Carbon Storage in Basalt.

The advantage of unsupervised learning

is that one does not require training examples

(usually human-curated), which can

be time consuming to generate and may be

subject to human bias or error. Instead, such

algorithms fi nd inherent structures or clusters

in the data, which can be interpreted as

meaningful (e.g., distinct behavioral states).

A potential disadvantage is that although

such learning algorithms do not have human

bias, they necessarily contain bias in their

underlying assumptions and parameters.

Vogelstein et al. carefully validated their

analysis post-hoc, fi nding several identifi -

able behaviors including turning, escaping,

and backing up.