TrojAI Literature Review

The list below contains curated papers and arXiv articles that are related to Trojan attacks, backdoor attacks, and data poisoning on neural networks and machine learning systems. They are ordered "approximately" from most to least recent and articles denoted with a "*" mention the TrojAI program directly. Some of the particularly relevant papers include a summary that can be accessed by clicking the "Summary" drop down icon underneath the paper link. These articles were identified using variety of methods including:

• A flair embedding created from the arXiv CS subset
• A trained ASReview random forest model
• A curated manual literature review

1. TA-CLEANER: A FINE-GRAINED TEXT ALIGNMENT BACKDOOR DEFENSE STRATEGY FOR MULTIMODAL CONTRASTIVE

2. WEAK-TO-STRONG BACKDOOR ATTACKS FOR LLMS WITH CONTRASTIVE KNOWLEDGE DISTILLATION

3. Data-centric NLP Backdoor Defense from the Lens of Memorization

4. Obliviate: Neutralizing Task-agnostic Backdoors within the Parameter-efficient Fine-tuning Paradigm

5. Large Language Models Are Better Adversaries: Exploring Generative Clean-Label Backdoor Attacks Against Text Classifiers

6. PoisonedRAG: Knowledge Corruption Attacks to Retrieval-Augmented Generation of Large Language Models

7. Here's a Free Lunch: Sanitizing Backdoored Models with Model Merge

8. TrojFM: Resource-efficient Backdoor Attacks against Very Large Foundation Models

9. A Survey of Backdoor Attacks and Defenses on Large Language Models: Implications for Security Measures

10. Unlearning Backdoor Threats: Enhancing Backdoor Defense in Multimodal Contrastive Learning via Local Token Unlearning

11. Transferring Backdoors between Large Language Models by Knowledge Distillation

12. Composite Backdoor Attacks Against Large Language Models

13. CleanGen: Mitigating Backdoor Attacks for Generation Tasks in Large Language Models

14. LoRA-as-an-Attack! Piercing LLM Safety Under The Share-and-Play Scenario

15. BadChain: Backdoor Chain-of-Thought Prompting for Large Language Models

16. BadAgent: Inserting and Activating Backdoor Attacks in LLM Agents

17. Chain-of-Scrutiny: Detecting Backdoor Attacks for Large Language Models

18. BEEAR: Embedding-based Adversarial Removal of Safety Backdoors in Instruction-tuned Language Models

19. Is poisoning a real threat to LLM alignment? Maybe more so than you think

20. ADAPTIVEBACKDOOR: Backdoored Language Model Agents that Detect Human Overseers

21. Mitigating Fine-tuning based Jailbreak Attack with Backdoor Enhanced Safety Alignment

22. AgentPoison: Red-teaming LLM Agents via Poisoning Memory or Knowledge Bases

23. Trojan Activation Attack: Red-Teaming Large Language Models using Activation Steering for Safety-Alignment

24. Scaling Laws for Data Poisoning in LLMs

25. BACKDOORLLM: A Comprehensive Benchmark for Backdoor Attacks on Large Language Models

26. Simple Probes can catch sleeper agents

27. Architectural Backdoors in Neural Networks

28. On the Limitation of Backdoor Detection Methods

29. Game of Trojans: Adaptive Adversaries Against Output-based Trojaned-Model Detectors

30. Mitigating Fine-tuning Jailbreak Attack with Backdoor Enhanced Alignment

31. Architectural Neural Backdoors from First Principles

32. ImpNet: Imperceptible and blackbox-undetectable backdoors in compiled neural networks

33. Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

34. Physical Adversarial Attack meets Computer Vision: A Decade Survey

35. Data Poisoning Attacks Against Multimodal Encoders

36. MARNet: Backdoor Attacks Against Cooperative Multi-Agent Reinforcement Learning

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

38. Evil vs evil: using adversarial examples against backdoor attack in federated learning

39. Auditing Visualizations: Transparency Methods Struggle to Detect Anomalous Behavior

40. Defending Backdoor Attacks on Vision Transformer via Patch Processing

41. Defense against backdoor attack in federated learning

42. SentMod: Hidden Backdoor Attack on Unstructured Textual Data

43. Adversarial poisoning attacks on reinforcement learning-driven energy pricing

44. Natural Backdoor Datasets

45. Backdoor Attacks and Defenses in Federated Learning: State-of-the-art, Taxonomy, and Future Directions

46. VulnerGAN: a backdoor attack through vulnerability amplification against machine learning-based network intrusion detection systems

