Adversarial Task Up-sampling for Meta-learning

Authors: Yichen WU, Long-Kai Huang, Ying Wei

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental On fewshot sine regression and image classification datasets, we empirically validate the marked improvement of ATU over state-of-the-art task augmentation strategies in the meta-testing performance and also the quality of up-sampled tasks.
Researcher Affiliation Collaboration Yichen Wu1,2 , Long-Kai Huang2 , Ying Wei1 1City University of Hong Kong, 2Tencent AI Lab
Pseudocode Yes We summarize the proposed ATU in Algorithm 1 in Appendix A.
Open Source Code No We exploit the open-source datasets and will release the code.
Open Datasets Yes We consider four datasets (base classes number): mini Imagenet-S (12), ISIC [18] (4), Dermnet-S (30), and Tabular Murris [5] (57) covering classification tasks on general natural images, medical images, and gene data. Note that the mini Imagenet-S and Dermnet-S are constructed by limiting the base classes of mini Imangenet [33] and Dermnet [1], respectively.
Dataset Splits Yes In the meta-training phase, each task contains K support and K target (K=10) examples. We adopt mean squared error (MSE) as the loss function. For the base model fθ, we adopt a small neural network, which consists of an input layer of size 1, 2 hidden layers of size 40 with Re LU and an output layer of size 1. We use one gradient update with a fixed step size α=0.01 in inner loop, and use Adam as the outer-loop optimizer following [9, 15]. Moreover, the meta-learner is trained on 240,000 tasks with meta batch-size being 4.
Hardware Specification Yes We use a workstation with Intel i7-2600 CPUs and NVIDIA GeForce RTX 2080 Ti GPUs. The training of the models takes roughly 2-3 days.
Software Dependencies No The paper mentions using 'Adam as the outer-loop optimizer' but does not specify software components with version numbers (e.g., Python, PyTorch/TensorFlow versions).
Experiment Setup Yes For the base model fθ, we adopt a small neural network, which consists of an input layer of size 1, 2 hidden layers of size 40 with Re LU and an output layer of size 1. We use one gradient update with a fixed step size α=0.01 in inner loop, and use Adam as the outer-loop optimizer following [9, 15]. Moreover, the meta-learner is trained on 240,000 tasks with meta batch-size being 4.