| 摘要 |
<p>Data processing device for selecting data words MS from data words DR contained in a dictionary which are the closest, according to a correspondence criterion, to a data word DO to be processed.
<??>The device comprises:
- first means (20) for imposing a mesh structure on a multidimensional space enclosing the set of data words DR of the dictionary,
- second means (24) (25) for attributing to each mesh a single attributed sub-dictionary which contains data words SDR selected from the data words DR of the dictionary, the data words SDR of an arbitrary sub-dictionary being those which satisfy the said correspondence criterion with respect to all the possible words which the mesh attributed to this arbitrary sub-dictionary can contain,
- third means (20) (24) (27) for using the said sub-dictionaries by determining, for each arbitrary data word DO to be processed, the mesh to which it belongs, then by determining, from the SDR words of the sub-dictionary attributed to the said mesh, the data word(s) MS which best correspond(s) to the said arbitrary data word DO to be processed.
<??>The mesh structure can be produced by a neural network in a layer or tree structure. The device can be used for information compression or data classification.
<IMAGE>
</p> |
