Few-Shot Non-Parametric Learning with Deep Latent Variable Model
Authors: Zhiying Jiang, Yiqin Dai, Ji Xin, Ming Li, Jimmy Lin
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We show that NPC-LV outperforms supervised methods on all three datasets on image classification in the low data regime and even outperforms semi-supervised learning methods on CIFAR-10. We compare our method with supervised learning, semi-supervised learning, non-parametric learning and traditional Non-Parametric learning by Compression (NPC) on MNIST, Fashion MNIST and CIFAR-10 [59, 60, 61]. |
| Researcher Affiliation | Collaboration | Zhiying Jiang1,2 Yiqin Dai2 Ji Xin1 Ming Li1 Jimmy Lin1 1 University of Waterloo 2 AFAIK. |
| Pseudocode | Yes | Algorithm 1 NPC-LV (use VAE and NCD as an example) |
| Open Source Code | Yes | The only new asset is our own code and we provide in supplemental material. |
| Open Datasets | Yes | We compare our method with supervised learning, semi-supervised learning, non-parametric learning and traditional Non-Parametric learning by Compression (NPC) on MNIST, Fashion MNIST and CIFAR-10 [59, 60, 61]. |
| Dataset Splits | Yes | For all datasets, we divide the data into training, validation and test splits using the same split ratios as in the official datasets. |
| Hardware Specification | Yes | All experiments were run on a single NVIDIA Tesla V100 GPU. |
| Software Dependencies | No | The model is implemented in PyTorch and we use the official API for each dataset. However, no specific version numbers for PyTorch or other software dependencies are provided. |
| Experiment Setup | Yes | We train a hierarchical latent generative model (details are in Appendix G) with an architecture of 4 layers and 64 hidden units for each layer. The model is trained using Adam optimizer with a learning rate of 0.001 and batch size of 128 for 500 epochs for MNIST and Fashion MNIST; 1000 epochs for CIFAR-10. |