MetaModulation: Learning Variational Feature Hierarchies for Few-Shot Learning with Fewer Tasks
Authors: Wenfang Sun, Yingjun Du, Xiantong Zhen, Fan Wang, Ling Wang, Cees G. M. Snoek
ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To verify our method, we conduct experiments on four few-task meta-learning benchmarks: mini Imagenet-S, ISIC, Derm Net-S, and Tabular Murris. We perform a series of ablation studies to investigate the benefits of using a learnable task modulation method at various levels of complexity. Our goal is to illustrate the advantages of increasing task diversity through such a method, as well as demonstrate the benefits of incorporating probabilistic variations in the few-task meta-learning framework. |
| Researcher Affiliation | Collaboration | 1 Hefei Institutes of Physical Science, Chinese Academy of Sciences, Hefei 230031, China/P. R. China. 2University of Science and Technology of China, Hefei 230026, China/P. R. China. 3University of Amsterdam, Amsterdam, the Netherlands. 4United Imaging Healthcare, Co., Ltd., China. |
| Pseudocode | No | The paper describes its methods through prose and diagrams but does not include formal pseudocode or algorithm blocks. |
| Open Source Code | Yes | Code available at: https: //github.com/lmsdss/Meta Modulation. |
| Open Datasets | Yes | Datasets. We conduct experiments on four few-task meta-learning challenges, i.e., mini Imagenet, ISIC, Derm Net and Tabular Murris (Cao et al., 2020). mini Imagenet (Vinyals et al., 2016) is constructed from Image Net (Deng et al., 2009)... ISIC (Milton, 2019) aims to classify dermoscopic images... |
| Dataset Splits | No | The paper describes how N-way k-shot tasks are sampled for meta-training and meta-testing but does not provide specific train/validation/test dataset splits (e.g., percentages or absolute counts) for the underlying datasets like mini Imagenet, ISIC, Derm Net, and Tabular Murris. The word 'validation' appears only in the context of related work. |
| Hardware Specification | No | The paper does not provide specific details about the hardware (e.g., CPU, GPU models, or cloud computing instances) used for running the experiments. |
| Software Dependencies | No | The paper mentions the use of the Adam optimizer but does not provide specific version numbers for any software libraries or dependencies (e.g., Python, PyTorch, TensorFlow, CUDA). |
| Experiment Setup | Yes | For all experiments, we use an initial learning rate of 10 3 and an SGD optimizer with Adam (Kingma & Ba, 2014). The variational neural network is parameterized by three feed-forward multiple-layer perception networks and a Re LU activation layer. The number of Monte Carlo samples is 20. The batch and query sizes of all datasets are set as 4 and 15. The total training iterations are 50,000. |