liveBook · Manning

Data representativeness

As explained in Build AI Drug Discovery Pipelines

Representativeness is the degree to which selected samples reflect the distribution and diversity of the data domain.

Data representativeness is the degree to which selected samples cover the input space and reflect the underlying distribution and diversity of a data domain. In active learning for virtual screening, it complements uncertainty: uncertainty finds confusing molecules, while representativeness prevents selection from concentrating in only a few regions. section 9.2.5

Active-learning acquisition strategies can promote representativeness by:

  • Selecting molecules maximally different from already labeled examples using distances in feature space. section 9.2.5
  • Combining uncertainty with density estimation so selected samples cover the data distribution rather than only model-confusion regions. section 9.2.5

How Build AI Drug Discovery Pipelines covers it

How deep it goes
Defines input-space coverage and names diversity and density-weighted strategies without providing a representativeness implementation.
Examples
Uses molecules, labeled examples, feature-space distances, and virtual-screening pools as examples of representative sampling.
What it assumes
Assumes familiarity with labeled and unlabeled samples, model predictions, feature spaces, and active-learning acquisition.
What you can do after
Apply it via diversity or density-weighted acquisition when selecting molecules for successive active-learning iterations.
How it connects
Connects active learning, acquisition functions, uncertainty sampling, diversity sampling, density estimation, and virtual screening.

Also explained in

Compare how 3 books explain data representativeness →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Build AI Drug Discovery Pipelines. The section links point to the parts of the book it draws on.