The experiments and plots were conducted with the code provided.
The experiment comparing the capacity of various K-Lipschitz constraints method was conducted by file:
00.0_2D_GP_vs_LP_vs_SN_vs_GCLP-Benchmark-All.ipynb
Category of Functions: 02.0_Plots_from_data.ipynb
Invex Function Visual Proof: 03.0_Invex_Visual_Proof.ipynb
GC-GP Function visualization: 04.0_Penalty and clipping function.ipynb
The experiment comparing the capacity of Invex function alongside Linear, Convex and Ordinary Neural Networks was conducted by:
2D Classification: 01.0_2D_cls_Logistic_vs_Convex_vs_Invex_vs_NN_Benchmark.ipynb
2D Regression: 01.1_2D_reg_Linear_vs_Convex_vs_Invex_vs_NN_Benchmark.ipynb
MNIST : MLP 05.0_Logistic_vs_Convex_vs_Invex_vs_NN_Mnist-Benchmark.ipynb
MNIST : CNN 05.1_Logistic_vs_Convex_vs_Invex_vs_NN_MnistCNN-Benchmark.ipynb
F-MNIST : CNN 05.2_Convex_vs_Invex_vs_NN_FMnistCNN-Benchmark.ipynb
1D-partial input (in/con)-vex neural network: 06.0_Energy_minimization_PIINN_vs_PICNN_diagrams.ipynb
Gaussian Mixture Model disconnected decision boundary: 07.0_GMM_decision_boundary_viz.ipynb
Convex vs Connected set for regions in 2D classification: 08.0_Toy_MultiConnected_set_classification_benchmark.ipynb
Invertible + Connected Classifier for Network Morphism in 2D classification: 08.1_Toy_MultiConnected_set_classification_add_remove_centers.ipynb
2D Invex embedding visualization of F-MNIST: 09.0_2D_Multi_Invex_vs_Ordinary_fMNIST.ipynb
2D Invex embedding visualization of CIFAR-10: 09.1_2D_Multi_Invex_vs_Ordinary_Cifar.ipynb
Multi-Invex classifier region interpretation F-MNIST: 10.0_Multi_Invex_Visualize_Centers_(BNvsAN)_fMNIST.ipynb
Multi-Invex classifier region interpretation CIFAR-10: 10.1_Multi_Invex_Visualize_Centers_(BNvsAN)_Cifar.ipynb
MNIST/FMNIST/CIFAR-10/CIFAR-100 invertible/ordinary + Connected-clf/MLP-clf experiment python bench_script.py
Uncertainity representation with local classifier: 11.0_Multi_Invex_MSE_classifier_2D_with_Uncertainity_(NotQuiteWorking).ipynb