Tensor TownISSUE #6 of 120

unsupervised learning · finding structure

NeuraVSThe Overfit Ogre

Neura saysUnsupervised learning finds structure in unlabeled data — clusters, patterns, compression.

With no labels, the model discovers structure on its own: clustering groups similar items, dimensionality reduction compresses data while keeping its shape, anomaly detection flags outliers. It's useful when labels are unavailable or you don't yet know what you're looking for — exploratory by nature. The trade-off: harder to evaluate since there's no "right answer".

Power-ups you unlock

No labels — find structure itself
Clustering, dimensionality reduction, anomaly detection
Good when labels are unavailable
Harder to evaluate (no ground truth)

The Overfit Ogre attacks — common mistakes

Expecting clean "answers" from clustering
Over-interpreting discovered clusters
Skipping validation because there are no labels

Boss battleName a task where unsupervised learning fits better than supervised.

Example code

<!doctype html><html><head><meta charset="utf-8"></head>
<body style="background:#06040d;color:#e6e0ff;font-family:monospace;padding:20px"><pre>no labels → find structure
  cluster customers by behavior
  detect anomalies (fraud)
  compress features (PCA)</pre></body></html>

▶ Open the interactive comic issue

‹ Supervised Learning · Labeled Data Reinforcement Learning · Reward Signals ›