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. Simple Probes can catch sleeper agents

2. Architectural Backdoors in Neural Networks

3. On the Limitation of Backdoor Detection Methods

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

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

6. Architectural Neural Backdoors from First Principles

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

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

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

10. Data Poisoning Attacks Against Multimodal Encoders

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

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

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

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

15. Defending Backdoor Attacks on Vision Transformer via Patch Processing

16. Defense against backdoor attack in federated learning

17. SentMod: Hidden Backdoor Attack on Unstructured Textual Data

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

19. Natural Backdoor Datasets

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

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

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

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

24. Camouflaged Poisoning Attack on Graph Neural Networks

25. BackdoorBench: A Comprehensive Benchmark of Backdoor Learning

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

27. Backdoor Attacks on Bayesian Neural Networks using Reverse Distribution

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

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

30. Model-Contrastive Learning for Backdoor Defense

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

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

33. Poisoning attack toward visual classification model

34. Verifying Neural Networks Against Backdoor Attacks

35. VPN: Verification of Poisoning in Neural Networks

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

37. A Study of the Attention Abnormality in Trojaned BERTs

38. Universal Post-Training Backdoor Detection

39. Planting Undetectable Backdoors in Machine Learning Models

40. Natural Backdoor Attacks on Deep Neural Networks via Raindrops

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

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

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

44. Toward Realistic Backdoor Injection Attacks on DNNs using Rowhammer

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

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

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

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

49. Backdoor Attack through Frequency Domain

50. NTD: Non-Transferability Enabled Backdoor Detection

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

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

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

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

55. BFClass: A Backdoor-free Text Classification Framework

56. Backdoor Attacks on Federated Learning with Lottery Ticket Hypothesis

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

58. DOES DIFFERENTIAL PRIVACY DEFEAT DATA POISONING?

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

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

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

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

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

64. Trojan Signatures in DNN Weights

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

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

67. Backdoor Attack and Defense for Deep Regression

68. Use Procedural Noise to Achieve Backdoor Attack

69. Excess Capacity and Backdoor Poisoning

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

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

72. Backdoor Attacks on Network Certification via Data Poisoning

73. Identifying Physically Realizable Triggers for Backdoored Face Recognition Networks

74. Simtrojan: Stealthy Backdoor Attack

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

76. Quantization Backdoors to Deep Learning Models

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

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

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

80. Accumulative Poisoning Attacks on Real-time Data

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

82. Stealthy Targeted Data Poisoning Attack on Knowledge Graphs

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

84. On the Effectiveness of Poisoning against Unsupervised Domain Adaptation

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

86. Data Poisoning Attacks Against Outcome Interpretations of Predictive Models

87. BDDR: An Effective Defense Against Textual Backdoor Attacks

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

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

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

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

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

93. Poisoning Attacks via Generative Adversarial Text to Image Synthesis

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

95. Poison Ink: Robust and Invisible Backdoor Attack

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

97. Text Backdoor Detection Using An Interpretable RNN Abstract Model

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

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

100. Poisoning Knowledge Graph Embeddings via Relation Inference Patterns

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

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

103. Rethinking Stealthiness of Backdoor Attack against NLP Models

104. Robust Learning for Data Poisoning Attacks

105. SPECTRE: Defending Against Backdoor Attacks Using Robust Statistics

106. Poisoning the Search Space in Neural Architecture Search

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

108. Accumulative Poisoning Attacks on Real-time Data

109. Backdoor Attack on Machine Learning Based Android Malware Detectors

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

111. Indirect Invisible Poisoning Attacks on Domain Adaptation

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

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

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

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

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

117. Poisoning and Backdooring Contrastive Learning

118. AdvDoor: Adversarial Backdoor Attack of Deep Learning System

119. Defending against Backdoor Attacks in Natural Language Generation

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

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

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

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

124. Poisoning the Unlabeled Dataset of Semi-Supervised Learning

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

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

127. Towards Robustness Against Natural Language Word Substitutions

128. Concealed Data Poisoning Attacks on NLP Models

129. Covert Channel Attack to Federated Learning Systems

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

131. Backdoor Attacks on Self-Supervised Learning

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

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

134. Explanation-Guided Backdoor Poisoning Attacks Against Malware Classifiers

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

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

137. A Backdoor Attack against 3D Point Cloud Classifiers

138. Explainability-based Backdoor Attacks Against Graph Neural Networks

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

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

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

142. Online Defense of Trojaned Models using Misattributions

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

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

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

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

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

148. Hidden Backdoor Attack against Semantic Segmentation Models

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

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

151. Provable Defense Against Delusive Poisoning

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

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

