This project is the code for AAAI 2023 paper "Understanding Representation Learnability of Nonlinear Self-Supervised Learning".
We conduct simulation experiments on the nonlinear SSL model and the nonlinear SL model, and their update processes are in the following files:
- SSL_update_process.py
- SL_update_process.py
We will illustrate the role of hyperparameters. The hyperparameters d, n, T,
You can modify the above hyperparameters to get different results for the SSL or SL model. The only thing worth noting is that you need to change the hyperparameter path to the path where the result is stored when running the program.
In the folder, we include four programs:
- Figure_learning_curve.py
- Figure_learning_results.py
- Figure_projection_e1.py
- Figure_projection_e2.py
, and each program contains a function "get_data". You only need to change the results name in this function to get the final figure.