KREAM@ICCS2013
This document proposes a new approach to conducting scientific research that continuously updates hypotheses as new data becomes available over time. It involves ingesting relevant datasets, representing hypotheses as executable programs, and periodically re-running the programs on the aggregated data to test if conclusions have changed. As a case study, the document applies this approach to re-evaluate a published hypothesis about long-term climate trends in Australia over a wider timescale and geographical scope using continuously updated temperature data sources. It concludes the proposed approach leverages existing tools to support science through large-scale computation and data integration but challenges remain in fully representing the scientific inquiry process and its computational demands.
KREAM@ICCS2013
- 1. Jaakko Lappalainen Computer Science department University of Alcalá, Spain
- 2. Overview • • • • • • • • The problem Proposed approach The method Results Conclusions Strengths and weaknesses Future work Questions
- 3. The problem • Researchers focus on a particular time frame and scope for testing their hypotheses. • But the conclusions of the research are projected to the future. • Paradox: the work that predicts things for tomorrow, becomes a snapshot of what happened until today.
- 4. Proposed approach • New data relevant to some hypotheses gets continuously aggregated as time passes. • With common semantics, it can be combined or related to other datasets. • Represent the hypothesis as programs that are executed repeatedly.
- 5. The method • The case of study – Lenten, L. J., & Moosa, I. A. (2003). An empirical investigation into long-term climate change in Australia. Environmental Modelling & Software, 18(1), 59-70. • The authors claim that the temperature series has some a trend feature.
- 6. The method (II) • Let’s find some data sources. – ACORN-SAT, from the Australian Bureau of Meteorology. This uses LD!! – NOAA weather data, not in LD but easy to parse… • Periodically ingest data (e.g., into a relational database) • An R script checks if the trend on the data has changed… • Ingested data is semantically tagged…
- 7. Results • We are checking for Lenten & Moosa’s hypothesis every week. – More extensive time scope. – Wider geographical scope, to all data available for Australia. • The snapshot becomes a movie. • Executable paper
- 8. Conclusions • The tools we already have allows us to use large-scale computation infrastructures easily to support science. – The agINFRA project • Massive data ingestion. • Data integration and interlinking. • User-tailored service execution.
- 9. Strengths • Data availability – The data is ingested (from LD sources, but not only) and published. • Data interoperability – The data is not stored by itself. • Actionable data – Ready to be addressed, used and generate new actionable data.
- 10. Weaknesses • Represent ‘science inquiry’ as a data model is not trivial. • CPU-consuming tasks are even more consuming.
- 11. Future work • Further dataset interlinking – More plural value for physical parameters. – Dataset value error detection. • Advance in hypothesis representation – Machine readable research processes.
- 12. Questions?
- 13. Thank you very much! jkk.lapp@uah.es