (1985)

  18. Description and Expert Knowledge: Lohse 1985 Classification Descriptive Classification Expert Classification Geon 1Geon 2Knowledge Base: Point Types Lanceolate Simple lanceolate "Large un-named lanceolate" Windust C Cascade A Cascade B Cascade C Shouldered lanceolate Windust A Windust B Lind Coulee Mahkin Shouldered Triangular Side-notched triangular Cold Springs Side-notched Plateau Side-notched Corner-removed Nespelem Bar Rabbit Island A Rabbit Island B Corner-notched Columbia Corner-notched A Quilomene Bar Corner-notched A Quilomene Bar Corner-notched B Columbia Corner-notched B Wallula Rectangular-stemmed Basal-notched Quilomene Bar Basal-notched A Quilomene Bar Basal-notched B Columbia Stemmed A Columbia Stemmed B Columbia Stemmed C

  19. Informational displays: lots of information … lots of choice Cascade A Cascade C Lohse (1985)

  20. Data summaries

  21. Visualizations Windust Rabbit Island Cascade Lind Coulee

  22. Geon – archetype – mental model … pattern matching

  23. Pattern Matching – Classification = assign large = assign small = assign early = assign late = assign Cold Springs Side-notched = assign Plateau Side-notched A9A007 A9A029 = assign Cascade A = assign Cascade B = assign Cascade C A9A032, A9A033, A9A034, A9A038, A9A043, A9A044 Windust Cave type specimens, SIGGI-AACS Project

  24. Lanceolate forms as abstract geometric shapes simple lanceolate compound lanceolate 11 Large lanceolate 15 Windust C 23 Cascade C 21 Cascade A 22 Cascade B 12 Lind Coulee 13 Windust A 14 Windust B

  25. Lanceolate Forms as abstract shapes and paradigmatic descriptions

  26. Other Lanceolate Forms as abstract shapes and paradigmatic descriptions

  27. Expert knowledge elicitation project • Northwestern Plains Point Classification Conference • 6 archaeological experts to participate • Each brings an authenticated data base • Each brings 1 advanced student • Preconference: each expert submits digital images and data for training SIGGI; each submits drawings of type geons • Conference: 6 sessions over three days • Result: creation of an accepted classification system for Northwestern Plains

  28. Session Goals • Session 1: compare geons, discuss types and classification systems • Session 2: use decision center to elicit knowledge to construct typology using geons • Session 3: experts interact with SIGGI directing classification of geons • Session 4: experts interact with SIGGI to classify grouped image data • Session 5: experts in conjunction with SIGGI review and evaluate classification • Session 6: experts discuss outliers in the classification and finalize the Northwestern Plains projectile point classification

  29. Ontology or classification is the goal • Ontology: philosophical meaning: • re-usable representations as in a database of context independent (or cross-contextual) concepts or objects • Machine translations may be impossible beyond a certain level on grammar and lexicon alone • assume that most meanings in the mind are metaphoric • To ‘understand’ metaphors we need a built-in knowledge of real world objects and properties in detailed, text-independent, ontological form

  30. Project objectives • Significance: • Archaeological research • Knowledge elicitation protocols • Use of computer decision center • Dissemination using an educational web site • Participating archaeologists are stakeholders • Creation of an agreed upon measurement and classification system • Projection of how this data should be managed and used in the future • Will establish a clear agenda for future research

  31. Recommended References • Dickinson, S., Christensen, H., Tsotsos, J., and Olofsson, G. (1994). Active Object Recognition Integrating Attention and Viewpoint Control, Proceedings, ECCV '94, Stockholm, May, 1994. http://citeseer.ist.psu.edu/dickinson94active.html • Kampis, George. Dynamic Models of the Mind. http://www.jaist.ac.jp/~g kampis/DBF/Dynamic_Models_of_the_Mind.html • Lohse, E.S. (1985). Rufus Woods Lake Projectile Point Chronology. In Summary of Results: Chief Joseph Dam Cultural Resources Project, Washington, Sarah Campbell (editor), pp.317-364. Report to the U.S. Army Corps of Engineers. Office of Public Archaeology, University of Washington, Seattle. • Lohse, E.S., C. Schou, A. Strickland, D. Sammons and R. Schlader (2004). Automated Classification of Stone Projectile Points in a Neural Network, with In Magistrat der Stadt Wien – Referat Ulturelles Erbe – Stadarchaeologie Wien (eds.), Enter the Past: The E-Way into the four Dimensions of Cultural Heritage, BAR International Series 1227, pp. 431-433. Oxford. • Scaife, Mike and Yvonne Rogers (1996). External cognition: how do graphic representations work? Int. J. Human-Computer Studies 45: 185-213.

More Related