# Reconstructing Network Inputs with Additive Perturbation Signatures

## TL;DR

This paper explores how limited access to a model's outputs and additive input perturbations can be used to reconstruct significant information about the model's secret inputs.

## Contribution

It introduces a method to recover input information from models using additive perturbation signatures with minimal output access.

## Key findings

- Partial input reconstruction demonstrated
- Effective with limited model output access
- Potential implications for model privacy

## Abstract

In this work, we present preliminary results demonstrating the ability to recover a significant amount of information about secret model inputs given only very limited access to model outputs and the ability evaluate the model on additive perturbations to the input.