100 likes | 216 Views
This overview discusses ongoing research in machine creativity spearheaded by Simon Colton and his team at the Universities of Edinburgh and York. The research focuses on various aspects of creativity including literary, scientific theory formation, and cognitive modeling. Notable projects include the HR program which generates concepts, proofs, and computational humor. Collaboration with experts like Alison Pease and Roy McCasland enhances the exploration of creativity in mathematical domains and bioinformatics. The multi-agent approach aims to tackle complex problem-solving in large datasets.
E N D
Machine Creativity Research @ Edinburgh Simon Colton Universities of Edinburgh and York
Overview • Players • Research • Contacts • Possibilities
Creativity Researchers • Graeme Ritchie • Literary creativity, assessment of creativity • Simon Colton • Scientific theory formation • Alison Pease • Cognitive modelling • Alan Bundy? • Roy McCasland?
Graeme Ritchie • Literary/Linguistic creativity • Computational humour • With Kim Binsted: JAPE joke generator • See Binsted PhD, AISB’00 paper • Assessment of creative programs • Take into account the inspiring set • Fine tuning, creative set (with Pease & Colton) • Shotgun approach • See AISB’01 paper, ICCBR’01 workshop paper
Simon Colton The HR program Overview • Scientific theory formation • Implemented in the HR program • Starts with ML-style background info • Invents concepts (definitions and examples) • Makes, proves, disproves hypotheses • Used in mathematical domains • Integrates with ATP, CAS, CSP, Databases • Applied to mathematical discovery
The Application of HR • Number theory • Invention of integer sequences & theorems • Constraint invention (with Ian Miguel) • Speed up CSPs, 10x for QG4-quasigroups • ATP (with Geoff Sutcliffe) • Lemma generation, theorems to break provers • Puzzle generation • Study of machine creativity • Cross-domain, meta-theory, multi-agent, interestingness
HR for Bioinformatics • HR is now independent of maths • Theory extends to other sciences • E.g., making of empirically false hypotheses • Multi-agent approach for large datasets • Machine learning problems • Concept identification: forward look-ahead • Prediction: uses the whole theory • Very preliminary • Application to ML datasets • Comparison of methods next
Alison Pease • Phd proposal: • A computational model of mathematical creativity via Interaction • Using HR to perform cognitive modelling • Multi-agent setting (see IAT paper) • Lakatos-style reasoning • Fixing faulty hypotheses (see ECAI paper) • Conjecture-driven concept formation • Implications for creativity • Fit into Boden’s framework (see ICCBR’01 paper)
Contacts • Edinburgh • UK national centre for E-science (GRID) • Bioinformatics group • York • Machine learning group • Imperial • Bioinformatics group (Muggleton)
Possibilities • Problem with Large Datasets • Multi-agent creativity (split data) • Domain knowledge • Cognitive Modelling • HR applied to Bioinformatics • Serious Case Study (Roy McCasland) • EPSRC 1-year fellowship (fingers crossed) • Using HR to Study Zariski Spaces