|
PATTERN RECOGNITION
|
SHAPER
is a pattern recognition engine that enables software applications to recognize noisy, ill-defined patterns with
great accuracy and speed. |
| As a benchmark the recognition of handwritten symbols
is a difficult classification task. With 30 distinct classes of symbols, 20 training examples
in each symbol class, where each symbol example contains 300 bytes of information, handwriting
recognition performance on a 300 Mhz PC is the following: |
|
i. The first training creating the primitives takes 1 hour. Subsequent training takes 5 minutes. |
|
ii. Recognition speed is 0.001 seconds per symbol. |
|
iii. Accuracy is greater than 98% on untrained examples and 100% on trained examples. |
|
iv. The engine uses 30k bytes of memory. |
|
The accuracy increases considerably as the number of examples in each symbol class increases.
When there are 200 examples in each class, the accuracy is greater than 99%. Unlike other
recognition algorithms, the training time increases linearly as a function of the size of the
training space. |
SHAPER is more accurate at distinguishing gene expression patterns
in microarray data than handwriting patterns. |
