Style Adaptation and Uncertainty Estimation for Multi-Source Blended-Target Domain Adaptation

Authors: Yuwu Lu, Haoyu Huang, Xue Hu

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

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments conducted on several challenging DA benchmarks, including the Image CLEF-DA, Office-Home, Vis DA 2017, and Domain Net datasets, demonstrate the superiority of our method over the state-of-the-art (SOTA) approaches.
Researcher Affiliation Academia Yuwu Lu , Haoyu Huang, and Xue Hu School of Artificial Intelligence, South China Normal University {luyuwu2008, hyhuang99, hx1430940232}@163.com
Pseudocode Yes Algorithm 1 SAUE for MBDA
Open Source Code Yes The source code of SAUE is provide in the Supplementary Material.
Open Datasets Yes Four standard benchmark datasets are used to validate the effectiveness of our proposed method. The Image CLEF-DA [40]... The Office-Home [41]... The Domain Net [14]... The Vis DA 2017 [42] dataset...
Dataset Splits No The paper mentions training data and test data implicitly through its use of benchmarks, but it does not specify a distinct validation set split or how it was used beyond 'training process'.
Hardware Specification Yes All experiments are run on a single Ge Force RTX-4090 GPU, and the batch size of both the source and blended-target domains are set to 32.
Software Dependencies Yes We utilize Py Torch framework [43] to perform our experiments; the Py Torch version is 1.13.1 and CUDA version is 11.7.
Experiment Setup Yes The optimizer is Stochastic Gradient Descent (SGD) with a momentum parameter of 0.9 and a weight decay of 1e-3. The learning rate is 1e-3 and updated by the Lambda LR [43] during the training process. All experiments are run on a single Ge Force RTX-4090 GPU, and the batch size of both the source and blended-target domains are set to 32. The hyper-parameters λe and λd, maximum iteration I, and mini-batch size B are also mentioned in Algorithm 1.