Provable Discriminative Hyperspherical Embedding for Out-of-Distribution Detection
Authors: Zhipeng Zou, Sheng Wan, Guangyu Li, Bo Han, Tongliang Liu, Lin Zhao, Chen Gong
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results demonstrate the effectiveness of our proposed DHE, which showcases a remarkable reduction in FPR95 (i.e., 5.37% on CIFAR-100) and more than doubling the computational efficiency when compared with the state-of-the-art methods. |
| Researcher Affiliation | Academia | 1School of Computer Science and Engineering, Nanjing University of Science and Technology, China 4Hong Kong Baptist University, China 5Sydney AI Centre, The University of Sydney, Sydney 6Department of Automation, Institute of Image Processing and Pattern Recognition, Shanghai Jiao Tong University, China EMAIL, EMAIL |
| Pseudocode | No | The paper describes the proposed method, DHE, with theoretical analyses, mathematical formulations, and a framework overview (Figure 1), but does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | Code https://github.com/Canoeszzp/DHE. |
| Open Datasets | Yes | We use the CIFAR10 (Krizhevsky, Hinton et al. 2009) and CIFAR100 (Krizhevsky, Hinton et al. 2009) as our ID datasets, which have been commonly adopted in this field. For evaluation of OOD detection, we use five commonly-used datasets, including SVHN (Netzer et al. 2011), Places365 (Zhou et al. 2017), Texture (Cimpoi et al. 2014), LSUN (Yu et al. 2015), and i SUN (Xu et al. 2015). |
| Dataset Splits | Yes | We use the CIFAR10 (Krizhevsky, Hinton et al. 2009) and CIFAR100 (Krizhevsky, Hinton et al. 2009) as our ID datasets, which have been commonly adopted in this field. For evaluation of OOD detection, we use five commonly-used datasets, including SVHN (Netzer et al. 2011), Places365 (Zhou et al. 2017), Texture (Cimpoi et al. 2014), LSUN (Yu et al. 2015), and i SUN (Xu et al. 2015). |
| Hardware Specification | No | The paper mentions using ResNet-18 and ResNet-34 models and discusses training time, but does not provide specific details about the hardware (e.g., GPU/CPU models, memory) used for the experiments. |
| Software Dependencies | No | The paper does not provide specific software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions, or CUDA versions) that were used to replicate the experiments. |
| Experiment Setup | Yes | Specifically, we set the initial learning rate to 0.5 with cosine scheduling, maintain a batch size of 512, and conduct training for a duration of 500 epochs. The embedding dimension D is set to 128 for our projector... The temperature t in our method is set to 0.1. The model is trained by using stochastic gradient descent (SGD) with the momentum and the weight decay setting to 0.9 and 10 4, respectively. |