NP-Match: When Neural Processes meet Semi-Supervised Learning
Authors: Jianfeng Wang, Thomas Lukasiewicz, Daniela Massiceti, Xiaolin Hu, Vladimir Pavlovic, Alexandros Neophytou
ICML 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conducted extensive experiments on four public datasets, and NP-Match outperforms state-of-the-art (SOTA) results or achieves competitive results on them, which shows the effectiveness of NP-Match and its potential for SSL. |
| Researcher Affiliation | Collaboration | 1Department of Computer Science, University of Oxford, UK. 2Institute of Logic and Computation, TU Wien, Austria. 3Microsoft Research, Cambridge, UK. 4Department of Computer Science and Technology, Tsinghua University, Beijing, China. 5Department of Computer Science, Rutgers University, New Jersey, USA. 6Microsoft, Applied Science Group, Reading, UK. |
| Pseudocode | No | The paper contains diagrams (e.g., Figure 1 for an overview of NP-Match), but no structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | The source code is available at: https://github.com/Jianf-Wang/NP-Match. |
| Open Datasets | Yes | We conducted our experiments on four widely used public SSL benchmarks, including CIFAR-10 (Krizhevsky et al., 2009), CIFAR-100 (Krizhevsky et al., 2009), STL10 (Coates et al., 2011), and Image Net (Deng et al., 2009). |
| Dataset Splits | Yes | We evaluated NP-match on these two datasets following the evaluation settings used in previous works (Sohn et al., 2020; Zhang et al., 2021; Li et al., 2021). |
| Hardware Specification | Yes | GeForce GTX 1080 Ti GPUs were used for the experiments on CIFAR-10, CIFAR-100, and STL-10, while Tesla V100 SXM2 GPUs were used for the experiments on Image Net. |
| Software Dependencies | No | The paper mentions using stochastic gradient descent (SGD) and a cosine decay schedule (Loshchilov & Hutter, 2016), but does not provide specific version numbers for software libraries or environments used for implementation. |
| Experiment Setup | Yes | The deep neural network configuration and training details are summarized in Table 5. As for the NP-Match related hyperparameters, we set the lengths of both memory banks (Q) to 2560. The coefficient (β) is set to 0.01, and we sample T = 10 latent vectors for each target point. The uncertainty threshold (τu) is set to 0.4 for CIFAR-10, CIFAR-100, and STL-10, and it is set to 1.2 for Image Net. NP-Match is trained by using stochastic gradient descent (SGD) with a momentum of 0.9. The initial learning rate is set to 0.03 for CIFAR-10, CIFAR-100, and STL-10, and it is set to 0.05 for Image Net. The learning rate is decayed with a cosine decay schedule (Loshchilov & Hutter, 2016), and NP-Match is trained for 220 iterations. |