47. Hiding Needles in a Haystack: Towards Constructing Neural Networks that Evade Verification

48. TrojanZoo: Towards Unified, Holistic, and Practical Evaluation of Neural Backdoors

49. Camouflaged Poisoning Attack on Graph Neural Networks

50. BackdoorBench: A Comprehensive Benchmark of Backdoor Learning

51. Fooling a Face Recognition System with a Marker-Free Label-Consistent Backdoor Attack

52. Backdoor Attacks on Bayesian Neural Networks using Reverse Distribution

53. Design of AI Trojans for Evading Machine Learning-based Detection of Hardware Trojans

54. PoisonedEncoder: Poisoning the Unlabeled Pre-training Data in Contrastive Learning

55. Model-Contrastive Learning for Backdoor Defense

56. Robust Anomaly based Attack Detection in Smart Grids under Data Poisoning Attacks

57. Disguised as Privacy: Data Poisoning Attacks against Differentially Private Crowdsensing Systems

58. Poisoning attack toward visual classification model

59. Verifying Neural Networks Against Backdoor Attacks

60. VPN: Verification of Poisoning in Neural Networks

61. LinkBreaker: Breaking the Backdoor-Trigger Link in DNNs via Neurons Consistency Check

62. A Study of the Attention Abnormality in Trojaned BERTs

63. Universal Post-Training Backdoor Detection

64. Planting Undetectable Backdoors in Machine Learning Models

65. Natural Backdoor Attacks on Deep Neural Networks via Raindrops

66. MPAF: Model Poisoning Attacks to Federated Learning based on Fake Clients

67. PiDAn: A Coherence Optimization Approach for Backdoor Attack Detection and Mitigation in Deep Neural Networks

68. ADFL: A Poisoning Attack Defense Framework for Horizontal Federated Learning

69. Toward Realistic Backdoor Injection Attacks on DNNs using Rowhammer

70. Execute Order 66: Targeted Data Poisoning for Reinforcement Learning via Minuscule Perturbations

71. A Feature Based On-Line Detector to Remove Adversarial-Backdoors by Iterative Demarcation

72. BlindNet backdoor: Attack on deep neural network using blind watermark

73. DBIA: Data-free Backdoor Injection Attack against Transformer Networks

74. Backdoor Attack through Frequency Domain

75. NTD: Non-Transferability Enabled Backdoor Detection

76. Romoa: Robust Model Aggregation for the Resistance of Federated Learning to Model Poisoning Attacks

77. Generative strategy based backdoor attacks to 3D point clouds: Work in Progress

78. Deep Neural Backdoor in Semi-Supervised Learning: Threats and Countermeasures

79. FooBaR: Fault Fooling Backdoor Attack on Neural Network Training

80. BFClass: A Backdoor-free Text Classification Framework

81. Backdoor Attacks on Federated Learning with Lottery Ticket Hypothesis

82. Data Poisoning against Differentially-Private Learners: Attacks and Defenses

83. DOES DIFFERENTIAL PRIVACY DEFEAT DATA POISONING?

84. Check Your Other Door! Establishing Backdoor Attacks in the Frequency Domain

85. HaS-Nets: A Heal and Select Mechanism to Defend DNNs Against Backdoor Attacks for Data Collection Scenarios

86. SanitAIs: Unsupervised Data Augmentation to Sanitize Trojaned Neural Networks

87. COVID-19 Diagnosis from Chest X-Ray Images Using Convolutional Neural Networks and Effects of Data Poisoning

88. Interpretability-Guided Defense against Backdoor Attacks to Deep Neural Networks

89. Trojan Signatures in DNN Weights

90. HOW TO INJECT BACKDOORS WITH BETTER CONSISTENCY: LOGIT ANCHORING ON CLEAN DATA

91. A Synergetic Attack against Neural Network Classifiers combining Backdoor and Adversarial Examples

92. Backdoor Attack and Defense for Deep Regression

