Weakly supervised learning

As explained in Machine Learning System Design

Weakly supervised learning is a machine-learning approach that learns from labels that are incomplete, noisy, imprecise, or otherwise weaker than fully accurate annotations.

Weakly supervised learning uses inaccurate or partial labels when fully detailed annotation is too slow, expensive, expertise-intensive, or unsupported by available tools. The labeling choice affects model options and ultimately constrains the product. section 6.2.4

Consider weak supervision when:

• Properly labeling each item would take too much time. section 6.2.4
• Too few qualified experts are available to cover enough data. section 6.2.4
• The required labeling toolset or level of detail is not ready. section 6.2.4

Medical-image labeling

A labeling task may ask whether image pixels represent malignancy. If pixel-level annotation is impractical, incomplete or inaccurate labels become a possible alternative, subject to product goals and needs. section 6.2.4

Figure 6.2 Various approaches to data labeling

Compare labeling approaches before choosing the annotation detail, tools, expertise, and model implications. section 6.2.4

How Machine Learning System Design covers it

How deep it goes

Defines the motivation and tradeoffs but stops short of algorithms, implementation, or evaluation procedures.

Examples

Uses medical-image pixel labeling and constraints involving annotation time, expert availability, and tool readiness.

What it assumes

Assumes defined product goals and an understanding of the required labeling detail, expertise, tools, and model implications.

What you can do after

Evaluate when partial or inaccurate labels are a practical alternative to fully detailed annotation.

How it connects

Connects weak supervision to data labeling, labeling tools, model selection, product goals, and crowdsourcing.

liveBookmanning.com

Read Machine Learning System Design on liveBook Get the book at manning.com

Related concepts

• labeluses · A label is a value or annotation associated with an item that identifies its target category, outcome, or reference answer for learning or evaluation.

This explainer was generated by Manning's knowledge graph from the text of Machine Learning System Design. The section links point to the parts of the book it draws on.