Analogical Predictions

This paper deals with exchangeable analogical predictions,
and proposes a Bayesian model for such predictions.
The paper first discerns two kinds of analogical
predictions, based on similarity of individuals and of types
respectively. It then introduces a Bayesian framework that
employs hypotheses for making predictions. This framework
is used to describe predictions based on the similarity
of individuals, and further relates exchangeable predictions
with a specific partition of hypotheses on types. Exchangeable
predictions based on type similarity are determined by
prior probabilities over the partition, but the partition obstructs
the control over the similarity relations. Finally the
paper develops a model for exchangeable predictions
based on type similarity, which employs hypotheses on
similarity between individuals, thereby offering a better
control over the similarity relations.