93. Use Procedural Noise to Achieve Backdoor Attack

94. Excess Capacity and Backdoor Poisoning

95. BatFL: Backdoor Detection on Federated Learning in e-Health

96. Poisonous Label Attack: Black-Box Data Poisoning Attack with Enhanced Conditional DCGAN

97. Backdoor Attacks on Network Certification via Data Poisoning

98. Identifying Physically Realizable Triggers for Backdoored Face Recognition Networks

99. Simtrojan: Stealthy Backdoor Attack

100. Back to the Drawing Board: A Critical Evaluation of Poisoning Attacks on Federated Learning

101. Quantization Backdoors to Deep Learning Models

102. Multi-Target Invisibly Trojaned Networks for Visual Recognition and Detection

103. A Countermeasure Method Using Poisonous Data Against Poisoning Attacks on IoT Machine Learning

104. FederatedReverse: A Detection and Defense Method Against Backdoor Attacks in Federated Learning

105. Accumulative Poisoning Attacks on Real-time Data

106. Inaudible Manipulation of Voice-Enabled Devices Through BackDoor Using Robust Adversarial Audio Attacks

107. Stealthy Targeted Data Poisoning Attack on Knowledge Graphs

108. BinarizedAttack: Structural Poisoning Attacks to Graph-based Anomaly Detection

109. On the Effectiveness of Poisoning against Unsupervised Domain Adaptation

110. Simple, Attack-Agnostic Defense Against Targeted Training Set Attacks Using Cosine Similarity

111. Data Poisoning Attacks Against Outcome Interpretations of Predictive Models

112. BDDR: An Effective Defense Against Textual Backdoor Attacks

113. Poisoning attacks and countermeasures in intelligent networks: status quo and prospects

114. The Devil is in the GAN: Defending Deep Generative Models Against Backdoor Attacks

115. BadEncoder: Backdoor Attacks to Pre-trainedEncoders in Self-Supervised Learning

116. BadEncoder: Backdoor Attacks to Pre-trained Encoders in Self-Supervised Learning

117. Can You Hear It? Backdoor Attacks via Ultrasonic Triggers

118. Poisoning Attacks via Generative Adversarial Text to Image Synthesis

119. Ant Hole: Data Poisoning Attack Breaking out the Boundary of Face Cluster

120. Poison Ink: Robust and Invisible Backdoor Attack

121. MT-MTD: Muti-Training based Moving Target Defense Trojaning Attack in Edged-AI network

122. Text Backdoor Detection Using An Interpretable RNN Abstract Model

123. Garbage in, Garbage out: Poisoning Attacks Disguised with Plausible Mobility in Data Aggregation

124. Classification Auto-Encoder based Detector against Diverse Data Poisoning Attacks

125. Poisoning Knowledge Graph Embeddings via Relation Inference Patterns

126. Adversarial Training Time Attack Against Discriminative and Generative Convolutional Models

127. Poisoning of Online Learning Filters: DDoS Attacks and Countermeasures

128. Rethinking Stealthiness of Backdoor Attack against NLP Models

129. Robust Learning for Data Poisoning Attacks

130. SPECTRE: Defending Against Backdoor Attacks Using Robust Statistics

131. Poisoning the Search Space in Neural Architecture Search

132. Data Poisoning Won’t Save You From Facial Recognition

133. Accumulative Poisoning Attacks on Real-time Data

134. Backdoor Attack on Machine Learning Based Android Malware Detectors

135. Understanding the Limits of Unsupervised Domain Adaptation via Data Poisoning

136. Indirect Invisible Poisoning Attacks on Domain Adaptation

137. Fight Fire with Fire: Towards Robust Recommender Systems via Adversarial Poisoning Training

138. Putting words into the system’s mouth: A targeted attack on neural machine translation using monolingual data poisoning

