Sampling-Based Techniques for Training Deep Neural Networks with Limited Computational Resources

Approx_LSH: Contains implementation of ALSH-approx. Run main.py to see the results. The configuration can be changed in main.py.

Regular_NN: Contains implementation of vanilla neural network. Run main.py to see the results.

DropOut: Contains implementations of Dropout and Adaptive Dropout. Run main.py to see the results.

Activation function: ReLU; weight initialization: Kaiming initialization

Loss Function: Negative Log-Likelihood

Datasets: MNIST, NORB, Fashion-MNIST

Input and output dimension must change according to the dataset.

Dataset	Input	Output
NORB	9216	5
MNIST	784	10
Fashion-MNIST	784	10

Check out MC-approx here: https://github.com/acsl-technion/approx

Find the supplementary material in Appendix.pdf.