Short answers, pulled from the story.
Statisticians introduced discriminant analysis in 1936 to assign classes based on data patterns. This early method laid the groundwork for modern pattern recognition systems.
Supervised learning relies on training sets where humans manually label every instance with the correct output. Unsupervised learning operates without any pre-existing labels or human guidance and searches for inherent similarities within raw data itself.
Banks adopted stylus-based signature verification starting in 1990 to confirm customer identity. This application falls under the broader category of computer-aided diagnosis and identification tasks.
Probabilistic classifiers output not just a single best label but also a probability score for that choice. This confidence value allows systems to abstain from making decisions when certainty falls below a threshold.
The Branch-and-Bound algorithm reduces complexity but remains intractable for medium to large feature sets. Feature extraction transforms raw vectors into lower dimensions using techniques like principal components analysis instead.