139. SUBNET REPLACEMENT: DEPLOYMENT-STAGE BACKDOOR ATTACK AGAINST DEEP NEURAL NETWORKS IN GRAY-BOX SETTING

140. Spinning Sequence-to-Sequence Models with Meta-Backdoors

141. Sleeper Agent: Scalable Hidden Trigger Backdoors for Neural Networks Trained from Scratch

142. Poisoning and Backdooring Contrastive Learning

143. AdvDoor: Adversarial Backdoor Attack of Deep Learning System

144. Defending against Backdoor Attacks in Natural Language Generation

145. De-Pois: An Attack-Agnostic Defense against Data Poisoning Attacks

146. Poisoning MorphNet for Clean-Label Backdoor Attack to Point Clouds

147. Provable Guarantees against Data Poisoning Using Self-Expansion and Compatibility

148. MLDS: A Dataset for Weight-Space Analysis of Neural Networks

149. Poisoning the Unlabeled Dataset of Semi-Supervised Learning

150. Regularization Can Help Mitigate Poisioning Attacks. . . With The Right Hyperparameters

151. Witches' Brew: Industrial Scale Data Poisoning via Gradient Matching

152. Towards Robustness Against Natural Language Word Substitutions

153. Concealed Data Poisoning Attacks on NLP Models

154. Covert Channel Attack to Federated Learning Systems

155. Backdoor Attacks Against Deep Learning Systems in the Physical World

156. Backdoor Attacks on Self-Supervised Learning

157. Transferable Environment Poisoning: Training-time Attack on Reinforcement Learning

158. Investigation of a differential cryptanalysis inspired approach for Trojan AI detection

159. Explanation-Guided Backdoor Poisoning Attacks Against Malware Classifiers

160. Robust Backdoor Attacks against Deep Neural Networks in Real Physical World

161. The Design and Development of a Game to Study Backdoor Poisoning Attacks: The Backdoor Game

162. A Backdoor Attack against 3D Point Cloud Classifiers

163. Explainability-based Backdoor Attacks Against Graph Neural Networks

164. DeepSweep: An Evaluation Framework for Mitigating DNN Backdoor Attacks using Data Augmentation

165. Rethinking the Backdoor Attacks' Triggers: A Frequency Perspective

166. PointBA: Towards Backdoor Attacks in 3D Point Cloud

167. Online Defense of Trojaned Models using Misattributions

168. Be Careful about Poisoned Word Embeddings: Exploring the Vulnerability of the Embedding Layers in NLP Models

169. SPECTRE: Defending Against Backdoor Attacks Using Robust Covariance Estimation

170. Black-box Detection of Backdoor Attacks with Limited Information and Data

171. TOP: Backdoor Detection in Neural Networks via Transferability of Perturbation

172. T-Miner : A Generative Approach to Defend Against Trojan Attacks on DNN-based Text Classification

173. Hidden Backdoor Attack against Semantic Segmentation Models

174. What Doesn't Kill You Makes You Robust(er): Adversarial Training against Poisons and Backdoors

175. Red Alarm for Pre-trained Models: Universal Vulnerabilities by Neuron-Level Backdoor Attacks

176. Provable Defense Against Delusive Poisoning

177. An Approach for Poisoning Attacks Against RNN-Based Cyber Anomaly Detection

178. Backdoor Scanning for Deep Neural Networks through K-Arm Optimization

179. TAD: Trigger Approximation based Black-box Trojan Detection for AI*

180. WaNet - Imperceptible Warping-based Backdoor Attack

181. Data Poisoning Attack on Deep Neural Network and Some Defense Methods

182. Baseline Pruning-Based Approach to Trojan Detection in Neural Networks*

183. Covert Model Poisoning Against Federated Learning: Algorithm Design and Optimization

184. Property Inference from Poisoning

185. TROJANZOO: Everything you ever wanted to know about neural backdoors (but were afraid to ask)

186. A Master Key Backdoor for Universal Impersonation Attack against DNN-based Face Verification

187. Detecting Universal Trigger's Adversarial Attack with Honeypot

188. ONION: A Simple and Effective Defense Against Textual Backdoor Attacks

