Effective Meta-Regularization by Kernelized Proximal Regularization
Authors: Weisen Jiang, James Kwok, Yu Zhang
NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results confirm the advantage of the proposed algorithm. |
| Researcher Affiliation | Academia | 1 Department of Computer Science and Engineering, Southern University of Science and Technology 2 Department of Computer Science and Engineering, Hong Kong University of Science and Technology 3 Peng Cheng Laboratory |
| Pseudocode | Yes | Algorithm 1 MAML [12]. Algorithm 2 Common Mean [7]. Algorithm 3 Meta Prox. |
| Open Source Code | No | The paper does not provide concrete access to source code for the methodology described. It mentions using 'CVXPYLayers package [1]' but does not state that their own implementation code is released. |
| Open Datasets | Yes | Data sets. Experiments are performed on three data sets. (i) Sine. ... (ii) Sale. This is a real-world dataset from [36] ... (iii) QMUL, which is a multiview face dataset [16] from Queen Mary University of London. ... We use the standard 5-way K-shot setting (K = 1 or 5) on the mini-Image Net [39] dataset, which consists of 100 randomly chosen classes from ILSVRC-2012 [33]. |
| Dataset Splits | Yes | We randomly generate a meta-training set of 8000 tasks, a meta-validation set of 1000 tasks for early stopping, and a meta-testing set of 2000 tasks for performance evaluation. ... We randomly split the tasks into a meta-training set of 600 tasks, a meta-validation set of 100 tasks, and a meta-testing set of 111 tasks. ... the 100 classes are randomly split into 64 for meta-training, 16 for meta-validation, and 20 for meta-testing. |
| Hardware Specification | No | The paper does not provide specific hardware details (exact GPU/CPU models, processor types, or memory amounts) used for running its experiments. |
| Software Dependencies | No | The paper mentions 'Adam optimizer [20]' and 'CVXPYLayers package [1]' but does not provide specific version numbers for these software dependencies to ensure reproducibility. |
| Experiment Setup | Yes | We use the Adam optimizer [20] with a learning rate of 0.001. Each minibatch has 16 tasks. For Sine and Sale, the model (φ and fθ) is meta-trained for 40, 000 iterations. ... We train the model for 80, 000 iterations, and each mini-batch has 4 tasks. We use the Adam optimizer [20] with an initial learning rate of 0.001, which is then reduced by half every 2, 500 iterations. |