Learning to Learn Kernels with Variational Random Features
Authors: Xiantong Zhen, Haoliang Sun, Yingjun Du, Jun Xu, Yilong Yin, Ling Shao, Cees Snoek
ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate our Meta VRF on several few-shot learning problems for both regression and classification. We demonstrate the benefit of exploring task dependency by implementing a baseline Meta VRF (12) without using the LSTM, which infers the random base solely from the support set. We also conduct further analysis to validate the effectiveness of our Meta VRF by showing its performance with deep embedding architectures, different numbers of bases, and under versatile and challenging settings with inconsistent training and test conditions. |
| Researcher Affiliation | Academia | 1Inception Institute of Artificial Intelligence, UAE 2Informatics Institute, University of Amsterdam, The Netherlands 3School of Software, Shandong University, China 4College of Computer Science, Nankai University, China 5Mohamed bin Zayed University of Artificial Intelligence, Abu Dhabi, UAE. |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide any explicit statement about making the source code available, nor does it include a link to a code repository. |
| Open Datasets | Yes | The classification experiments are conducted on three commonly-used benchmark datasets, i.e., Omniglot (Lake et al., 2015), mini Image Net (Vinyals et al., 2016) and CIFAR-FS (Krizhevsky et al., 2009); for more details, please refer to the supplementary material. |
| Dataset Splits | No | The paper describes an episodic training strategy using 'support sets' and 'query sets' within tasks during meta-training and meta-testing stages. However, it does not explicitly provide percentages or counts for a separate 'validation' dataset split in the traditional sense, or reference predefined global validation splits for reproducibility. |
| Hardware Specification | No | The paper does not provide specific details regarding the hardware used to run the experiments, such as GPU/CPU models, memory, or cloud instance specifications. |
| Software Dependencies | No | The paper does not provide specific software dependencies or their version numbers, such as programming languages, libraries, or frameworks used for implementation. |
| Experiment Setup | Yes | For a fair comparison, we compute the feature embedding using a small multi-layer perception (MLP) with two hidden layers of size 40, following the same settings used in MAML.The inference network φ( ) is a three-layer MLP with 256 units in the hidden layers and rectifier non-linearity where input sizes are 256 and 512 for the vanilla and bidirectional LSTMs, respectively.The key hyperparameter for the number of bases D in (7) is set to D = 780 for Meta VRF in all experiments, while we use RFFs with D = 2048 as this produces the best performance. |