189. Neural Attention Distillation: Erasing Backdoor Triggers from Deep Neural Networks

190. Data Poisoning Attacks to Deep Learning Based Recommender Systems

191. Backdoors hidden in facial features: a novel invisible backdoor attack against face recognition systems

192. One-to-N & N-to-One: Two Advanced Backdoor Attacks against Deep Learning Models

193. DeepPoison: Feature Transfer Based Stealthy Poisoning Attack

194. Policy Teaching via Environment Poisoning:Training-time Adversarial Attacks against Reinforcement Learning

195. Composite Backdoor Attack for Deep Neural Network by Mixing Existing Benign Features

196. SPA: Stealthy Poisoning Attack

197. Backdoor Attack with Sample-Specific Triggers

198. Explainability Matters: Backdoor Attacks on Medical Imaging

199. Escaping Backdoor Attack Detection of Deep Learning

200. Just How Toxic is Data Poisoning? A Unified Benchmark for Backdoor and Data Poisoning Attacks

201. Poisoning Attacks on Cyber Attack Detectors for Industrial Control Systems

202. Fair Detection of Poisoning Attacks in Federated Learning

203. Deep Feature Space Trojan Attack of Neural Networks by Controlled Detoxification*

204. Stealthy Poisoning Attack on Certified Robustness

205. Machine Learning with Electronic Health Records is vulnerable to Backdoor Trigger Attacks

206. Data Security for Machine Learning: Data Poisoning, Backdoor Attacks, and Defenses

207. Detection of Backdoors in Trained Classifiers Without Access to the Training Set

208. TROJANZOO: Everything you ever wanted to know about neural backdoors(but were afraid to ask)

209. HaS-Nets: A Heal and Select Mechanism to Defend DNNs Against Backdoor Attacks for Data Collection Scenarios

210. DeepSweep: An Evaluation Framework for Mitigating DNN Backdoor Attacks using Data Augmentation

211. Poison Attacks against Text Datasets with Conditional Adversarially Regularized Autoencoder

212. Strong Data Augmentation Sanitizes Poisoning and Backdoor Attacks Without an Accuracy Tradeoff

213. BaFFLe: Backdoor detection via Feedback-based Federated Learning

214. Detecting Backdoors in Neural Networks Using Novel Feature-Based Anomaly Detection

215. Mitigating Backdoor Attacks in Federated Learning

216. FaceHack: Triggering backdoored facial recognition systems using facial characteristics

217. Customizing Triggers with Concealed Data Poisoning

218. Backdoor Learning: A Survey

219. Rethinking the Trigger of Backdoor Attack

220. AEGIS: Exposing Backdoors in Robust Machine Learning Models

221. Weight Poisoning Attacks on Pre-trained Models

222. Poisoned classifiers are not only backdoored, they are fundamentally broken

223. Input-Aware Dynamic Backdoor Attack

224. Reverse Engineering Imperceptible Backdoor Attacks on Deep Neural Networks for Detection and Training Set Cleansing

225. BAAAN: Backdoor Attacks Against Autoencoder and GAN-Based Machine Learning Models

226. Don’t Trigger Me! A Triggerless Backdoor Attack Against Deep Neural Networks

227. Toward Robustness and Privacy in Federated Learning: Experimenting with Local and Central Differential Privacy

228. CLEANN: Accelerated Trojan Shield for Embedded Neural Networks

229. Witches’ Brew: Industrial Scale Data Poisoning via Gradient Matching

230. Intrinsic Certified Robustness of Bagging against Data Poisoning Attacks

231. Can Adversarial Weight Perturbations Inject Neural Backdoors?

232. Trojaning Language Models for Fun and Profit

233. Practical Detection of Trojan Neural Networks: Data-Limited and Data-Free Cases

234. Class-Oriented Poisoning Attack

235. Noise-response Analysis for Rapid Detection of Backdoors in Deep Neural Networks

236. Cassandra: Detecting Trojaned Networks from Adversarial Perturbations

237. Backdoor Learning: A Survey

238. Backdoor Attacks and Countermeasures on Deep Learning: A Comprehensive Review

