# Not All Poisons are Created Equal: Robust Training against Data Poisoning

arXiv: https://arxiv.org/abs/2210.09671 · HTML: https://mlsanity.com/paper/2210.09671/

## Key facts

- Cited by 30 papers in the map (data through 14 Oct 2026).
- Placed in Backdoor and data poisoning attacks, one of 12 directions in LLM safety and security.
- Builds on 29 of its 40 references that are in the map.
- First posted on arXiv 18 Oct 2022.

## Placed in

- LLM safety and security / [Backdoor and data poisoning attacks](https://mlsanity.com/safety/backdoor-and-data-poisoning-attacks/)

## Builds on (most cited first, 29 of 29)

- [Very Deep Convolutional Networks for Large-Scale Image Recognition](https://mlsanity.com/paper/1409.1556/) (2014)
- [Towards Deep Learning Models Resistant to Adversarial Attacks](https://mlsanity.com/paper/1706.06083/) (2017)
- [Deep Learning with Differential Privacy](https://mlsanity.com/paper/1607.00133/) (2016)
- [Convex optimization](https://mlsanity.com/paper/2106.01946/) (2021)
- [BadNets: Identifying Vulnerabilities in the Machine Learning Model Supply Chain](https://mlsanity.com/paper/1708.06733/) (2017)
- [Targeted Backdoor Attacks on Deep Learning Systems Using Data Poisoning](https://mlsanity.com/paper/1712.05526/) (2017)
- [Poison Frogs! Targeted Clean-Label Poisoning Attacks on Neural Networks](https://mlsanity.com/paper/1804.00792/) (2018)
- [Spectral Signatures in Backdoor Attacks](https://mlsanity.com/paper/1811.00636/) (2018)
- [Detecting Backdoor Attacks on Deep Neural Networks by Activation Clustering](https://mlsanity.com/paper/1811.03728/) (2018)
- [Hidden Trigger Backdoor Attacks](https://mlsanity.com/paper/1910.00033/) (2019)
- [Not All Samples Are Created Equal: Deep Learning with Importance Sampling](https://mlsanity.com/paper/1803.00942/) (2018)
- [Certified Defenses for Data Poisoning Attacks](https://mlsanity.com/paper/1706.03691/) (2017)
- [Anti-Backdoor Learning: Training Clean Models on Poisoned Data](https://mlsanity.com/paper/2110.11571/) (2021)
- [Deep learning versus kernel learning: an empirical study of loss landscape geometry and the time evolution of the Neural Tangent Kernel](https://mlsanity.com/paper/2010.15110/) (2020)
- [Witches' Brew: Industrial Scale Data Poisoning via Gradient Matching](https://mlsanity.com/paper/2009.02276/) (2020)
- [Stronger Data Poisoning Attacks Break Data Sanitization Defenses](https://mlsanity.com/paper/1811.00741/) (2018)
- [Transferable Clean-Label Poisoning Attacks on Deep Neural Nets](https://mlsanity.com/paper/1905.05897/) (2019)
- [MetaPoison: Practical General-purpose Clean-label Data Poisoning](https://mlsanity.com/paper/2004.00225/) (2020)
- [RAB: Provable Robustness Against Backdoor Attacks](https://mlsanity.com/paper/2003.08904/) (2020)
- [Just How Toxic is Data Poisoning? A Unified Benchmark for Backdoor and Data Poisoning Attacks](https://mlsanity.com/paper/2006.12557/) (2020)
- [Deep Partition Aggregation: Provable Defense against General Poisoning Attacks](https://mlsanity.com/paper/2006.14768/) (2020)
- [Sleeper Agent: Scalable Hidden Trigger Backdoors for Neural Networks Trained from Scratch](https://mlsanity.com/paper/2106.08970/) (2021)
- [Lazier Than Lazy Greedy](https://mlsanity.com/paper/1409.7938/) (2014)
- [Strong Data Augmentation Sanitizes Poisoning and Backdoor Attacks Without an Accuracy Tradeoff](https://mlsanity.com/paper/2011.09527/) (2020)
- [On the Effectiveness of Mitigating Data Poisoning Attacks with Gradient Shaping](https://mlsanity.com/paper/2002.11497/) (2020)
- [Data Poisoning against Differentially-Private Learners: Attacks and Defenses](https://mlsanity.com/paper/1903.09860/) (2019)
- [Bullseye Polytope: A Scalable Clean-Label Poisoning Attack with Improved Transferability](https://mlsanity.com/paper/2005.00191/) (2020)
- [Better Safe Than Sorry: Preventing Delusive Adversaries with Adversarial Training](https://mlsanity.com/paper/2102.04716/) (2021)
- [On Evaluating Neural Network Backdoor Defenses](https://mlsanity.com/paper/2010.12186/) (2020)

