Task Affinity with Maximum Bipartite Matching in Few-Shot Learning
Authors: Cat Phuoc Le, Juncheng Dong, Mohammadreza Soltani, Vahid Tarokh
ICLR 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Using extensive simulations, we demonstrate that our approach of utilizing only the related training data is an effective method for boosting the performance of the few-shot model with less number of parameters in both 5-way 1-shot and 5-way 5-shot settings for various benchmark datasets. Experimental results on mini Image Net (Vinyals et al., 2016), tiered Image Net (Ren et al., 2018), CIFARFS (Bertinetto et al., 2018), and FC-100 (Oreshkin et al., 2018) datasets are provided demonstrating the efficacy of the proposed approach compared to other state-of-the-art few-shot learning methods. |
| Researcher Affiliation | Academia | Cat P. Le, Juncheng Dong, Mohammadreza Soltani, Vahid Tarokh Department of Electrical and Computer Engineering, Duke University |
| Pseudocode | Yes | Algorithm 1: Few-Shot Learning with Task Affinity Score |
| Open Source Code | No | The paper does not provide any concrete access to source code, such as a specific repository link, an explicit code release statement, or code in supplementary materials. |
| Open Datasets | Yes | Experimental results on mini Image Net (Vinyals et al., 2016), tiered Image Net (Ren et al., 2018), CIFARFS (Bertinetto et al., 2018), and FC-100 (Oreshkin et al., 2018) datasets are provided demonstrating the efficacy of the proposed approach compared to other state-of-the-art few-shot learning methods. |
| Dataset Splits | Yes | The mini Image Net dataset consists of 100 classes, sampled from Image Net (Russakovsky et al., 2015), and randomly split into 64, 16, and 20 classes for training, validation, and testing, respectively. Similarly, the tiered Image Net dataset is also a derivative of Image Net, containing a total of 608 classes from 34 categories. It is split into 20, 6, and 8 categories (or 351, 97, and 160 classes) for training, validation, and testing, respectively. |
| Hardware Specification | No | The paper does not provide specific hardware details such as exact GPU/CPU models, processor types, or memory amounts used for running its experiments. |
| Software Dependencies | No | The paper mentions optimizers and classifiers but does not specify any software libraries or their version numbers, such as Python, PyTorch, or TensorFlow, that are needed to replicate the experiment. |
| Experiment Setup | Yes | For training the Whole-Classification network, the SGD optimizer is applied with the momentum of 0.9, and the learning rate is initialized at 0.05 with the decay factor of 0.1 for all experiments. In mini Image Net experiment, we train the model for 100 epochs with a batch size of 64, and the learning rate decaying at epoch 90. For tiered Image Net, we train for 120 epochs with a batch size of 64, and the learning rate decaying two times at epochs 40 and 80. In the episodic fine-tuning phase, we use the SGD optimizer with momentum 0.9, and the learning rate is set to 0.001. The batch size is set to 4, where each batch of data includes 4 few-shot tasks. |