Non-parametric Outlier Synthesis
Authors: Leitian Tao, Xuefeng Du, Jerry Zhu, Yixuan Li
ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments show that NPOS can achieve superior OOD detection performance, outperforming the competitive rivals by a significant margin. [...] In this section, we present empirical evidence to validate the effectiveness of our method on real-world classification tasks. We describe the setup in Section 4.1, followed by the results and comprehensive analysis in Section 4.2 Section 4.5. |
| Researcher Affiliation | Academia | Leitian Tao School of Information Engineering Wuhan University taoleitian@gmail.com Xuefeng Du, Xiaojin Zhu, Yixuan Li Department of Computer Sciences University of Wisconsin Madison {xfdu,jerryzhu,sharonli}@cs.wisc.edu |
| Pseudocode | Yes | Our algorithm is summarized in Algorithm 1 (Appendix C). [...] Algorithm 1 NPOS: Non-parametric Outlier Synthesis |
| Open Source Code | Yes | Code is publicly available at https://github.com/deeplearning-wisc/npos. |
| Open Datasets | Yes | We use both standard CIFAR-100 benchmark (Krizhevsky et al., 2009) and the largescale Image Net dataset (Deng et al., 2009) as the in-distribution data. [...] For OOD datasets, we adopt the same ones as in (Huang & Li, 2021), including subsets of i Naturalist (Van Horn et al., 2018), SUN (Xiao et al., 2010), PLACES (Zhou et al., 2017), and TEXTURE (Cimpoi et al., 2014). |
| Dataset Splits | No | No explicit mention of validation dataset splits (e.g., percentages or sample counts for a validation set) is provided for reproducibility. The paper discusses training and testing. |
| Hardware Specification | Yes | We use Python 3.8.5 and Py Torch 1.11.0, and 8 NVIDIA Ge Force RTX 2080Ti GPUs. |
| Software Dependencies | Yes | We use Python 3.8.5 and Py Torch 1.11.0, and 8 NVIDIA Ge Force RTX 2080Ti GPUs. |
| Experiment Setup | Yes | Experimental details. We employ a two-layer MLP with a Re LU nonlinearity for ϕ, with a hidden layer dimension of 16. We train the model using stochastic gradient descent with a momentum of 0.9, and weight decay of 10 4. For Image Net-100, we train the model for a total of 20 epochs, where we only use Equation 8 for representation learning for the first ten epochs. We train the model jointly with our outlier synthesis loss (Equation 6) in the last 10 epochs. We set the learning rate to be 0.1 for the Rclosed branch, and 0.01 for the MLP in the Ropen branch. For the Image Net-1k dataset, we train the model for 60 epochs, where the first 20 epochs are trained with Equation 8. |