high-performance computing for visual analytics · 2017-01-31 · high-performance computing for...
TRANSCRIPT
FachgebietComputergrafischeSystemeProf.Dr.JürgenDöllnerMasterprojektSommersemester2017
High-PerformanceComputingforVisualAnalyticsBackgroundVisualanalyticsisoneofthekeytechnologicaldirectionsthatshapehowwecanhandle,manage,andusebigdataacrossdifferentapplicationdomains.Efficientalgorithmsanddatastructuresareessentialforanoperationalsystemprocessingbigdatavolumes.Thisisthemasterproject'soveralltopic.
DescriptionThisprojectaimsatdesigningand implementingselectedvisualanalyticshigh-performance techniques.Theyaretargetedatnext-generationtoolsandapplicationsforvisualanalytics.Thesubtopicsinclude.
• Efficient Processing of Sensor Data: How toaggregate, and summarizemassive sensor databased on measurement time or measurementvalues while, at the same time, preserving thedataset’s key characteristics relevant for visualanalytics?Howtovisualizemassive sensordataand its dynamics to assist a user in identifyingsensor data patterns (measurements anddynamics) typically preceding failure events?How to derive the criticality of a sensor-observedprocessandhowtomapthiscriticalityontovisual variablesof a “sensordatamap”or“3D plant map”. This subtopic is bound to anongoingjointresearchproject.
• Efficient Deep Learning on Point Clouds:
Geospatial point clouds represent capturedreality by means of unstructured, massive 3Dpoint collections. Identifying objects andclassifying regions can both be efficiently andeffectively performed by deep learning systemsas thenatureofpointcloudsexactly fits to theircorestrengths.Thissubtopicisboundtoongoingjointresearchprojects.
ThemasterprojectreferstoanumberofcurrentresearchandsoftwareprojectsoftheHPI'sComputerGraphicsSystemsgroup.It isespeciallysuitedforfurtherresearchinthecontextofamasterthesisorafuturedoctoralthesis.Further,themasterprojectcanlayafoundationforworkingasastudentassistantorsoftwaredeveloperatourresearchpartners.
ContactResearchGroupComputerGraphicsSystem,Prof.Dr.JürgenDöllner([email protected]),BenjaminHagedorn&JanKlimke(SensorData),andRicoRichter&JoannesWolf(PointClouds).