Computer tool with sparse representation
AssigneesInria Institut National De Recherche En Informatique Et En Automatique, Universite De Rennes 1
InventorsChristine Guillemot, Joaquin Zepeda, Ewa Kijak
Life of this patent
Estimating the term…
Abstract
Computer tool characterised in that it comprises: a working memory (206) to receive input data, an auxiliary memory (204) to store several approximation dictionaries, a reducer (202) arranged for the iterative calculation of the approximation data of the input data, where the successive iterations are of the “matching pursuit” type, a selector (200) arranged to choose an approximation dictionary in the auxiliary memory based on a selection rule based on at least some of the preceding iterations, in which the reducer (202) calls the selector (200) in each iteration, and uses the dictionary thus selected in the adaptive matching.
Claims7
A computer tool comprising:
- a working memory to receive input data,
- an auxiliary memory to store several approximation dictionaries,
- a reducer arranged to calculate iteratively approximation data on the input data, in which the successive iterations are of the matching pursuit type,
- a selector arranged to select an approximation dictionary from the several approximation dictionaries in the auxiliary memory as a function of a selection rule based on at least some of the preceding iterations,
- in which:
- the reducer calls the selector in each iteration, and uses the dictionary thus selected in the matching pursuit;
- the selection rule comprises the selection of the approximation dictionary as a function of the number of the current iteration and/or the approximation data calculated in the preceding iterations;
- the approximation dictionaries are interconnected by a tree structure, and the selection rule comprises the selection of the approximation dictionary at a level of the tree taken from the number of the current iteration.
Dependencies are inferred from the claim text.