Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

InkDrop: Invisible Backdoor Attacks Against Dataset Condensation

He Yang, Dongyi Lv, Song Ma, Wei Xi, Zhi Wang, Hanlin Gu, Yajie Wang

Abstract

Dataset Condensation (DC) is a data-efficient learning paradigm that synthesizes small yet informative datasets, enabling models to match the performance of full-data training. However, recent work exposes a critical vulnerability of DC to backdoor attacks, where malicious patterns (\textit{e.g.}, triggers) are implanted into the condensation dataset, inducing targeted misclassification on specific inputs. Existing attacks always prioritize attack effectiveness and model utility, overlooking the crucial dimension of stealthiness. To bridge this gap, we propose InkDrop, which enhances the imperceptibility of malicious manipulation without degrading attack effectiveness and model utility. InkDrop leverages the inherent uncertainty near model decision boundaries, where minor input perturbations can induce semantic shifts, to construct a stealthy and effective backdoor attack. Specifically, InkDrop first selects candidate samples near the target decision boundary that exhibit latent semantic affinity to the target class. It then learns instance-dependent perturbations constrained by perceptual and spatial consistency, embedding targeted malicious behavior into the condensed dataset. Extensive experiments across diverse datasets validate the overall effectiveness of InkDrop, demonstrating its ability to integrate adversarial intent into condensed datasets while preserving model utility and minimizing detectability. Our code is available at https://github.com/lvdongyi/InkDrop.

InkDrop: Invisible Backdoor Attacks Against Dataset Condensation

Abstract

Dataset Condensation (DC) is a data-efficient learning paradigm that synthesizes small yet informative datasets, enabling models to match the performance of full-data training. However, recent work exposes a critical vulnerability of DC to backdoor attacks, where malicious patterns (\textit{e.g.}, triggers) are implanted into the condensation dataset, inducing targeted misclassification on specific inputs. Existing attacks always prioritize attack effectiveness and model utility, overlooking the crucial dimension of stealthiness. To bridge this gap, we propose InkDrop, which enhances the imperceptibility of malicious manipulation without degrading attack effectiveness and model utility. InkDrop leverages the inherent uncertainty near model decision boundaries, where minor input perturbations can induce semantic shifts, to construct a stealthy and effective backdoor attack. Specifically, InkDrop first selects candidate samples near the target decision boundary that exhibit latent semantic affinity to the target class. It then learns instance-dependent perturbations constrained by perceptual and spatial consistency, embedding targeted malicious behavior into the condensed dataset. Extensive experiments across diverse datasets validate the overall effectiveness of InkDrop, demonstrating its ability to integrate adversarial intent into condensed datasets while preserving model utility and minimizing detectability. Our code is available at https://github.com/lvdongyi/InkDrop.

Paper Structure

This paper contains 37 sections, 21 equations, 6 figures, 10 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Dataset Condensation
  4. Backdoor against Dataset Condensation
  5. Methodology
  6. Threat Model
  7. Attack Scenario
  8. Attacker’s Goal
  9. Stealthiness (STH)
  10. Attack Effectiveness
  11. Clean Model Utility
  12. Overview of InkDrop
  13. Detailed Design of InkDrop
  14. Candidate Pool Construction
  15. Source Class Selection
  16. ...and 22 more sections

Figures (6)

  • Figure 1: Illustration of backdoor attacks against dataset condensation.
  • Figure 2: Visualization of the stealth limitations in existing backdoor condensation methods. Both synthetic data and trigger-injected test samples exhibit visible artifacts or unnatural structures, exposing the trigger and undermining the attack's stealthiness.
  • Figure 3: Illustration of constructing candidate pool.
  • Figure 4: Radar plots showing normalized performance across attack success rate (ASR), clean task accuracy (CTA), and stealthiness (STE) for each method. Each subplot corresponds to a different dataset. Curves extending toward the outer edge indicate stronger overall performance.
  • Figure 5: Comparison of stealthiness across different datasets using PSNR, SSIM, and IS$^\dagger$ metrics.
  • ...and 1 more figures