239. Live Trojan Attacks on Deep Neural Networks

240. Odyssey: Creation, Analysis and Detection of Trojan Models

241. Data Poisoning Attacks Against Federated Learning Systems

242. Blind Backdoors in Deep Learning Models

243. Deep Learning Backdoors

244. Attack of the Tails: Yes, You Really Can Backdoor Federated Learning

245. Backdoor Attacks on Facial Recognition in the Physical World

246. Graph Backdoor

247. Backdoor Attacks to Graph Neural Networks

248. You Autocomplete Me: Poisoning Vulnerabilities in Neural Code Completion

249. Reflection Backdoor: A Natural Backdoor Attack on Deep Neural Networks

250. Trembling triggers: exploring the sensitivity of backdoors in DNN-based face recognition

251. Just How Toxic is Data Poisoning? A Unified Benchmark for Backdoor and Data Poisoning Attacks

252. Adversarial Machine Learning -- Industry Perspectives

253. ConFoc: Content-Focus Protection Against Trojan Attacks on Neural Networks

254. Model-Targeted Poisoning Attacks: Provable Convergence and Certified Bounds

255. Deep Partition Aggregation: Provable Defense against General Poisoning Attacks

256. The TrojAI Software Framework: An OpenSource tool for Embedding Trojans into Deep Learning Models*

257. Influence Function based Data Poisoning Attacks to Top-N Recommender Systems

258. BadNL: Backdoor Attacks Against NLP Models

     Summary

     • Introduces first example of backdoor attacks against NLP models using Char-level, Word-level, and Sentence-level triggers (these different triggers operate on the level of their descriptor)
       • Word-level trigger picks a word from the target model’s dictionary and uses it as a trigger
       • Char-level trigger uses insertion, deletion or replacement to modify a single character in a chosen word’s location (with respect to the sentence, for instance, at the start of each sentence) as the trigger.
       • Sentence-level trigger changes the grammar of the sentence and use this as the trigger
     • Authors impose an additional constraint that requires inserted triggers to not change the sentiment of text input
     • Proposed backdoor attack achieves 100% backdoor accuracy with only a drop of 0.18%, 1.26%, and 0.19% in the models utility, for the IMDB, Amazon, and Stanford Sentiment Treebank datasets

259. Neural Network Calculator for Designing Trojan Detectors*

260. Dynamic Backdoor Attacks Against Machine Learning Models

261. Vulnerabilities of Connectionist AI Applications: Evaluation and Defence

262. Backdoor Attacks on Federated Meta-Learning

263. Defending Support Vector Machines against Poisoning Attacks: the Hardness and Algorithm

264. Backdoors in Neural Models of Source Code

265. A new measure for overfitting and its implications for backdooring of deep learning

266. An Embarrassingly Simple Approach for Trojan Attack in Deep Neural Networks

267. MetaPoison: Practical General-purpose Clean-label Data Poisoning

268. Backdooring and Poisoning Neural Networks with Image-Scaling Attacks

269. Bullseye Polytope: A Scalable Clean-Label Poisoning Attack with Improved Transferability

270. On the Effectiveness of Mitigating Data Poisoning Attacks with Gradient Shaping

271. A Survey on Neural Trojans

272. STRIP: A Defence Against Trojan Attacks on Deep Neural Networks

     Summary

     • Authors introduce a run-time based trojan detection system called STRIP or STRong Intentional Pertubation which focuses on models in computer vision
     • STRIP works by intentionally perturbing incoming inputs (ie. by image blending) and then measuring entropy to determine whether the model is trojaned or not. Low entropy violates the input-dependance assumption for a clean model and thus indicates corruption
     • Authors validate STRIPs efficacy on MNIST,CIFAR10, and GTSRB acheiveing false acceptance rates of below 1%

