Meta-Weight-Net: Learning an Explicit Mapping For Sample Weighting
Authors: Jun Shu, Qi Xie, Lixuan Yi, Qian Zhao, Sanping Zhou, Zongben Xu, Deyu Meng
NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Synthetic and real experiments substantiate the capability of our method for achieving proper weighting functions in class imbalance and noisy label cases, fully complying with the common settings in traditional methods, and more complicated scenarios beyond conventional cases. This naturally leads to its better accuracy than other state-of-the-art methods. |
| Researcher Affiliation | Academia | Jun Shu1, Qi Xie1, Lixuan Yi1, Qian Zhao1, Sanping Zhou1, Zongben Xu1, and Deyu Meng*2,1 1Xi an Jiaotong University 2The Macau University of Science and Technology |
| Pseudocode | Yes | Algorithm 1 The MW-Net Learning Algorithm |
| Open Source Code | Yes | Source code is available at https://github.com/xjtushujun/meta-weight-net. |
| Open Datasets | Yes | We use Long-Tailed CIFAR dataset [60], that reduces the number of training samples per class according to an exponential function... Two benchmark datasets are employed: CIFAR-10 and CIFAR-100 [62]... We conduct experiments on the Clothing1M dataset [64] |
| Dataset Splits | Yes | We randomly selected 10 images per class in validation set as the meta-data set. |
| Hardware Specification | No | The paper does not provide specific hardware details (exact GPU/CPU models, processor types, or memory amounts) used for running its experiments. |
| Software Dependencies | No | The algorithm can be easily implemented using popular deep learning frameworks like Py Torch [36]. |
| Experiment Setup | Yes | We trained Res Net-32 [61] with softmax cross-entropy loss by SGD with a momentum 0.9, a weight decay 5 × 10−4, an initial learning rate 0.1. The learning rate of Res Net-32 is divided by 10 after 80 and 90 epoch (for a total 100 epochs), and the learning rate of WN-Net is fixed as 10−5. |