Design for IDSS Liam Page CSE 435 23

Design for IDSS Liam Page CSE 435 23 October 2006

What is design? n n Construction of an artifact from single parts that may be either known and given or newly created for this particular effect (Börner 1998) Design systems assist a user in producing better designs in shorter amount of time

What is design? n How does design help with: ¡ decreasing design times? ¡ increasing design quality? ¡ improving design predictability?

Classifying Design Task n Three classifications: ¡ Routine Design ¡ Innovative Design ¡ Creative Design

Routine Design n n State space is well defined using potential designs New designs can be derived entirely from existing designs Outcomes known before hand Final design agrees with configurable constraints Used mostly in KB-systems

Innovative Design n Well defined state space of potential designs, non-routine design desired Values for variables may change Solution is similar to old designs, but also appears to be new due to variables

Complex Design n n Non-routine design New variables ¡ Extends/moves state space of potential designs

Complex and Innovative Tasks (1) n n n Often unsure what the final design constraints will be Typically ordered in accordance to preference criteria Abstract -> Concrete ¡ Reduction of design space

Complex and Innovative Tasks (2) n Ideal system ¡ ¡ ¡ Assists user, not automated User interface logically constructed for type of design task Learns from past solutions and user’s response to solutions (accept, correct, refuse)

Case Based Design n Themes of case based designed systems (Maher and Gomez de Silva Garza 1997) ¡ ¡ ¡ representation and management of complex cases case augmentation using generalized design knowledge formalization of informal knowledge

Case Based Design n What can be a complex case? ¡ ¡ ¡ Sample of larger data model Data represented structurally (graphs) Non-static variables Flexible – may have multiple interpretations Adaptable to solve new problems

Case Based Design n Implications of complex cases ¡ ¡ ¡ Must be able to reinterpret and reformulate new problems Overlapping of problem and past cases must be identified Parts must be chosen for transfer and combination Similarity functions must be flexible Joint consideration of case aspects is possible

Example of Complex Case Usage Case: Deluxe. Bathroom 1 Transformed Solution • Dimensions = 30’ x 30’ Case: Deluxe. Bathroom 2 • Dimensions = (20’-40’)x (20’-40’) • Doors = 1 • Dimensions = ( 30’ 50’)x (30’x 50’) • Doors = 1 – 2 • Doors = 2 – 3 • Outlets = 4 – 6 • Outlets = 6 • Outlets – 6 – 10 • Hot tub = yes • Deluxe Standing Shower = yes • …

Case Based Design n Generalized design knowledge to augment cases ¡ Includes causal models, state interactions, heuristic models/rules, geometric constraints ¡ Typically not available for innovative and creative tasks

Case Based Design n Need formalization of knowledge for CBR automation Problem: human knowledge of design is difficult to formalize into rules and variables that the system can utilize In cases where it is only possible to create an informal body of knowledge, system should be developed to merely support a human designer

Knowledge Representation n Four knowledge containers in CBR ¡ ¡ Vocabulary Case base Similarity measure Solution transformation

Vocabulary n Vocabulary – task and domain dependent ¡ ¡ Should capture all important features of design Supports problem solving in relevant domain

Case Base n n Represent past design experience Usage – abnormal/normal Granularity – grain size of cases is equal to grain size of design task Level of Abstraction ¡ ¡ ¡ Ossified cases – general rules of thumb Paradigmatic cases – represent learned expertise Stories – complex, relate to large number of circumstance

Case Base (cont) n Perspective ¡ ¡ State-oriented – case represents problem and solution Solution-path – case refer to problem or operator that determines solution from problem description

Similarity Measure n Two different approaches to similarity assessment ¡ ¡ Computational (similarity) approach Representational approach

Computational Approach n n Unstructured organization Usefulness of cases based on presence or absence of features Many cases Are Called – candidate cases Few Are Chosen – structural comparison between problem and possible solutions

Representational Approach n n n Pre-structured case base (indexing structure) Neighboring cases are assumed to be similar Probes constraints in memory to determine possible solutions

