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; details will be provided later.
• A trained ASReview random forest model
• A curated manual literature review

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

2. Data Poisoning Attacks Against Multimodal Encoders

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

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

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

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

7. Defending Backdoor Attacks on Vision Transformer via Patch Processing

8. Defense against backdoor attack in federated learning

9. SentMod: Hidden Backdoor Attack on Unstructured Textual Data

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

11. Natural Backdoor Datasets

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

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

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

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

16. Camouflaged Poisoning Attack on Graph Neural Networks

17. BackdoorBench: A Comprehensive Benchmark of Backdoor Learning

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

19. Backdoor Attacks on Bayesian Neural Networks using Reverse Distribution

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

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

22. Model-Contrastive Learning for Backdoor Defense

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

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

25. Poisoning attack toward visual classification model

26. Verifying Neural Networks Against Backdoor Attacks

27. VPN: Verification of Poisoning in Neural Networks

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

29. A Study of the Attention Abnormality in Trojaned BERTs

30. Universal Post-Training Backdoor Detection

31. Planting Undetectable Backdoors in Machine Learning Models

32. Natural Backdoor Attacks on Deep Neural Networks via Raindrops

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

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

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

36. Toward Realistic Backdoor Injection Attacks on DNNs using Rowhammer

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

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

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

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

41. Backdoor Attack through Frequency Domain

42. NTD: Non-Transferability Enabled Backdoor Detection

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

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

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

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

47. BFClass: A Backdoor-free Text Classification Framework

48. Backdoor Attacks on Federated Learning with Lottery Ticket Hypothesis

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

50. DOES DIFFERENTIAL PRIVACY DEFEAT DATA POISONING?

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

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

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

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

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

56. Trojan Signatures in DNN Weights

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

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

59. Backdoor Attack and Defense for Deep Regression

60. Use Procedural Noise to Achieve Backdoor Attack

61. Excess Capacity and Backdoor Poisoning

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

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

64. Backdoor Attacks on Network Certification via Data Poisoning

65. Identifying Physically Realizable Triggers for Backdoored Face Recognition Networks

66. Simtrojan: Stealthy Backdoor Attack

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

68. Quantization Backdoors to Deep Learning Models

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

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

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

72. Accumulative Poisoning Attacks on Real-time Data

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

74. Stealthy Targeted Data Poisoning Attack on Knowledge Graphs

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

76. On the Effectiveness of Poisoning against Unsupervised Domain Adaptation

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

78. Data Poisoning Attacks Against Outcome Interpretations of Predictive Models

79. BDDR: An Effective Defense Against Textual Backdoor Attacks

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

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

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

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

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

85. Poisoning Attacks via Generative Adversarial Text to Image Synthesis

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

87. Poison Ink: Robust and Invisible Backdoor Attack

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

89. Text Backdoor Detection Using An Interpretable RNN Abstract Model

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

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

92. Poisoning Knowledge Graph Embeddings via Relation Inference Patterns

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

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

95. Rethinking Stealthiness of Backdoor Attack against NLP Models

96. Robust Learning for Data Poisoning Attacks

97. SPECTRE: Defending Against Backdoor Attacks Using Robust Statistics

98. Poisoning the Search Space in Neural Architecture Search

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

100. Accumulative Poisoning Attacks on Real-time Data

101. Backdoor Attack on Machine Learning Based Android Malware Detectors

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

103. Indirect Invisible Poisoning Attacks on Domain Adaptation

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

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

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

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

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

109. Poisoning and Backdooring Contrastive Learning

110. AdvDoor: Adversarial Backdoor Attack of Deep Learning System

111. Defending against Backdoor Attacks in Natural Language Generation

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

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

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

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

116. Poisoning the Unlabeled Dataset of Semi-Supervised Learning

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

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

119. Towards Robustness Against Natural Language Word Substitutions

120. Concealed Data Poisoning Attacks on NLP Models

121. Covert Channel Attack to Federated Learning Systems

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

123. Backdoor Attacks on Self-Supervised Learning

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

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

126. Explanation-Guided Backdoor Poisoning Attacks Against Malware Classifiers

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

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

129. A Backdoor Attack against 3D Point Cloud Classifiers

130. Explainability-based Backdoor Attacks Against Graph Neural Networks

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

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

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

134. Online Defense of Trojaned Models using Misattributions

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

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

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

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

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

140. Hidden Backdoor Attack against Semantic Segmentation Models

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

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

143. Provable Defense Against Delusive Poisoning

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

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

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

147. WaNet - Imperceptible Warping-based Backdoor Attack

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

