[arXiv] This is an implementation of Large Deviation Analysis for Hypothesis Testing for Score based Models
See requirements.txt
- Experimental control are configured in
config.yml
- Use
make.sh
to generate run script withmake.py
- Use
make.py
to generate exp script toscripts
- Use
make_dataset.py
to prepare datasets - Use
process.py
to process exp results - Experimental setup are listed in
make.py
- Hyperparameters can be found in
config.yml
andprocess_control()
ofmodule/hyper.py
- Test of Multivariate Normal (MVN) distribution with pertubation
$\sigma_{ptb} = 0.02$ on$\mu$ for theoretical limitpython test_ht.py --control_name MVN_mvn_lrt-t_0.02-0.0_1
- Test of KDDCUP dataset (KDDCUP99) with "back" adversarial network traffic on
$W$ of Gauss-Benoulli RBM for empirical limit ($N=10$ )python test_ht.py --control_name KDDCUP99_rbm_hst-t_back_1_10
- Large deviation analysis of likelihood-based and sore-based hypothesis testing for multivariate normal distribution with perturbation on
$\mu$ and$\sigma_{ptb} = 0.02$ .
Enmao Diao
Taposh Banerjee
Vahid Tarokh