Active Learning with Cross-Class Similarity Transfer

Authors: Yuchen Guo, Guiguang Ding, Yue Gao, Jungong Han

AAAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on three datasets demonstrate that the proposed approach outperforms significantly the state-of-the-art related approaches. The performance curves on three datasets are plotted in Figure 1. It can be observed that our approach significantly outperforms the baselines, which verifies its effectiveness.
Researcher Affiliation Academia Tsinghua National Laboratory for Information Science and Technology (TNList) School of Software, Tsinghua University, Beijing 100084, China Northumbria University, Newcastle, NE1 8ST, UK
Pseudocode Yes Algorithm 1 AL with Cross-class Similarity Transfer
Open Source Code No The paper does not provide any explicit statement about making the source code available or a link to a code repository.
Open Datasets Yes The first is CIFAR10 (Krizhevsky 2009) which has 10 object classes and each class has 6, 000 images. The second is Animals with Attributes (Aw A) (Lampert, Nickisch, and Harmeling 2014) which has 50 animal classes. The third is a Pascal-a Yahoo (a PY) (Farhadi et al. 2009) which has two subsets.
Dataset Splits No The paper mentions that parameters were chosen by "class-wise cross validation" but does not specify a distinct validation split (e.g., a percentage or fixed count) of the dataset separate from the training and test sets.
Hardware Specification No The paper does not provide specific details about the hardware used to run the experiments (e.g., GPU models, CPU types, or memory specifications).
Software Dependencies No The paper mentions using "the Caffe tool (Donahue et al. 2014) with the pretrained Alex Net (Krizhevsky, Sutskever, and Hinton 2012)", but it does not specify version numbers for these software components, which is required for reproducibility.
Experiment Setup Yes The parameter λ in Eq. (10) is set to 0.5 in all experiments. The values of τ and η in Eq. (13) are chosen by class-wise cross validation... These parameters are chosen from {0.1, 1, 10}. In line 10 of Algorithm 1, we select and transfer Q = |Sc| = 200, 100, 100 samples for each target domain class for three datasets. To construct sample similarity graph in Eq. (7), we randomly choose 500 samples from Ds. For the matrix Kus in Eq. (13), we randomly choose 1, 000 samples from Ds.