40 likes | 47 Views
DataForge: SourceForge for Datasets Preserving and Sharing Experimental Data. Prabal Dutta. CACM – September 2004, Vol. 47, No.9. Too often data “graduates” with a student (job security?) Good experimental design and data collection Is time-consuming Days, weeks, months Overhead-laden
E N D
DataForge: SourceForge for Datasets Preserving and Sharing Experimental Data Prabal Dutta
Too often data “graduates” with a student (job security?) Good experimental design and data collection Is time-consuming Days, weeks, months Overhead-laden 80% in some experiments Requires infrastructure and physical access Creates a barrier to entry End-to-end data providence Needed for traceability from source to interpretation Includes experimental design, setup, collection, scrubbing, selection, fusion, analysis, and conclusions Open science is good science Why bother preserving or sharing data?
Where do we go from here? • Key Questions: How to • Establish data ownership, precedence, and providence? • Motivate academic/industrial researchers to share data? • Recognize and reward the contributions of experimentalists? • Mitigate and manage conflicts of interests? • Disseminate knowledge of datasets? • Archive data in truly reusable form? • Potential Impacts • Dramatically improve research ROI through reuse • Lower barriers to entry for emerging groups • Capture and retain the products of scientific inquiry • The CENTS opportunity • Leadership role in driving adoption of “sensor data schema” • Distribute (BSD-style) open-source experimental datasets • Promote a “sensor data-abstract services”