290 likes | 413 Views
Concept Formation in a Design Optimization Tool Wei Peng and John S. Gero 7, July, 2006. Outlines. Design Optimization Concept formation Concept formation from a situated lens A situated agent-based design optimization tool The agent’s experience and concept formation engine
E N D
Concept Formation in a Design Optimization Tool Wei Peng and John S. Gero 7, July, 2006
Outlines • Design Optimization • Concept formation • Concept formation from a situated lens • A situated agent-based design optimization tool • The agent’s experience and concept formation engine • Prototype system • Testing results and future direction
Design Optimization • Three major tasks • Interactive process • Design knowledge requirement • Application scenario – how the agent learn to recognize design optimization problem
Design Optimization Knowledge • Recognition of appropriate optimization model is fundamental to design decision problems • Can be expressed into semantic relationships between design elements • For example • Focus on learning and adapting the knowledge of recognizing an optimization problem
Concept Formation (CF) • Concept learning – given a set of examples of some concept/class/category, determine if a given example is an instance of concept • Concept formation – incremental unsupervised acquisition of categories and their intentional descriptions • Concept in designing – a consequence of the situatedness of designing
Concept – Coupled Interactions in Designing Situated Agent Sensor Concept Formation Experience Effector Designer Virtual Knowledge Flows between two Worlds Interactions in Designing
Concept Formation through a Situated Lens • Situatedness – notion of conceptual situations that are based on the observers’ experiences and inseparable from interactions (Dewey, 1902) • The concept formation process – the way agent orders its experience in time (Clancey,1999) as conceptual coordination • Concept formation framework –in a situated agent (Gero and Fujii, 2000)
time t C what I’m-doing-now C2 Perceptual Categorization 2 C1 Perceptual Categorization 1 time t’ Situated Concept Formation Situated concept formation Concept as higher order categorization of a sequence
A Situated Agent I • A situated agent contains sensors, effectors, experience and a concept formation engine • A concept formation engine consists of a perceptor, a cue_Maker, a conceptor, a hypothesizer, a validator and related processes • Sense data takes the form of a sequence of actions and their initial descriptions S (t) {…… “click on objective function text field”, key stroke of “x”, “(”, “1”, “)”, “+”, “x”, “(, “2”, “)”…} • Percepts are intermediate data structures of environment states with multimodal information. It can be described as (Objective Function Object, Objective_Function, “x(1)+x(2)”)
A Situated Agent II • Proto-concepts are initial or intermediate concept structures • Tree or rule structures • Hypotheses depict the agent’s explanations about failures in correctly predicting a situation • Backward chaining rules • Validationallows concepts and hypotheses to be evaluated in interactions • Concepts are grounded proto-concepts or hypotheses • Invariants about the agent’s experience
Concept Formation I Recast Concept Formation in A Constructive Memory Model
Concept Formation II Recast Concept Formation in A Constructive Memory Model
System Architecture Situated Agent-based Design Optimization Tool
Test I • Using similar design tasks – linear programming
Test II • Using novel design optimization scenarios • {L, Q, Q, L, NL, Q, NL, L, L, NL, Q, Q, L, L, L} • Initial experience – a quadratic experience • Behaviour charts and characteristics • Performance (prediction rate) for a static, reactive and situated system:
Summary and Future Work • Concept formation in a situated agent • New concept (new knowledge structure) • Interaction plays a role in shaping structures and behaviours • Co-evolution relation between structures and behaviours • Future direction 1: maintaining user models in design interactions • Future direction 2: learning from enriched contexts in design optimisation
The End Thanks!