273. TrojDRL: Trojan Attacks on Deep Reinforcement Learning Agents

274. Demon in the Variant: Statistical Analysis of DNNs for Robust Backdoor Contamination Detection

275. Regula Sub-rosa: Latent Backdoor Attacks on Deep Neural Networks

276. Februus: Input Purification Defense Against Trojan Attacks on Deep Neural Network Systems

277. TBT: Targeted Neural Network Attack with Bit Trojan

278. Bypassing Backdoor Detection Algorithms in Deep Learning

279. A backdoor attack against LSTM-based text classification systems

280. Invisible Backdoor Attacks Against Deep Neural Networks

281. Detecting AI Trojans Using Meta Neural Analysis

282. Label-Consistent Backdoor Attacks

283. Detection of Backdoors in Trained Classifiers Without Access to the Training Set

284. ABS: Scanning neural networks for back-doors by artificial brain stimulation

285. NeuronInspect: Detecting Backdoors in Neural Networks via Output Explanations

286. Universal Litmus Patterns: Revealing Backdoor Attacks in CNNs

287. Programmable Neural Network Trojan for Pre-Trained Feature Extractor

288. Demon in the Variant: Statistical Analysis of DNNs for Robust Backdoor Contamination Detection

289. TamperNN: Efficient Tampering Detection of Deployed Neural Nets

290. TABOR: A Highly Accurate Approach to Inspecting and Restoring Trojan Backdoors in AI Systems

291. Design of intentional backdoors in sequential models

292. Design and Evaluation of a Multi-Domain Trojan Detection Method on ins Neural Networks

293. Poison as a Cure: Detecting & Neutralizing Variable-Sized Backdoor Attacks in Deep Neural Networks

294. Data Poisoning Attacks on Stochastic Bandits

295. Hidden Trigger Backdoor Attacks

296. Deep Poisoning Functions: Towards Robust Privacy-safe Image Data Sharing

297. A new Backdoor Attack in CNNs by training set corruption without label poisoning

298. Deep k-NN Defense against Clean-label Data Poisoning Attacks

299. Transferable Clean-Label Poisoning Attacks on Deep Neural Nets

300. Revealing Backdoors, Post-Training, in DNN Classifiers via Novel Inference on Optimized Perturbations Inducing Group Misclassification

301. Explaining Vulnerabilities to Adversarial Machine Learning through Visual Analytics

302. Subpopulation Data Poisoning Attacks

303. TensorClog: An imperceptible poisoning attack on deep neural network applications

304. DeepInspect: A black-box trojan detection and mitigation framework for deep neural networks

305. Resilience of Pruned Neural Network Against Poisoning Attack

306. Spectrum Data Poisoning with Adversarial Deep Learning

307. Neural cleanse: Identifying and mitigating backdoor attacks in neural networks

308. SentiNet: Detecting Localized Universal Attacks Against Deep Learning Systems

     Summary

     • Authors develop SentiNet detection framework for locating universal attacks on neural networks
     • SentiNet is ambivalent to the attack vectors and uses model visualization / object detection techniques to extract potential attacks regions from the models input images. The potential attacks regions are identified as being the parts that influence the prediction the most. After extraction, SentiNet applies these regions to benign inputs and uses the original model to analyze the output
     • Authors stress test the SentiNet framework on three different types of attacks— data poisoning attacks, Trojan attacks, and adversarial patches. They are able to show that the framework achieves competitive metrics across all of the attacks (average true positive rate of 96.22% and an average true negative rate of 95.36%)

309. PoTrojan: powerful neural-level trojan designs in deep learning models

310. Hardware Trojan Attacks on Neural Networks

311. Spectral Signatures in Backdoor Attacks

     Summary

     • Identified a "spectral signatures" property of current backdoor attacks which allows the authors to use robust statistics to stop Trojan attacks
     • The "spectral signature" refers to a change in the covariance spectrum of learned feature representations that is left after a network is attacked. This can be detected by using singular value decomposition (SVD). SVD is used to identify which examples to remove from the training set. After these examples are removed the model is retrained on the cleaned dataset and is no longer Trojaned. The authors test this method on the CIFAR 10 image dataset.