CBR for Innovative and Creative Design n Flexible case retrieval ¡ ¡ ¡ n Retrieved cases show similar aspects to the problem Different similarity measures have to be dynamically composed during retrieval Fish and Shrink Algorithm Structural similarity assessment ¡ Structural cases are processed and represented as variables taking the role of problem or solution variables

Solution Transformation and Case Adaptation n n New situations often different from old solutions Solutions must be adapted to fit the constraints of the problem using parts from other past solutions

Solution Transformation and Case Adaptation n Three kinds of adaptation (Cunningham and Slattery 1993) ¡ ¡ ¡ Parametric adaptation – modifying parameters Structural adaptation – adaptation operators (grammar rules) Generative Adaptation – reuse and adaptation for derivations of past problem -solving episodes

Fish and Shrink n n n Algorithm for flexible case retrieval Allows for rapid searching through case base (even if significant aspects are combined at query time) Can be stopped at any time and still produce usable results (though not complete)

Fish and Shrink original case → αname → Ωname T 1 Ωname δname distances C 1 n Similarity measure of emphasized attributes between all cases and a set of test cases are retrieved and stored

Fish and Shrink (2) Represents similarity distances between cases and emphasized attributes n n n Reduce range of possible similarity of any case to problem by utilizing the predetermined similarity to test cases Find similarity distance from test cases to problem Use predetermined similarity of cases to test cases to derive the possible similarity of cases to problem Reduce similarity range to a single estimate by overlaying similarity ranges to test case

Structural Similarity n n Used to solve design problems involving a representative structure Determines candidate solutions via maximal common subgraph (mcs)

Structural Similarity n Several functions are required ¡ ¡ Compile – translates attribute representations of objects and relations into graphs Recompile – converts graph back to attributes that may be depicted graphically Retrieve – gets candidate cases Match – finds mcs between graphs

Structural Similarity n n Best mcs transferred to problem Vertices and edges of other candidate cases may be used to augment solution

Structural Similarity

Structural Similarity n Arrows represent spatial relations (touches, overlaps, etc)

Case Study – EADOCS n EADOCS ¡ ¡ Interactive, multi-level, and hybrid expert system for aircraft sandwich panel structures Structure of design defines the set of components, their configuration and parameter values

EADOCS (2) n Innovative design ¡ ¡ Plans for designing components are not available Only partial models for evaluating behavior are available

EADOCS (3) n Object Oriented class structure ¡ ¡ Design cases are instances of design problems containing objects that define its behavior For EADOCS, cases contain knowledge of the structural behavior of the design, such as an ability for a material to maintain its shape at a particular air pressure

EADOCS (4) n Retrieving a solution 1. 2. 3. Best solutions are selected and configured into prototype solutions A best prototype defining an optimal design space is selected and a conceptual solution is retrieved If no conceptual solution fitting the requirements can be retrieved, next best prototype is selected and 2 is repeated

EADOCS (5) n Case Combination ¡ ¡ Sub-targets are identified within the conceptual solution that do not match the design requirements New target for retrieval is defined Cases are retrieved to satisfy the new target Adaptations are retrieved based on differences in functionality between cases with a similar structure to the conceptual solution and the case satisfying the new target

EADOCS (6)

Final Remarks n IDSS can significantly help with design tasks by: ¡ ¡ ¡ Decreasing design times by automating aspects of the design process Increasing design quality by insuring constraints of design are respected Improving the predictability of designs by using learning algorithms to reduce design space

References n Arcos, J. L. and Enric Plaza. “The ABC of adaptation: Towards a Software Architecture for Adaptation-Centered CBR Systems. ” 12 November 1999. 22 October 2006 <http: //www. iiia. csic. es/Projects/cbr/ABC/abc-report. html> n Bergmann, Ralph. “Experience Management for Electronic Design Reuse. ” Experience Management : Foundations, Development Methodology, and Internet-Based Applications. Springer Berlin/Heidelberg, 2002. 2 August 2003. 6 October 2006. n Börner, Katy. “CBR for Design. ” Case-Based Reasoning Technology: From Foundations to Applications. Springer Berlin/Heidelberg, 1998. Springer Link. 20 May 2003. 6 October 2006.