iurada/robust-segmentation

DieHardNET

Repository for a reliable Deep Neural Network (DNN) model. DieHardNet stands for Die (integrated circuit) Hardened Neural Network

Directories

The directories are organized as follows:

• hg_noise_injector - A module to inject realistic errors in the training process
• eval_fault_injection_cfg - Configuration files for NVBITFI for fault injection
• pytorch_scripts - PyTorch scripts for training and inference for the used DNNs. For information read the README.

Run the training process

1. With the Hans Gruber error injector

$ ./main.py ....

2. Without the error injector

$ ./main.py ....

Run the inference

An example of running the inference with the pretrained hardened model

$ ./main.py ....

To cite this work

@INPROCEEDINGS{diehardnet,
    author = {AUTHORS},
    title = {TITLE},
    booktitle = {PROCEEDINGS},
    series = {EVENT},
    year = {2022},
    isbn = {ISBN},
}