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

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

151. Property Inference from Poisoning

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

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

154. Detecting Universal Trigger's Adversarial Attack with Honeypot

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

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

157. Data Poisoning Attacks to Deep Learning Based Recommender Systems

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

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

160. DeepPoison: Feature Transfer Based Stealthy Poisoning Attack

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

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

163. SPA: Stealthy Poisoning Attack

164. Backdoor Attack with Sample-Specific Triggers

165. Explainability Matters: Backdoor Attacks on Medical Imaging

166. Escaping Backdoor Attack Detection of Deep Learning

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

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

169. Fair Detection of Poisoning Attacks in Federated Learning

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

171. Stealthy Poisoning Attack on Certified Robustness

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

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

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

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

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

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

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

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

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

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

182. Mitigating Backdoor Attacks in Federated Learning

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

184. Customizing Triggers with Concealed Data Poisoning

185. Backdoor Learning: A Survey

186. Rethinking the Trigger of Backdoor Attack

187. AEGIS: Exposing Backdoors in Robust Machine Learning Models

188. Weight Poisoning Attacks on Pre-trained Models

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

190. Input-Aware Dynamic Backdoor Attack

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

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

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

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

195. CLEANN: Accelerated Trojan Shield for Embedded Neural Networks

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

197. Intrinsic Certified Robustness of Bagging against Data Poisoning Attacks

198. Can Adversarial Weight Perturbations Inject Neural Backdoors?

199. Trojaning Language Models for Fun and Profit

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

201. Class-Oriented Poisoning Attack

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

203. Cassandra: Detecting Trojaned Networks from Adversarial Perturbations

204. Backdoor Learning: A Survey

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

206. Live Trojan Attacks on Deep Neural Networks

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

208. Data Poisoning Attacks Against Federated Learning Systems

209. Blind Backdoors in Deep Learning Models

210. Deep Learning Backdoors

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

212. Backdoor Attacks on Facial Recognition in the Physical World

213. Graph Backdoor

214. Backdoor Attacks to Graph Neural Networks

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

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

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

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

219. Adversarial Machine Learning -- Industry Perspectives

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

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

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

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

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

225. 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

226. Neural Network Calculator for Designing Trojan Detectors*

227. Dynamic Backdoor Attacks Against Machine Learning Models

228. Vulnerabilities of Connectionist AI Applications: Evaluation and Defence

229. Backdoor Attacks on Federated Meta-Learning

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

231. Backdoors in Neural Models of Source Code

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

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

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

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

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

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

238. A Survey on Neural Trojans

239. 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%

240. TrojDRL: Trojan Attacks on Deep Reinforcement Learning Agents

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

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

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

244. TBT: Targeted Neural Network Attack with Bit Trojan

245. Bypassing Backdoor Detection Algorithms in Deep Learning

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

247. Invisible Backdoor Attacks Against Deep Neural Networks

248. Detecting AI Trojans Using Meta Neural Analysis

249. Label-Consistent Backdoor Attacks

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

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

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

253. Universal Litmus Patterns: Revealing Backdoor Attacks in CNNs

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

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

256. TamperNN: Efficient Tampering Detection of Deployed Neural Nets

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

258. Design of intentional backdoors in sequential models

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

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

261. Data Poisoning Attacks on Stochastic Bandits

262. Hidden Trigger Backdoor Attacks

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

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

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

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

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

268. Explaining Vulnerabilities to Adversarial Machine Learning through Visual Analytics

269. Subpopulation Data Poisoning Attacks

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

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

272. Resilience of Pruned Neural Network Against Poisoning Attack

273. Spectrum Data Poisoning with Adversarial Deep Learning

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

275. 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%)

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

277. Hardware Trojan Attacks on Neural Networks

278. 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.

279. Defending Neural Backdoors via Generative Distribution Modeling

280. 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

281. Model-Reuse Attacks on Deep Learning Systems

282. How To Backdoor Federated Learning

283. Trojaning Attack on Neural Networks

284. 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

285. 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

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

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

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

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

290. Data Poisoning Attacks in Contextual Bandits

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

292. Generative Poisoning Attack Method Against Neural Networks

293. 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

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

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

296. Neural Trojans

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

298. Certified defenses for data poisoning attacks

299. Data Poisoning Attacks on Factorization-Based Collaborative Filtering

300. Data poisoning attacks against autoregressive models

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

302. Poisoning Attacks against Support Vector Machines

303. Backdoor Attacks against Learning Systems

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

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