Title :
An efficient gradient computation approach to discriminative fusion optimization in semantic concept detection
Author :
Ma, Chengyuan ; Lee, Chin-Hui
Author_Institution :
Sch. of Electr. & Comput. Eng., Georgia Inst. of Technol., Atlanta, GA
Abstract :
In this paper, we propose an efficient gradient computation approach for discriminative fusion optimization in TRECVID high-level feature extraction. Numerical approximation was exploited in gradient calculation and model parameter update. The gradient of the performance measure was approximated by a sum of instance point-wise gradient instead of instance pair-wise gradient used in maximum figure-of-merit learning such that performance metrics like average precision can be optimized directly and efficiently on large training set. Experiments on the TRECVID 2005 high-level feature extraction test set showed that the proposed algorithm can improve the mean average precision from 0.254 of a state-of-the-art baseline system to 0.285.
Keywords :
approximation theory; feature extraction; gradient methods; learning (artificial intelligence); optimisation; video signal processing; TRECVID feature extraction; discriminative fusion optimization; figure-of-merit learning; gradient calculation; gradient computation; instance point-wise gradient; mean average precision; model parameter update; numerical approximation; performance metrics; semantic concept detection; state-of-the-art baseline system; Broadcasting; Detectors; Face detection; Feature extraction; Indexing; Measurement; Multimedia communication; Statistics; System testing; Training data;
Conference_Titel :
Pattern Recognition, 2008. ICPR 2008. 19th International Conference on
Conference_Location :
Tampa, FL
Print_ISBN :
978-1-4244-2174-9
Electronic_ISBN :
1051-4651
DOI :
10.1109/ICPR.2008.4761471