Meta-AdaM: An Meta-Learned Adaptive Optimizer with Momentum for Few-Shot Learning

Authors: Siyuan Sun, Hongyang Gao

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct experiments to evaluate the proposed Meta-Ada M using three benchmark datasets. Extensive experimental results on benchmark datasets demonstrate the effectiveness of the proposed Meta-Ada M.
Researcher Affiliation Academia Siyuan Sun Department of Computer Science Iowa State University Ames, IA 50011 sxs14473@iastate.edu Hongyang Gao Department of Computer Science Iowa State University Ames, IA 50011 hygao@iastate.edu
Pseudocode Yes Algorithm 1 Meta Momentum Inner. Algorithm 2 Meta-Ada M.
Open Source Code No The paper does not provide a link to open-source code for the described methodology or explicitly state that the code is released.
Open Datasets Yes We evaluate the proposed methods using three datasets: Mini-Image Net [10], Tiered Image Net [33], and Cifar100 [16] datasets. These datasets are bench-marking datasets in the FSL domain.
Dataset Splits Yes Following previous works [30], we split 100 classes into 64, 16, and 20 class groups for training, validation, and testing.
Hardware Specification Yes We use the experimental setting of 5way-1-shot on the Mini-Image Net dataset and record the running time on an NVIDIA RTX A4000 GPU.
Software Dependencies No The paper does not specify any software dependencies with version numbers (e.g., Python version, specific deep learning frameworks like PyTorch or TensorFlow versions).
Experiment Setup Yes We show the hyperparamters in Table 2, which applies to each dataset and each backbone. Table 2: Hyperparameters for experiment. Hyperpameter Value tasks batch size 2 inner learning rate η 0.01 outer learning rate α, β 0.001 # inner fine-tune step 5 # training epochs 100 # outer steps in each epoch 500