An Embarrassingly Simple Approach to Semi-Supervised Few-Shot Learning
Authors: Xiu-Shen Wei, H.-Y. Xu, Faen Zhang, Yuxin Peng, Wei Zhou
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct experiments on four widely-used few-shot learning benchmark datasets for general object recognition and fine-grained classification, including mini Image Net [25], tiered Image Net [26], CIFAR-FS [2] and CUB [34]. |
| Researcher Affiliation | Collaboration | Xiu-Shen Wei1,2, He-Yang Xu1, Faen Zhang3, Yuxin Peng4 , Wei Zhou5 1School of Computer Science and Engineering, Nanjing University of Science and Technology 2State Key Laboratory of Integrated Services Networks, Xidian University 3Qingdao AInnovation Technology Group Co., Ltd 4Wangxuan Institute of Computer Technology, Peking University 5CICC Alpha (Beijing) Private Equity |
| Pseudocode | Yes | Algorithm 1 Pseudo-code of the proposed MUSIC |
| Open Source Code | Yes | Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [Yes] See Section 4.1. |
| Open Datasets | Yes | We conduct experiments on four widely-used few-shot learning benchmark datasets for general object recognition and fine-grained classification, including mini Image Net [25], tiered Image Net [26], CIFAR-FS [2] and CUB [34]. |
| Dataset Splits | Yes | For fair comparisons, we obey the protocol of data splits in [9, 15, 36] to train the feature embedding function and conduct experiments for evaluations in SSFSL. |
| Hardware Specification | Yes | All experiments are conducted by Mind Spore with a Ge Force RTX 3060 GPU. |
| Software Dependencies | No | The paper mentions 'Mind Spore' but does not provide specific version numbers for it or any other software dependencies. |
| Experiment Setup | Yes | For optimization, Stochastic Gradient Descent (SGD) with momentum of 0.9 and weight decay of 5 10 4 is adopted as the optimizer to train the feature extractor from scratch. The initial learning rate is 0.1, and decayed as 6 10 3, 1.2 10 3 and 2.4 10 4 after 60, 70 and 80 epochs, by following [38]. Regarding the hyper-parameters in MUSIC, the reject option δ in Eqn. (4) is set to 1 c and the trade-off parameter over Eqn. (6) is set to 1 as default for all experiments and iterations, which shows its practicality and non-tricky. |