Acquisition

What it is

Every piece of research begins with something to work on. Acquisition is the act of obtaining that raw material. Sometimes it means making a measurement: pointing a telescope at the sky, recording a signal from a detector, running a chemical reaction and recording what comes out. Sometimes it means retrieval instead: downloading files from a public repository, copying records out of a database, or gathering published papers. The word covers both, because what matters is not the method but the result, which is data that did not exist in the researcher's hands before and now does.

Raw data is rarely tidy. It arrives in whatever form the instrument or archive produces: images, spectra, time series, text, long tables of numbers. Acquisition stops at that point. Deciding what to keep, correcting for faults, converting units and arranging things into a usable table all come later, and are usually treated as separate steps. Keeping acquisition distinct matters because it is the step that fixes what is possible: nothing can be analysed that was never obtained, and a flaw introduced here is carried through everything that follows.

Why scientists use it

Scientists separate out acquisition because it is where the questions of coverage and bias are settled. If a telescope only looks at part of the sky, or an archive only holds molecules that someone thought worth depositing, every later conclusion inherits that narrowness. Writing the step down makes those choices visible and lets other people judge or repeat them.

The step has grown more important as machine learning has spread through the sciences. These methods learn patterns from many examples rather than following rules written by hand, so they need large amounts of data, and their quality tracks the quality of what they were given. That has pushed researchers towards instruments and robots that acquire data continuously, and towards mining archives and published literature for material that was gathered for other purposes. Acquisition in that setting stands in for the older, slower business of a person running one experiment at a time.

In the record

Limits

Acquisition says nothing about whether the data is any good. An instrument can be miscalibrated, an archive can be skewed towards whatever was easy or fashionable to collect, and a download can be large while still missing the cases that matter. Volume is especially misleading: a great many measurements of nearly the same thing carry little more information than a few. Nor does obtaining data make it usable. Raw output often needs heavy correction before it means anything, and some of it will simply be discarded.

Written by AIxSci: the background from general knowledge, the examples from checked records.

Every article