Intelligent Database Systems Lab Presenter : YAN-SHOU SIE Authors : MARCO PIASTRA NN Self-organizing adaptive map: Autonomous learning of curves and surfaces from point samples
Intelligent Database Systems Lab Outlines Motivation Objectives Methodology Experiments Conclusions Comments
Intelligent Database Systems Lab Motivation In here we want from a point cloud image to reconstruct it original structure, but preliminary version SOAM algorithm is can not effective to produced the expected topology.
Intelligent Database Systems Lab Objectives In here we present a improve version SOAM algorithm, its has a much more predictability and includes some new concepts.
Intelligent Database Systems Lab Methodology Topological and geometrical background Term – homeomorphic – manifold – Voronoi cell – Delaunay triangulation
Intelligent Database Systems Lab – Restricted Delaunay complex : – Homeomorphism and ε –sample – Witness complex Methodology
Intelligent Database Systems Lab Methodology – Finite sets of witnesses and noise Growing self-organizing networks – Positioning the units: ‘gas-like’ dynamics adaptation strategy of the first kind
Intelligent Database Systems Lab second kind of strategy – Competitive Hebbian learning and dynamic units – Growing networks, insertion threshold Methodology
Intelligent Database Systems Lab Self-Organizing Adaptive Map (SOAM) – Stateful units Methodology
Intelligent Database Systems Lab – Adaptive insertion thresholds – The SOAM algorithm Methodology
Intelligent Database Systems Lab Suppose that we have a document with four concepts: ‘Ad,’‘Bert,’ ‘Cees,’ and ‘Dirk.’ If the window size is 2, the following windows are created for this document: {Ad}, {Ad, Bert}, {Bert, Cees},{Cees, Dirk}, and {Dirk}. ex : ‘System’ appears in documents {1,3,6,8} and windows {1,5,10,14,18,20,28}; ‘Process’ appears in documents {1,3,6,12} and windows {1,5,12,14,18,25,30}. the similarities are converted to distances: Methodology-distance measures -document co-occurrence similarity -window-based similarity window similarity : document similarity : Avg = 0.15
Intelligent Database Systems Lab Experiments Experimental setup
Intelligent Database Systems Lab Experiments – Algorithm behavior
Intelligent Database Systems Lab Experiments – Performances
Intelligent Database Systems Lab Experiments – Undersampling and noise: when things go wrong – Boundaries and non-manifold units
Intelligent Database Systems Lab Conclusions The SOAM algorithm represents an interesting alternative to deformable models in that it can effectively deal with changes in topology and execution speedup.
Intelligent Database Systems Lab Comments Advantages -SOAM can be dynamically self-growth, and the results will be generated close to the result we want, for the field of 3D technology has considerable value.. Applications - medical imaging, 3D sample, etc.