Efficient Heterogeneous Meta-Learning via Channel Shuffling Modulation

Authors: Minh Hoang, Carl Kingsford

ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our proposed methods on two public datasets, namely Mini-ImageNet and Tiered-ImageNet [35, 34]. For Mini-ImageNet, each image is of size 84 × 84. It contains 100 classes with 600 images per class. Following the common setting [3, 23], we divide Mini-ImageNet into 64, 16, and 20 classes for meta-training, meta-validation, and meta-testing, respectively. For Tiered-ImageNet, each image is of size 84 × 84. It contains 608 classes which are divided into 34, 10, and 16 classes for meta-training, meta-validation, and meta-testing, respectively. We follow the standard few-shot classification setting for all experiments. In particular, we evaluate our method on the 5-way 1-shot/5-shot settings. For a fair comparison, we repeat each experiment 3 times using different random seeds and report the average performance.
Researcher Affiliation Academia Hanyang University, Korea
Pseudocode No The paper describes the method algorithmically but does not provide a formal pseudocode or algorithm block.
Open Source Code No The paper does not provide a statement or link indicating the availability of open-source code for the described methodology.
Open Datasets Yes We evaluate our proposed methods on two public datasets, namely Mini-ImageNet and Tiered-ImageNet [35, 34]. For Mini-ImageNet, each image is of size 84 × 84. It contains 100 classes with 600 images per class. Following the common setting [3, 23], we divide Mini-ImageNet into 64, 16, and 20 classes for meta-training, meta-validation, and meta-testing, respectively. For Tiered-ImageNet, each image is of size 84 × 84. It contains 608 classes which are divided into 34, 10, and 16 classes for meta-training, meta-validation, and meta-testing, respectively.
Dataset Splits Yes For Mini-ImageNet, each image is of size 84 × 84. It contains 100 classes with 600 images per class. Following the common setting [3, 23], we divide Mini-ImageNet into 64, 16, and 20 classes for meta-training, meta-validation, and meta-testing, respectively. For Tiered-ImageNet, each image is of size 84 × 84. It contains 608 classes which are divided into 34, 10, and 16 classes for meta-training, meta-validation, and meta-testing, respectively.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper does not provide specific software dependencies with version numbers (e.g., PyTorch 1.9, Python 3.8).
Experiment Setup Yes For all few-shot classification experiments, we use ResNet-18 [10] as the backbone. In the meta-training stage, we use stochastic gradient descent (SGD) as the optimizer with a learning rate of 0.001. We train our model for 100 epochs, and the learning rate is decayed by a factor of 0.1 at 80 epochs. We use a batch size of 256. For data augmentation, we use random horizontal flip, random crop, and color jittering. In the meta-testing stage, for fair comparison, we use the same evaluation protocol as [23, 24]. In particular, we randomly sample 600 episodes for each setting and report the average accuracy and 95% confidence interval.