Page 35 - CyberSecurityEssentialsEnglish

FIGURE 1-4: FROM DATA TO TRAINING, INFERENCE, AND UPDATING

What to notice: A model learns from data during training and produces outputs for new
cases during use.

We test it before adoption and monitor it afterward because the environment, audience,
and workflow may change.
Source: Prepared by the author based on references [4–6].

TABLE 1-5: MAIN STAGES FROM TRAINING TO USE

Stage Main question Example

What patterns should the Teach a model to distinguish
Training
system learn? request types.

Does the model improve on
Development testing Compare more than one
examples it did not
(validation) configuration or version.
memorize?

Does it succeed on new
Final evaluation Test it on examples excluded from
cases that represent
(testing) training.
reality?

How does it produce an Submit a new request and receive
Use (inference)
output for a new case? a category or content.

Monitoring and Is performance still suitable Review performance when the

updating after the data changes? audience or workflow changes.

34