Computational experiments in Science: Horse wrangling in the digital age

Abstract : The ready availability of massive amounts of data in numerous scientific fields, while an obvious boon for research, may also occasionally have a pernicious effect, as it lulls scientists into a false sense of confidence in their experimental results. Indeed, while all medical double-blind studies come with the caveat of limited sample size, and no result is considered acquired until it has been consistently duplicated by several teams, computer data analysis studies routinely boast decimal-point precision percentages as proof of the validity of their approach, considering that the size of their experimental dataset guarantees its representativity. Cue subsequent announcements of superior decimal-point precision percentages, in a process we call Progress. This performance-driven approach to research is probably unavoidable, and it gives evaluation data a crucial importance. A high level of scrutiny is therefore necessary, both of the data themselves and of the way they are used.
Document type :
Conference papers
Complete list of metadatas

https://hal.archives-ouvertes.fr/hal-01635373
Contributor : Mathieu Lagrange <>
Submitted on : Thursday, November 16, 2017 - 9:49:35 AM
Last modification on : Friday, May 17, 2019 - 9:22:06 AM
Long-term archiving on : Saturday, February 17, 2018 - 12:51:24 PM

Files

lagrangeHorse16.pdf
Files produced by the author(s)

Identifiers

Collections

Citation

Mathieu Lagrange, Mathias Rossignol. Computational experiments in Science: Horse wrangling in the digital age. Research workshop on “Horses” in Applied Machine Learning, Oct 2016, London, United Kingdom. ⟨10.1109/TMM.2014.2330697⟩. ⟨hal-01635373⟩

Share

Metrics

Record views

97

Files downloads

20