Toward Reliable Neural Inference for Scientific Analysis: Data-aware Bounding of Quantities of Interest

Overview

This repository contains the code and implementation details for our paper:

"Toward Reliable Neural Inference for Scientific Analysis: Data-aware Bounding of Quantities of Interest"

Running the Experiments

This section provides step-by-step instructions for reproducing our results.

1️ Data Preparation

Before running experiments, ensure that you have the necessary datasets. The full combustion data is property, and access can only be granted by PI.

2️ Training the Model

To train the model with default settings, run:

python [task]/train.py --input path_to_input_tensor.pth --target path_to_target_tensor.pth --checkpoint path_to_checkpoint.pth

For custom configurations, modify the config/ directory.

3 Quantizing the Model

To acquire quantized models of various formats:

python [task]/quantization.py --checkpoint path_to_checkpoint.pth --quantized path_to_output_folder

4 Evaluating the Model

To evaluate a trained model:

python [task]/evaluate.py --input path_to_input_tensor.pth --reduced path_to_reduced_tensor.pth --checkpoint path_to_checkpoint.pth