# Reconstructing Training Data from Diverse ML Models by Ensemble Inversion

arXiv: https://arxiv.org/abs/2111.03702 · HTML: https://mlsanity.com/paper/2111.03702/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- Builds on 12 of its 25 references that are in the map.
- First posted on arXiv 5 Nov 2021.

## Builds on (most cited first, 12 of 12)

- [Deep Residual Learning for Image Recognition](https://mlsanity.com/paper/1512.03385/) (2015)
- [Very Deep Convolutional Networks for Large-Scale Image Recognition](https://mlsanity.com/paper/1409.1556/) (2014)
- [PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space](https://mlsanity.com/paper/1706.02413/) (2017)
- [Visualizing and Understanding Convolutional Networks](https://mlsanity.com/paper/1311.2901/) (2013)
- [Membership Inference Attacks against Machine Learning Models](https://mlsanity.com/paper/1610.05820/) (2016)
- [Conditional Generative Adversarial Nets](https://mlsanity.com/paper/1411.1784/) (2014)
- [Federated Learning: Strategies for Improving Communication Efficiency](https://mlsanity.com/paper/1610.05492/) (2016)
- [VGGFace2: A dataset for recognising faces across pose and age](https://mlsanity.com/paper/1710.08092/) (2017)
- [Conditional Image Synthesis With Auxiliary Classifier GANs](https://mlsanity.com/paper/1610.09585/) (2016)
- [Snapshot Ensembles: Train 1, get M for free](https://mlsanity.com/paper/1704.00109/) (2017)
- [The Secret Revealer: Generative Model-Inversion Attacks Against Deep Neural Networks](https://mlsanity.com/paper/1911.07135/) (2019)
- [Data-Free Learning of Student Networks](https://mlsanity.com/paper/1904.01186/) (2019)

## Built on by (most cited first, 5 of 5)

- [Is it an i or an l: Test-time Adaptation of Text Line Recognition Models](https://mlsanity.com/paper/2308.15037/) (2023)
- [Ensembler: Protect Collaborative Inference Privacy from Model Inversion Attack via Selective Ensemble](https://mlsanity.com/paper/2401.10859/) (2024)
- [Responsible Active Learning via Human-in-the-loop Peer Study](https://mlsanity.com/paper/2211.13587/) (2022)
- [Rethinking Model Inversion Attacks With Patch-Wise Reconstruction](https://mlsanity.com/paper/2312.07040/) (2023)
- [Supervised machine learning for microbiomics: bridging the gap between current and best practices](https://mlsanity.com/paper/2402.17621/) (2024)

## Abstract (arXiv)

Model Inversion (MI), in which an adversary abuses access to a trained Machine Learning (ML) model attempting to infer sensitive information about its original training data, has attracted increasing research attention. During MI, the trained model under attack (MUA) is usually frozen and used to guide the training of a generator, such as a Generative Adversarial Network (GAN), to reconstruct the distribution of the original training data of that model. This might cause leakage of original training samples, and if successful, the privacy of dataset subjects will be at risk if the training data contains Personally Identifiable Information (PII). Therefore, an in-depth investigation of the potentials of MI techniques is crucial for the development of corresponding defense techniques. High-quality reconstruction of training data based on a single model is challenging. However, existing MI literature does not explore targeting multiple models jointly, which may provide additional information and diverse perspectives to the adversary. We propose the ensemble inversion technique that estimates the distribution of original training data by training a generator constrained by an ensemble (or set) of trained models with shared subjects or entities. This technique leads to noticeable improvements of the quality of the generated samples with distinguishable features of the dataset entities compared to MI of a single ML model. We achieve high quality results without any dataset and show how utilizing an auxiliary dataset that's similar to the presumed training data improves the results. The impact of model diversity in the ensemble is thoroughly investigated and additional constraints are utilized to encourage sharp predictions and high activations for the reconstructed samples, leading to more accurate reconstruction of training images.