Pattern Recognition Lecture 1: Introduction and Applications

In pattern recognition, what is the informal definition of recognizing patterns in data?

In the context of medical diagnostics, what are the classes and objects?

What is the application for recognizing acoustic signals?

What application involves recognizing all possible characters, such as a, b, c,..., z?

In the toy application of fish sorting, what is the classifier being used?

Using fish length as the discriminating feature resulted in a classification error of 34%.

The best length threshold (L) found for classifying fish was 12.

When using fish lightness as the discriminating feature, the classification error was 8%.

Combining both length and lightness features resulted in a classification error of 4%.

A complicated decision boundary that doesn't generalize well to new data is known as underfitting.

Pattern recognition involves assigning an object or an event to one of several pre-specified categories.

The toy application of fish sorting involves sorting salmon and camera on a conveyer belt.

In medical diagnostics application, the classes are tumors and cancer.

In pattern recognition, the classes for the application of recognizing all possible characters (a, b, c,..., z) are phonemes.

An underfitting decision boundary is known to generalize well to new data.

The classifier performed ideally on the new data, with 0% classification error.

Using fish length as the discriminating feature alone resulted in a classification error of 34%.

The best length threshold (L) found for classifying fish was 9.

Overfitting the data results in complicated boundaries that do not generalize well to new data.

Combining both length and lightness features resulted in a classification error of 8%.