Labeling Complicated Objects: Multi-View Multi-Instance Multi-Label Learning

Authors: Cam-Tu Nguyen, Xiaoliang Wang, Jing Liu, Zhi-Hua Zhou

AAAI 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We perform experiments on 2 multi-view datasets and 3 single-view datasets.
Researcher Affiliation Academia Cam-Tu Nguyen1,3, Xiaoliang Wang1, Jing Liu2 and Zhi-Hua Zhou1 1 National Key Laboratory for Novel Software Technology, Nanjing University, 210023, China 2 National Key Laboratory of Pattern Recognition, Institute of Automation Chinese Academy of Sciences, China 3 VNU-University of Engineering and Technology, Hanoi, Vietnam
Pseudocode Yes Algorithm 1 Generative Process for MIMLmix, Algorithm 2 Training with MIMLmix, Algorithm 3 Testing with MIMLmix
Open Source Code No The paper does not contain any statement about releasing open-source code, nor does it provide a link to a code repository for the methodology described.
Open Datasets Yes Citeseerx-10k1 contains scientific papers in two views, i.e, content (v1) and citations (v2). Image CLEF (M uller et al. 2010) contains images with two views: visual (v1) and textual (v2). Here, we use the same subset that has been used in (Nguyen, Zhan, and Zhou 2013).1collected from http://citeseerx.ist.psu.edu/index ... Letter Carroll, MSRC-v2 were collected by (Briggs, Xiaoli, and Raich 2012); and IAPRTC-12 dataset was selected from (Escalante et al. 2010).
Dataset Splits Yes We conduct 30 times evaluation for Image CLEF, each time we use 1000 examples for training and 1000 examples for testing; 10-fold crossvalidation is conducted for the other datasets.
Hardware Specification Yes Table 4 shows training times of MIML methods on 3 large datasets on the same computer (CPU of 3.3Gz, 4GB memory).
Software Dependencies No The paper mentions using 'LIBSVM' and describes parameters like 'RBF kernel with C=23 and γ=.5' but does not specify exact version numbers for any software dependencies.
Experiment Setup Yes For MIMLmix methods, we set 0 = 0.1, K = 200 as default for both datasets, set = .3, = 5 for Citeseerx-10k; and = 10 and = 0 for Image CLEF. M3LDA is conducted with the same setting as in (Nguyen, Zhan, and Zhou 2013) on Image CLEF; and with K = 200, λ = .5, and the number of sampling iterations of 300 on Citeseerx-10k.