/AAAI2022-RTA-Iterative-Search-Mindspore

AAAI2022 Data-Centric Robust Learning on ML Models Rank 10th

Primary LanguageJupyter Notebook

paper link: http://alisec-competition.oss-cn-shanghai.aliyuncs.com/competition_papers/20211201/rank10.pdf

image

This work is sponsored by Natural Science Foundation of China(62276242), CAAI-Huawei MindSpore Open Fund(CAAIXSJLJJ-2021-016B), Anhui Province Key Research and Development Program(202104a05020007), and USTC Research Funds of the Double First-Class Initiative(YD2350002001)”。

We use Mindspore to develop our algorithm.

The generated partial image candidate set

image

Experiment results

image

How to run

  1. Obtain all Cifar10 data from the official website to the folder CIFAR-10-FANs-PY;
  2. Create two empty folders. new_data and train_model;
  3. Run gen_corr_data.ipynb to generate corruption data;
  4. Run final_gen_data_iter to generate the first generation of the data;
  5. Run train.py to generate the first generation model;
  6. Generate the next generation of the data.
  7. Loop 5 ~ 6 until num = 6. In general, it takes 5 times to train the model to get the final submitted model, and 4 times to get the final submitted data.npy.