Bias Specification Language

Context:
- …
Example(s):
- A graph can be specific as bias when a user provides a graph structure of a model.
- …
Counter-Example(s):
- Hypothesis Language.
See: Learning as Search, Hypothesis Space, Instance-based Learning, Inductive Logic Programming.

References

(Sammut & Webb, 2017) ⇒ (2017). "Bias Specification Language". In: (Sammut & Webb, 2017). DOI: 10.1007/978-1-4899-7687-1_440
- QUOTE: A bias specification language is a language in which a user can specify a Language Bias. The language bias of a learner is the set of hypotheses (or hypothesis descriptions) that this learner may return.
  In contrast to the hypothesis language, the bias specification language allows us to describe not single hypotheses but sets (languages) of hypotheses.
  (...) In learning approaches based on graphical models or artificial neural networks, whenever the user provides the graph structure of the model, he or she is specifying a bias. The “language” used to specify this bias, in this case, consists of graphs. Figure 1 shows examples of such graphs. Not every kind of bias can necessarily be expressed by some bias specification language; for instance, the bias defined by the Bayesian network structure in Fig. 1 cannot be expressed using a Markov network. Bayesian networks and Markov networks have a different expressiveness, when viewed as bias specification languages.