arxiv, NeurIPS 2021, OpenReview

@misc{yuan2021iterative,
      title={Iterative Teacher-Aware Learning}, 
      author={Luyao Yuan and Dongruo Zhou and Junhong Shen and Jingdong Gao and Jeffrey L. Chen and Quanquan Gu and Ying Nian Wu and Song-Chun Zhu},
      year={2021},
      eprint={2110.00137},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}

To collect data and plots for a cooperative or adversarial teacher for a particular experiment, run with the command line

python3 plot_band.py -s setting_name 

where setting_name is specified by ${experiment}_${type of teacher}. Detailed settings are described in the Settings section.
And an example command to collect data and plots for Linear Classifiers on MNIST Dataset with a cooperative teacher is

python3 plot_band.py -s mnist_coop 

To use the main.py or main_irl.py script, run with the command line

python3 main.py detailed_setting_name random_seed

where detailed_setting_name is specified by ${experiment}_${type of teacher}_${mode of teacher} followed by '_' and the imitate teacher's data dimension(MNIST, Equation) or data type(CIFAR).
An example command to collect data for Linear Classifiers on MNIST Dataset with a cooperative imitate teacher with data dimension 20 and random seed 0 is

python3 main.py mnist_coop_imit_20 0

The results will be saved as pdf in the corresponding folder inside the Experiments folder.

The ${experiment} part of a setting is specified by the table

The imitate teacher's data dimension(MNIST, Equation) or data type(CIFAR) part of a setting is specified by the table

The ${type of teacher} part of a setting is specified by the table

The ${mode of teacher} part of a setting is specified by the table

In this project, we used the following version of libraries:

Tensorflow v1.15
scikit-learn v0.22.1
numpy v1.18.2

https://github.com/yuanluya/CL_Human