F EATURE E XTRACTION & D ATA C UBE V ISUALIZATION T HROUGH T OPOLOGICAL D ATA A NALYSIS Paul Rosen Assistant Professor University of South Florida In collaboration with: Bei Wang Phillips, University of Utah Chris Johnson, University of Utah Jeff Kern, NRAO Betsy Mills, San Jose State (formerly NRAO)
• I NTRODUCTION TO THE TEAM Paul Rosen Bei Wang Phillips Chris Johnson Betsy Mills Jeff Kern Assistant Professor Assistant Professor Distinguished Professor Assistant Professor CASA Lead U of South Florida U of Utah U of Utah San Jose State NRAO PhD in CS PhD in CS PhD in Physics PhD in Astronomy PhD in Astrophysics
• M O ’ D ATA M O ’ P ROBLEM
• M O ’ D ATA M O ’ P ROBLEM
• M O ’ D ATA M O ’ P ROBLEM
• M O ’ D ATA M O ’ P ROBLEM
• T OPOLOGICAL D ATA A NALYSIS AND V ISUALIZATION (TDAV) • study of approaches to EXTRACT structure from NOISY or COMPLEX data and REPRESENT that data in an actionable form
• P ERSISTENT H OMOLOGY • a method for computing topological FEATURES of a space at DIFFERENT spatial RESOLUTIONS
• H OW D OES T HIS R ELATE TO R ADIO A STRONOMY ? • TDAV represents a DIVERSE toolbox capable of addressing analysis NEEDS in many contexts • Our development study addresses these needs specifically via the CONTOUR TREE
• T OPOLOGICAL S KELETON : C ONTOUR TREE
• C ONTOUR TREES
• C ONTOUR TREES
• C ONTOUR TREES
• C ONTOUR TREES
• C ONTOUR TREES
• C ONTOUR TREES
• C ONTOUR TREES
• C ONTOUR TREES
• A C LOSER L OOK AT THE C ONTOUR TREE Scalar Value of Event
• A C LOSER L OOK AT THE C ONTOUR TREE Scalar Value of Event Birth of the Feature
• A C LOSER L OOK AT THE C ONTOUR TREE Scalar Value of Event Death of the Feature Birth of the Feature
• A C LOSER L OOK AT THE C ONTOUR TREE Scalar Value of Event Persistence of the Feature
• F EATURE R EMOVAL
• F EATURE R EMOVAL
• F EATURE R EMOVAL
• S CALARFIELD S IMPLIFICATION
• S CALARFIELD S IMPLIFICATION
• S CALARFIELD S IMPLIFICATION
• S CALARFIELD S IMPLIFICATION
• S CALARFIELD S IMPLIFICATION
• R ESULTS • Simple Spinning Disk • from Anil Seth • Phys. & Astro. • University of Utah
• V ARYING P ERSISTENT S IMPLIFICATION
original simplified
original simplified
original simplified
original simplified
original simplified
• S TEPPING THROUGH S LICES
original simplified
original simplified
original simplified
original simplified
original simplified
original simplified
original simplified
original simplified
original simplified
original simplified
original simplified
M OMENT 0 A NALYSIS simplified original
• V OLUME R ENDERED
original simplified
original simplified
original simplified
• S UMMARY • Early results convincing • Open questions remain Scalar field simplification choice Scalability of software Related visualization needs Additional uses of the contour trees Scientific impact of simplification Other TDAV data structures
• S OFTWARE • Software will be publicly released before the end of the year • We invite interested users to contact us for early access
• Q UESTIONS ? • C ONTACT • Paul Rosen <prosen@usf.edu> • P ROJECT W EBSITE • http://alma-tda.cspaul.com
• A LTERNATIVE S IMPLIFICATIONS
• A LTERNATIVE S IMPLIFICATIONS
• A LTERNATIVE S IMPLIFICATIONS
• A LTERNATIVE S IMPLIFICATIONS
• P ROCESSING P IPELINE Feature Feature( Exploration Explora,on( Visualiza,on( Visualization eleva,on( elevation eleva,on( elevation azimuth( azimuth Feature Feature( azimuth( azimuth Feature(Extrac,on(Using( Data Transformation Feature Extraction Data(Transforma,on( Comparison Comparison( Contour=Trees( to(Scalar(Field(( to Scalar Field Using Contour Trees Visualization Visualiza,on(
• P ROCESSING P IPELINE Feature Feature( Exploration Explora,on( Visualiza,on( Visualization eleva,on( elevation eleva,on( elevation azimuth( azimuth Feature Feature( azimuth( azimuth Feature(Extrac,on(Using( Data Transformation Feature Extraction Data(Transforma,on( Comparison Comparison( Contour=Trees( to(Scalar(Field(( to Scalar Field Using Contour Trees Visualization Visualiza,on(
• P ROCESSING P IPELINE Feature Feature( Exploration Explora,on( Visualiza,on( Visualization eleva,on( elevation eleva,on( elevation azimuth( azimuth Feature Feature( azimuth( azimuth Feature(Extrac,on(Using( Data Transformation Feature Extraction Data(Transforma,on( Comparison Comparison( Contour=Trees( to(Scalar(Field(( to Scalar Field Using Contour Trees Visualization Visualiza,on(
• P ROCESSING P IPELINE Feature Feature( Exploration Explora,on( Visualiza,on( Visualization eleva,on( elevation eleva,on( elevation azimuth( azimuth Feature Feature( azimuth( azimuth Feature(Extrac,on(Using( Data Transformation Feature Extraction Data(Transforma,on( Comparison Comparison( Contour=Trees( to(Scalar(Field(( to Scalar Field Using Contour Trees Visualization Visualiza,on(
• R ESEARCH Q UESTIONS : D ATA T RANSFORMATION • How are the spectral lines represented in a 3D data cube MEANINGFULLY converted to scalar functions for contour tree-based analysis?
• R ESEARCH Q UESTIONS : F EATURE E XTRACTION • Once a contour tree is generated, there are many methods for selecting the important features of the tree. Therefore, how do we extract MEANINGFUL features of the data via contour tree simplification to suit the needs of astronomers?
• R ESEARCH Q UESTIONS : F EATURE E XPLORATION • What is a MEANINGFUL visualization of contour trees to enable feature exploration of a single data cube by the astronomers?
• R ESEARCH Q UESTIONS : F EATURE C OMPARISON • Can contour tree representations be used for MEANINGFUL feature comparisons among multiple data cubes to characterize secular changes with observed properties (for example, transition energy, molecular species and chemical families), or derived properties such as temperature and density?
• I NTERDISCIPLINARY R ESEARCH IS HARD You Me
Outstanding issue: multiple slices • How to co-simplify? • Multiple 2D vs 3D contour trees?
Outstanding issue: Local vs. global contour tree • Precomputation? • Data storage and query? • Efficient computation on parallel machine(s)?
Outstanding issue: Boundary Conditions
Outstanding issues: Boundary Conditions
Outstanding issues: Boundary Conditions • Are boundaries true critical points?
Recommend
More recommend