Zero → Max AI

EP03 · 8 min

Learning types + task types

Match problem statements to the correct learning paradigm and task category.

Simple definition
Different learning types fit different data situations and goals.
Precise definition
Learning paradigms are optimization setups defined by supervision signal availability and feedback structure.

Objective

By the end, you should map a business request to both a task type (classification, regression, clustering) and a learning type (supervised, unsupervised, self-supervised, RL).

Concept map

  • Supervised learning: labeled examples (spam/not spam).
  • Unsupervised learning: structure without labels (review topic clusters).
  • Self-supervised learning: labels generated from data structure itself.
  • Reinforcement learning: sequential decisions with reward feedback.

Task types:

  • Classification predicts categories.
  • Regression predicts continuous numbers.
  • Clustering groups similar items without predefined labels.

Worked example (online store)

  • Spam filter: supervised + classification.
  • Delivery ETA: supervised + regression.
  • Group customer behavior profiles: unsupervised + clustering.
  • Optimize discount strategy over repeated interactions: potentially RL.

Practical decision rule

Before choosing model families, ask two questions:

  1. Do we have trusted labels?
  2. Are decisions independent or sequential over time?

These two questions eliminate most wrong paths.

Quick check guidance

If someone says "predict" but there is no label, challenge the framing immediately. You may need data collection before model selection.

Three takeaways

  • Learning type depends on feedback signal.
  • Task type describes output shape.
  • Good framing is the highest-leverage engineering step.

Visual Stage

Interactive walkthrough

Visual walkthrough: learning paradigms

Tap each paradigm to inspect feedback signal.

Step Insight

Uses explicit labels. Example: spam vs not spam with historical labels.

Common traps
  • Forcing every task into supervised learning.
  • Treating clustering output as ground truth labels.
  • Using RL when a static supervised setup is enough.
Three takeaways
  • Start from available feedback signal, then choose learning type.
  • Task type and learning type are related but not identical.
  • Wrong framing causes expensive rework later.
Next lesson