Category-Aware Active Domain Adaptation

Authors: Wenxiao Xiao, Jiuxiang Gu, Hongfu Liu

ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Comprehensive experiments and in-depth explorations demonstrate the efficacy of our method on category-aware active DA over three datasets.
Researcher Affiliation Collaboration 1Computer Science Department, Brandeis University, Waltham, Massachusetts, USA 2Adobe Research, USA. Correspondence to: Wenxiao Xiao <wenxiaoxiao@brandeis.edu>.
Pseudocode Yes Algorithm 1 Category-Aware Active DA for Category c
Open Source Code Yes Our code is available at https://github.com/wxxiaoss/Category Aware DA.
Open Datasets Yes We choose three popular DA benchmark datasets, Office-Home (Venkateswara et al., 2017), Domain Net126 (Peng et al., 2019) and Vis Da-2017 (Peng et al., 2017) in our experiments.
Dataset Splits Yes For the validation loss Lv = ℓ(V ; ˆθ), V being the validation set, the change after removing a training sample can further be estimated using ˆθLv H 1 ˆθ ˆθℓ(xi, yi), as demonstrated in the work by Koh & Liang (2017). ... we use all queried target samples from previous iterations, i.e., V = Lt, to calculate the validation loss Lv in Eq. (2).
Hardware Specification Yes We implement our model using Py Torch(Paszke et al., 2019) and scikit-learn (Buitinck et al., 2013) with one NVIDIA TITAN RTX GPU.
Software Dependencies No The paper mentions 'Py Torch(Paszke et al., 2019) and scikit-learn (Buitinck et al., 2013)' but does not provide specific version numbers for these software components.
Experiment Setup Yes For all three datasets, we train the model for 5 iterations. For Office-Home (Venkateswara et al., 2017), we select 1% of the target data in each iteration. For the large datasets Vis Da-2017 (Peng et al., 2017) and Domain Net-126 (Peng et al., 2019), we select a fixed number of 100 samples in each iteration. ... We use the Res Net-50 (He et al., 2016) pre-trained on Image Net as the backbone feature extractor and train a DANN model as the base DA method following CLUE (Prabhu et al., 2021). ... In the following iterations, we actively query and annotate b target samples with Algorithm 1.