## Built on by (most cited first, 30 of 30)

- [CleanCLIP: Mitigating Data Poisoning Attacks in Multimodal Contrastive Learning](https://mlsanity.com/paper/2303.03323/) (2023)
- [Shadowcast: Stealthy Data Poisoning Attacks Against Vision-Language Models](https://mlsanity.com/paper/2402.06659/) (2024)
- [SmallToLarge (S2L): Scalable Data Selection for Fine-tuning Large Language Models by Summarizing Training Trajectories of Small Models](https://mlsanity.com/paper/2403.07384/) (2024)
- [Hidden Poison: Machine Unlearning Enables Camouflaged Poisoning Attacks](https://mlsanity.com/paper/2212.10717/) (2022)
- [Data and Model Poisoning Backdoor Attacks on Wireless Federated Learning, and the Defense Mechanisms: A Comprehensive Survey](https://mlsanity.com/paper/2312.08667/) (2023)
- [Rethinking Backdoor Attacks](https://mlsanity.com/paper/2307.10163/) (2023)
- [Mitigating Unauthorized Speech Synthesis for Voice Protection](https://mlsanity.com/paper/2410.20742/) (2024)
- [BELT: Old-School Backdoor Attacks can Evade the State-of-the-Art Defense with Backdoor Exclusivity Lifting](https://mlsanity.com/paper/2312.04902/) (2023)
- [Differentially Private Optimizers Can Learn Adversarially Robust Models](https://mlsanity.com/paper/2211.08942/) (2022)
- [No, of Course I Can! Deeper Fine-Tuning Attacks That Bypass Token-Level Safety Mechanisms](https://mlsanity.com/paper/2502.19537/) (2025)
- [A Proxy Attack-Free Strategy for Practically Improving the Poisoning Efficiency in Backdoor Attacks](https://mlsanity.com/paper/2306.08313/) (2023)
- [AdvBDGen: Adversarially Fortified Prompt-Specific Fuzzy Backdoor Generator Against LLM Alignment](https://mlsanity.com/paper/2410.11283/) (2024)
- [FCert: Certifiably Robust Few-Shot Classification in the Era of Foundation Models](https://mlsanity.com/paper/2404.08631/) (2024)
- [PureGen: Universal Data Purification for Train-Time Poison Defense via Generative Model Dynamics](https://mlsanity.com/paper/2405.18627/) (2024)
- [PoisonedParrot: Subtle Data Poisoning Attacks to Elicit Copyright-Infringing Content from Large Language Models](https://mlsanity.com/paper/2503.07697/) (2025)
- [Effective Backdoor Mitigation in Vision-Language Models Depends on the Pre-training Objective](https://mlsanity.com/paper/2311.14948/) (2023)
- [Clean Label Attacks against SLU Systems](https://mlsanity.com/paper/2409.08985/) (2024)
- [Unlearnable Examples Give a False Sense of Data Privacy: Understanding and Relearning](https://mlsanity.com/paper/2306.02064/) (2023)
- [HINT: Healthy Influential-Noise based Training to Defend against Data Poisoning Attacks](https://mlsanity.com/paper/2309.08549/) (2023)
- [Have You Poisoned My Data? Defending Neural Networks against Data Poisoning](https://mlsanity.com/paper/2403.13523/) (2024)
- [Vera Verto: Multimodal Hijacking Attack](https://mlsanity.com/paper/2408.00129/) (2024)
- [Generalizable Targeted Data Poisoning against Varying Physical Objects](https://mlsanity.com/paper/2412.03908/) (2024)
- [Angel or Demon: Investigating the Plasticity Interventions' Impact on Backdoor Threats in Deep Reinforcement Learning](https://mlsanity.com/paper/2605.14587/) (2026)
- [Mendata: A Framework to Purify Manipulated Training Data](https://mlsanity.com/paper/2312.01281/) (2023)
- [Robust Influence-based Training Methods for Noisy Brain MRI](https://mlsanity.com/paper/2403.10698/) (2024)
- [PureEBM: Universal Poison Purification via Mid-Run Dynamics of Energy-Based Models](https://mlsanity.com/paper/2405.19376/) (2024)
- [Like Oil and Water: Group Robustness Methods and Poisoning Defenses May Be at Odds](https://mlsanity.com/paper/2504.02142/) (2025)
- [Addressing The Devastating Effects Of Single-Task Data Poisoning In Exemplar-Free Continual Learning](https://mlsanity.com/paper/2507.04106/) (2025)
- [TraceGuard: Adaptive Multimodal Poison Filtering through Cross-Feature Rank Agreement](https://mlsanity.com/paper/2609.29099/) (2026)
- [SAGE: Similarity-Based Cleaning of Poisoned Training Data from Verified Examples](https://mlsanity.com/paper/2610.01788/) (2026)

## Abstract (arXiv)

Data poisoning causes misclassification of test time target examples by injecting maliciously crafted samples in the training data. Existing defenses are often effective only against a specific type of targeted attack, significantly degrade the generalization performance, or are prohibitive for standard deep learning pipelines. In this work, we propose an efficient defense mechanism that significantly reduces the success rate of various data poisoning attacks, and provides theoretical guarantees for the performance of the model. Targeted attacks work by adding bounded perturbations to a randomly selected subset of training data to match the targets' gradient or representation. We show that: (i) under bounded perturbations, only a number of poisons can be optimized to have a gradient that is close enough to that of the target and make the attack successful; (ii) such effective poisons move away from their original class and get isolated in the gradient space; (iii) dropping examples in low-density gradient regions during training can successfully eliminate the effective poisons, and guarantees similar training dynamics to that of training on full data. Our extensive experiments show that our method significantly decreases the success rate of state-of-the-art targeted attacks, including Gradient Matching and Bullseye Polytope, and easily scales to large datasets.