Cascaded Hierarchical Recognition¶
Recognize complex cases by moving attention through a hierarchy of coarse filters and fine discriminators instead of trying to inspect every possible feature at once.
The Diagnostic Story¶
Symptom: Every case arrives and the full battery of checks begins at once: fine-grained detail is inspected before broad class has even been established. Attention and computation are consumed on candidates that simple early cues could have ruled out. Some common cases are caught, but rare ones slip through because no stage owns ambiguity, and no one can explain afterward where the recognition failed.
Pivot: Arrange features into a hierarchy — cheap, broad, high-recall cues first; costly, fine, high-precision discriminators later — and let each stage narrow the candidate set before the next stage runs. Preserve uncertainty across handoffs so ambiguous or high-risk cases can escalate rather than fall silently.
Resolution: Cognitive and computational cost drops because effort concentrates only where candidates still need discrimination. Recognition becomes explainable because each stage has a defined question, threshold, and handoff. False positives from superficial resemblance and false negatives from premature closure both decrease.
Reach for this when you hear…¶
[emergency triage] “We check airway before we check anything else — you don't start listening to lung sounds on someone who isn't breathing.”
[document processing] “Route by document type before you try to extract fields — running NLP on a fax that turned out to be a blank cover sheet wastes everyone's time.”
[wildlife survey] “Confirm it's a raptor by silhouette before you zoom in for species — otherwise you're counting every hawk-shaped cloud.”
Mechanisms / Implementations¶
- Cascade Error Audit
- Coarse Screening Checklist
- Confidence Threshold Table: A maintained lookup table that turns model confidence and residual size into an action — pass, review, or escalate — indexed by stage and risk level.
- Drilldown Recognition Interface
- Expert Review Checkpoint
- Feature Pyramid or Hierarchical Model
- Multi-Stage Classifier Pipeline
- Salience Map or Attention Heatmap
- Stage Transition Log
- Triage Queue with Escalation Rules
Related Abstractions¶
Abstractions this archetype builds on — directly (a source ingredient) or as a related pattern. Links follow the typed catalog namespace.
Built directly on (3)
- Attention: The selective allocation of a fixed processing capacity to some inputs while the rest are filtered out, surfacing scarcity upstream of every decision.
- Hierarchy: Organizes elements into levels or ranks.
- Pattern Recognition: Identify regularities.
Also references 19 related abstractions
- Abstraction: Focus on core elements.
- Bounded Rationality: Limited decision capacity.
- Classification: Sorting entities into discrete categories by explicit rules, turning unbounded variation into a finite, reusable map for downstream reasoning and action.
- Cognitive Load: Mental effort.
- Cognitive Load And Attentional Capacity
- Compression: Reduce redundancy.
- Contrast: Emphasized difference.
- Emphasis: Highlighting priority element.
- Feedback: Outputs influence inputs.
- Layering: Segments systems into levels.
Variants¶
Narrower or domain-specific specializations that share this archetype's core structure. Recognized variants are established; candidate variants are provisional.
Visual Object Recognition Cascade · domain variant · recognized
Recognize visual objects by moving from broad spatial or silhouette cues to increasingly fine visual features and identity checks.
Diagnostic Triage Recognition Cascade · domain variant · recognized
Route clinical, operational, or incident cases from broad triage cues to targeted diagnostic tests while escalating ambiguity and red flags.
Attention and Feature Extraction Scaffolding · implementation variant · candidate
Teach or guide users to notice the right feature level in the right order before attempting final recognition.
Hierarchical Machine Classifier Pipeline · implementation variant · recognized
Implement the cascade through models or rules arranged so each classifier handles a narrower candidate set than the previous one.