Taxonomy-Structured Domain Adaptation

Authors: Tianyi Liu, Zihao Xu, Hao He, Guang-Yuan Hao, Guang-He Lee, Hao Wang

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Empirically, our method achieves state-of-the-art performance on both synthetic and real-world datasets with successful adaptation.
Researcher Affiliation Academia 1Rutgers University 2Massachusetts Institute of Technology 3The Chinese University of Hong Kong.
Pseudocode Yes We summarize the training procedure formally in Algorithm 1. ... The inference procedure of TSDA is formally presented in Algorithm 2.
Open Source Code Yes Code is available at https: //github.com/Wang-ML-Lab/TSDA.
Open Datasets Yes Image Net-Attribute-DT (Ouyang et al., 2015) builds on the animal images from Image Net... CUB-DT (He & Peng, 2019) contains 11,788 images of 200 bird categories.
Dataset Splits No The paper describes source and target domains for training and testing, but it does not specify explicit validation dataset splits (e.g., percentages, counts, or predefined sets) for model tuning or evaluation beyond the source-target distinction.
Hardware Specification Yes All experiments are run on NVDIA Ge Force RTX 2080 Ti GPUs.
Software Dependencies No The paper mentions using 'Py Torch' and 'Adam optimizer (Kingma & Ba, 2015)' but does not provide specific version numbers for these or other software dependencies.
Experiment Setup Yes We have λd, λt and λe as the weights that balance the discriminator loss, the taxonomist loss, and the predictor loss. ... λd and λe range from 0.1 to 1 and λt ranges from 0.1 to 10. ... We use Adam optimizer (Kingma & Ba, 2015) for all models with learning rates from 1 10 4 to 1 10 6. ... Table 4 shows the experiment results of various λd, λt combination on DT-14.