AdaNPC: Exploring Non-Parametric Classifier for Test-Time Adaptation
Authors: Yifan Zhang, Xue Wang, Kexin Jin, Kun Yuan, Zhang Zhang, Liang Wang, Rong Jin, Tieniu Tan
ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We test our model on extensive numerical experiments. Ada NPC significantly outperforms competitive baselines on various DG benchmarks. |
| Researcher Affiliation | Collaboration | 1School of Artificial Intelligence, University of Chinese Academy of Sciences (UCAS) 2MAIS, CRIPAC, CASIA. 3Work done during an internship at Alibaba Group. 4Machine Intelligence Technology, Alibaba Group. 5 Department of Mathematics at Princeton University. 6Center for Machine Learning Research, Peking University. 7AI for Science Institute, Beijing. 8Work done at Alibaba Group, and now affiliated with Meta. 9Nanjing University. |
| Pseudocode | Yes | Algorithm 1 Online optimization algorithm for optimizing the KNN loss function LKNN. |
| Open Source Code | Yes | Code is available at https://github.com/yfzhang114/Ada NPC. |
| Open Datasets | Yes | We use five popular OOD generalization benchmark datasets: Rotated MNIST (Ghifary et al., 2015), PACS (Li et al., 2017), VLCS (Torralba & Efros, 2011), Terra Incognita (Beery et al., 2018) and Domain Net (Peng et al., 2019). |
| Dataset Splits | Yes | Specifically, we split the data from each domain into 80% and 20% proportions, where the larger split is used for training and evaluation, and the smaller ones are used for select hyperparameters. We repeat the entire experiment twice using three different seeds to reduce the randomness. |
| Hardware Specification | Yes | We run our experiments mainly on Tesla-V100 (32G)x4 instances. |
| Software Dependencies | No | The paper mentions software like "Py Torch or Tensor Flow", "Faiss", "torch-vision", "timm", and "Domainbed codebase", but it does not specify exact version numbers for these software components or libraries, which is necessary for reproducible software description. |
| Experiment Setup | No | The paper mentions conducting a "random search of 20 trials over the hyperparameter distribution" and the use of a "KNN loss" and "BN retraining". It also mentions "batch size" in general terms. However, it does not explicitly provide the specific numerical values for key hyperparameters (e.g., learning rate, specific batch size values used for training, number of epochs, optimizer details) for the main experiments in a reproducible format within the main text. |