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

155. WaNet - Imperceptible Warping-based Backdoor Attack

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

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

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

159. Property Inference from Poisoning

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

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

162. Detecting Universal Trigger's Adversarial Attack with Honeypot

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

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

165. Data Poisoning Attacks to Deep Learning Based Recommender Systems

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

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

168. DeepPoison: Feature Transfer Based Stealthy Poisoning Attack

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

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

171. SPA: Stealthy Poisoning Attack

172. Backdoor Attack with Sample-Specific Triggers

173. Explainability Matters: Backdoor Attacks on Medical Imaging

174. Escaping Backdoor Attack Detection of Deep Learning

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

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

177. Fair Detection of Poisoning Attacks in Federated Learning

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

179. Stealthy Poisoning Attack on Certified Robustness

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

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

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

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

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

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

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

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

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

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

190. Mitigating Backdoor Attacks in Federated Learning

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

192. Customizing Triggers with Concealed Data Poisoning

193. Backdoor Learning: A Survey

194. Rethinking the Trigger of Backdoor Attack

195. AEGIS: Exposing Backdoors in Robust Machine Learning Models

196. Weight Poisoning Attacks on Pre-trained Models

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

198. Input-Aware Dynamic Backdoor Attack

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

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

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

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

203. CLEANN: Accelerated Trojan Shield for Embedded Neural Networks

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

205. Intrinsic Certified Robustness of Bagging against Data Poisoning Attacks

206. Can Adversarial Weight Perturbations Inject Neural Backdoors?

207. Trojaning Language Models for Fun and Profit

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

209. Class-Oriented Poisoning Attack

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

211. Cassandra: Detecting Trojaned Networks from Adversarial Perturbations

212. Backdoor Learning: A Survey

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

214. Live Trojan Attacks on Deep Neural Networks

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

216. Data Poisoning Attacks Against Federated Learning Systems

217. Blind Backdoors in Deep Learning Models

218. Deep Learning Backdoors

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

220. Backdoor Attacks on Facial Recognition in the Physical World

221. Graph Backdoor

222. Backdoor Attacks to Graph Neural Networks

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

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

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

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

227. Adversarial Machine Learning -- Industry Perspectives

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

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

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

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

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

233. 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

234. Neural Network Calculator for Designing Trojan Detectors*

235. Dynamic Backdoor Attacks Against Machine Learning Models

236. Vulnerabilities of Connectionist AI Applications: Evaluation and Defence

237. Backdoor Attacks on Federated Meta-Learning

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

239. Backdoors in Neural Models of Source Code

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

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

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

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

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

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

246. A Survey on Neural Trojans

247. 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%

248. TrojDRL: Trojan Attacks on Deep Reinforcement Learning Agents

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

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

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

252. TBT: Targeted Neural Network Attack with Bit Trojan

253. Bypassing Backdoor Detection Algorithms in Deep Learning

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

255. Invisible Backdoor Attacks Against Deep Neural Networks

256. Detecting AI Trojans Using Meta Neural Analysis

257. Label-Consistent Backdoor Attacks

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

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

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

261. Universal Litmus Patterns: Revealing Backdoor Attacks in CNNs

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

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

264. TamperNN: Efficient Tampering Detection of Deployed Neural Nets

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

266. Design of intentional backdoors in sequential models

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

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

269. Data Poisoning Attacks on Stochastic Bandits

270. Hidden Trigger Backdoor Attacks

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

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

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

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

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

276. Explaining Vulnerabilities to Adversarial Machine Learning through Visual Analytics

277. Subpopulation Data Poisoning Attacks

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

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

280. Resilience of Pruned Neural Network Against Poisoning Attack

281. Spectrum Data Poisoning with Adversarial Deep Learning

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

283. 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%)

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

285. Hardware Trojan Attacks on Neural Networks

286. 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.

287. Defending Neural Backdoors via Generative Distribution Modeling

288. 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

289. Model-Reuse Attacks on Deep Learning Systems

290. How To Backdoor Federated Learning

291. Trojaning Attack on Neural Networks

292. 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

293. 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

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

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

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

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

298. Data Poisoning Attacks in Contextual Bandits

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

300. Generative Poisoning Attack Method Against Neural Networks

301. 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

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

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

304. Neural Trojans

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

306. Certified defenses for data poisoning attacks

307. Data Poisoning Attacks on Factorization-Based Collaborative Filtering

308. Data poisoning attacks against autoregressive models

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

310. Poisoning Attacks against Support Vector Machines

311. Backdoor Attacks against Learning Systems

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

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