Investigating privacy leakage in dimensionality reduction methods via reconstruction attack

Chayadon Lumbut, Donlapark Ponnoprat

Abstract: Highlights•We propose a novel reconstruction attack on six dimensionality reduction methods.•PCA and Isomap show the highest vulnerability to reconstruction attacks.•Sparse random projection shows the greatest resistance to reconstruction attacks.•The methods that utilize the random initialization are challenging to attack.•Additive noise mechanisms are effective in mitigating the reconstruction attack.