Skip to content
forked from havelhuang/ReAsDL

The Reliability Assessment Model for Deep Learning systems

Notifications You must be signed in to change notification settings

TACPSLab/ReAsDL

 
 

Repository files navigation

ReAsDL

The Reliability Assessment Model (RAM) for Deep Learning Classifiers, presented at the AISafety'21 workshop co-located with IJCAI'21 (here is a video presentation). Ths repository contains the illustrative examples for three synthesized datasets, MNIST dataset, and CIFAR10 dataset used in the preprint.

Models trained on synthesized datasets

we train random forests on systhesized datasets and apply RAM method to evluate the reliability. To running example fold and run

python main

Deep learning models trained on MNIST and CIFAR10

we project the high dimensional inputs (images) into the latent space via Variantional Auto-Encoder (VAE), then apply RAM method get the final evaluation results. Type the command

python main('mnist', 'before', cell_size = 100, count_mh_steps = 250, count_particles = 10000)

About

The Reliability Assessment Model for Deep Learning systems

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%