312. Defending Neural Backdoors via Generative Distribution Modeling

313. Detecting Backdoor Attacks on Deep Neural Networks by Activation Clustering

     Summary

     • Proposes Activation Clustering approach to backdoor detection/ removal which analyzes the neural network activations for anomalies and works for both text and images
     • Activation Clustering uses dimensionality techniques (ICA, PCA) on the activations and then clusters them using k-means (k=2) along with a silhouette score metric to separate poisoned from clean clusters
     • Shows that Activation Clustering is successful on three different image/datasets (MNIST, LISA, Rotten Tomatoes) as well as in settings where multiple Trojans are inserted and classes are multi-modal

314. Model-Reuse Attacks on Deep Learning Systems

315. How To Backdoor Federated Learning

316. Trojaning Attack on Neural Networks

317. Poison Frogs! Targeted Clean-Label Poisoning Attacks on Neural Networks

     Summary

     • Proposes neural network poisoning attack that uses "clean labels" which do not require the adversary to mislabel training inputs
     • The paper also presents a optimization based method for generating their poisoning attacks and provides a watermarking strategy for end-to-end attacks that improves the poisoning reliability
     • Authors demonstrate their method by using generated poisoned frog images from the CIFAR dataset to manipulate different kinds of image classifiers

318. Fine-Pruning: Defending Against Backdooring Attacks on Deep Neural Networks

     Summary

     • Investigate two potential detection methods for backdoor attacks (Fine-tuning and pruning). They find both are insufficient on their own and thus propose a combined detection method which they call "Fine-Pruning"
     • Authors go on to show that on three backdoor techniques "Fine-Pruning" is able to eliminate or reduce Trojans on datasets in the traffic sign, speech, and face recognition domains

319. Technical Report: When Does Machine Learning FAIL? Generalized Transferability for Evasion and Poisoning Attacks

320. Backdoor Embedding in Convolutional Neural Network Models via Invisible Perturbation

321. Hu-Fu: Hardware and Software Collaborative Attack Framework against Neural Networks

322. Attack Strength vs. Detectability Dilemma in Adversarial Machine Learning

323. Data Poisoning Attacks in Contextual Bandits

324. BEBP: An Poisoning Method Against Machine Learning Based IDSs

325. Generative Poisoning Attack Method Against Neural Networks

326. BadNets: Identifying Vulnerabilities in the Machine Learning Model Supply Chain

     Summary

     • Introduce Trojan Attacks— a type of attack where an adversary can create a maliciously trained network (a backdoored neural network, or a BadNet) that has state-of-the-art performance on the user’s training and validation samples, but behaves badly on specific attacker-chosen inputs
     • Demonstrate backdoors in a more realistic scenario by creating a U.S. street sign classifier that identifies stop signs as speed limits when a special sticker is added to the stop sign

327. Towards Poisoning of Deep Learning Algorithms with Back-gradient Optimization

328. Targeted Backdoor Attacks on Deep Learning Systems Using Data Poisoning

329. Neural Trojans

330. Towards Poisoning of Deep Learning Algorithms with Back-gradient Optimization

331. Certified defenses for data poisoning attacks

332. Data Poisoning Attacks on Factorization-Based Collaborative Filtering

333. Data poisoning attacks against autoregressive models

334. Using machine teaching to identify optimal training-set attacks on machine learners

335. Poisoning Attacks against Support Vector Machines

336. Backdoor Attacks against Learning Systems

337. Targeted Backdoor Attacks on Deep Learning Systems Using Data Poisoning

338. Antidote: Understanding and defending against poisoning of anomaly detectors