Page 32 - CyberSecurityEssentialsEnglish

Category Typical output Workplace example How to test success

Perception
Transcribe audio or Error rate and performance
(image and Extracted text or
read a scanned with noise and different
audio objects
document accents
understanding)

Draft text or create an Relevance, consistency,
Generation New content
image or code and ease of review

Achievement of the
Optimization and Schedule resources
Plan or allocation objective while respecting
planning under constraints
constraints and exceptions

Decision safety,
Suggested action or Recommend an action
Decision support appropriate permissions,
recommendation or operate a tool
and reversibility

? Question to consider: Before choosing a tool, does your problem truly require
content generation, or does it require information retrieval, calculation, classification,

ranking, or the application of a constraint?

● Remember: Section summary
• Key idea: Begin with the output you need, because the type of output decides both the

tool and the way you test it.
• Practical decision: Name the task family before choosing a product: classification,
forecasting, ranking, perception, generation, optimization, or decision support.
• Mistake to avoid: Treating one application as a single tool when it is a chain of steps,
which makes an error impossible to locate.

1.4 How Do Systems Learn from Data?

◆ Key idea: A system does not learn in the same way a person does.

When we say that a model “learns,” we do not mean that it acquires experience or
understanding like a person. We mean that it adjusts internal numerical values to improve

performance on a specific task. The quality of its output depends on its training data, the
objective it was asked to achieve, and the way it was tested.

Machine learning begins with data that represents the problem, such as text, images,
numbers, or labeled examples. The model searches for recurring relationships and uses what
it has learned on new cases. It does not learn from “the entire world,” but from the data

made available to it and the way that data was prepared.

31