Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Deja Vu: Continual Model Generalization for Unseen Domains
Authors: Chenxi Liu, Lixu Wang, Lingjuan Lyu, Chen Sun, Xiao Wang, Qi Zhu
ICLR 2023 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on Digits, PACS, and Domain Net demonstrate that Ra TP significantly outperforms state-of-the-art works from Continual DA, Source-Free DA, Test-Time/Online DA, Single DG, Multiple DG and Unified DA&DG in TDG, and achieves comparable TDA and FA capabilities. |
| Researcher Affiliation | Collaboration | Chenxi Liu1 , Lixu Wang1 , Lingjuan Lyu2 , Chen Sun2, Xiao Wang1, Qi Zhu1 1Northwestern University, 2Sony AI EMAIL, EMAIL, EMAIL |
| Pseudocode | No | The paper describes algorithms and modules in text and diagrams (Figure 1) but does not include any explicit pseudocode blocks or algorithm listings. |
| Open Source Code | Yes | Our code is available at https://github.com/Sony AI/Ra TP. |
| Open Datasets | Yes | Digits consists of 5 different domains with 10 classes, including MNIST (MT), SVHN (SN), MNIST-M (MM), SYN-D (SD) and USPS (US). PACS contains 4 domains with 7 classes, including Photo (P), Art painting (A), Cartoon (C), and Sketch (S). Domain Net is the most challenging cross-domain dataset, including Quickdraw (Qu), Clipart (Cl), Painting (Pa), Infograph (In), Sketch (Sk) and Real (Re). Considering label noise and class imbalance, we follow Xie et al. (2022) to split a subset of Domain Net for our experiments. |
| Dataset Splits | No | In the source domain, we randomly split 80% data as the training set and the rest 20% as the testing set. In target domains, all data are used for training and testing. |
| Hardware Specification | No | The paper does not explicitly specify the hardware used for experiments, such as particular GPU or CPU models, or memory specifications. |
| Software Dependencies | No | The paper does not provide specific version numbers for software dependencies or libraries used in the experiments. |
| Experiment Setup | Yes | The SGD optimizer with an initial learning rate of 0.01 is used for Digits, and 0.005 for PACS and Domain Net. The exemplar memory size is set as 200 for all datasets, and the batch size is 64. For all experiments, we conduct multiple runs with three seeds (2022, 2023, 2024), and report the average performance. We use 800 steps for Digits, 50 steps for PACS and 70 steps for Domain Net. Training epoch is 30 for all datasets and domains. ... we use momentum of 0.9 and weight decay of 0.0005 to schedule the SGD. |