Causal structure learning algorithms have focused almost exclusively
on learning in “stable” environments in which the underlying
causal structure does not change. Such changes often occur,
however, without warning or signal in real-world environments. In
this paper, we present DOCL, a novel causal structure learning
algorithm that processes data in a dynamic, real-time manner and
tracks changes in the generating causal structure or
parameters. The algorithm learns in an online fashion from
sequential or ordered data rather than “batch-mode” from a full
dataset, and so supports causal learning in memory-limited
settings. We show by simulation that the algorithm performs
comparably to batch-mode learning when the causal structure is
stationary, and significantly better in non-stationary environments.
[ Paper ]