Optimizing Bag Features for Multiple-Instance Retrieval

Authors: Zhouyu Fu, Feifei Pan, Cheng Deng, Wei Liu

AAAI 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The experimental results demonstrate the effectiveness of the proposed approach in comparison with the alternatives for MI retrieval.
Researcher Affiliation Collaboration 1School of Computing, Engineering & Mathematics, University of Western Sydney, NSW, Australia 2New York Institute of Technology, New York, NY, USA 3School of Electronic Engineering, Xidian University, Xi an, China 4IBM T. J. Watson Research Center, Yorktown Heights, NY, USA z.fu@uws.edu.au, fpan@nyit.edu, chdeng@mail.xidian.edu.cn, weiliu@us.ibm.com
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any explicit statements about the release of source code or links to a code repository for the methodology described.
Open Datasets Yes We used four real-world benchmark data sets in our experiment, including two data sets on drug activity prediction (MUSK1 and MUSK2), two data sets on image categorization (COREL-10 and COREL-20).
Dataset Splits Yes For the COREL data sets, we randomly split each data set into two equal halves as the training and query sets. For the smaller MUSK data sets, we adopted 10-fold validation for each single training and query round.
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments.
Software Dependencies No The paper does not list specific software dependencies with version numbers.
Experiment Setup Yes For bag Feat, the γ parameter in eq. 6 was empirically set to the inverse of the instance feature dimension. Moreover, we tested different number of prototypes (32, 64 and 128) for all three bag Feat variants. For each method and data set, the experiment was repeated 10 times using different random data partitions. Query results for each method are ranked in increasing distances or decreasing similarities.