mlsanityData to 7 Oct 2026

Terminal Brain Damage: Exposing the Graceless Degradation in Deep Neural Networks Under Hardware Fault Attacks

arXiv 1906.01017 ↗First posted 3 Jun 2019cs.CR cs.LGcited by 55

Not on the map. Why?

Abstract (arXiv)

Deep neural networks (DNNs) have been shown to tolerate "brain damage": cumulative changes to the network's parameters (e.g., pruning, numerical perturbations) typically result in a graceful degradation of classification accuracy. However, the limits of this natural resilience are not well understood in the presence of small adversarial changes to the DNN parameters' underlying memory representation, such as bit-flips that may be induced by hardware fault attacks. We study the effects of bitwise corruptions on 19 DNN models---six architectures on three image classification tasks---and we show that most models have at least one parameter that, after a specific bit-flip in their bitwise representation, causes an accuracy loss of over 90%. We employ simple heuristics to efficiently identify the parameters likely to be vulnerable. We estimate that 40-50% of the parameters in a model might lead to an accuracy drop greater than 10% when individually subjected to such single-bit perturbations. To demonstrate how an adversary could take advantage of this vulnerability, we study the impact of an exemplary hardware fault attack, Rowhammer, on DNNs. Specifically, we show that a Rowhammer enabled attacker co-located in the same physical machine can inflict significant accuracy drops (up to 99%) even with single bit-flip corruptions and no knowledge of the model. Our results expose the limits of DNNs' resilience against parameter perturbations induced by real-world fault attacks. We conclude by discussing possible mitigations and future research directions towards fault attack-resilient DNNs.

Key facts. Cited by 55 papers in the map (data through 7 Oct 2026). First posted on arXiv 3 Jun 2019.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 14; 2022: 5; 2023: 9; 2024: 9; 2025: 8; 2026: 10.

55 papers.

Sort:

By area:

  1. 2021Advances in adversarial attacks and defenses in computer vision: A surveycited by 127
  2. 2021Bad Characters: Imperceptible NLP Attackscited by 69
  3. 2022"Real Attackers Don't Compute Gradients": Bridging the Gap Between Adversarial ML Research and Practicecited by 55
  4. 2021Targeted Attack against Deep Neural Networks via Flipping Limited Weight Bitscited by 49
  5. 2021RADAR: Run-time Adversarial Weight Attack Detection and Accuracy Recoverycited by 18
  6. 2021Towards Energy-Efficient and Secure Edge AI: A Cross-Layer Frameworkcited by 17
  7. 2021Qu-ANTI-zation: Exploiting Quantization Artifacts for Achieving Adversarial Outcomescited by 16
  8. 2022Versatile Weight Attack via Flipping Limited Bitscited by 13
  9. 2021FitAct: Error Resilient Deep Neural Networks via Fine-Grained Post-Trainable Activation Functionscited by 12
  10. 2021RA-BNN: Constructing Robust & Accurate Binary Neural Network to Simultaneously Defend Adversarial Bit-Flip Attack and Improve Accuracycited by 11
  11. 2023DeepAxe: A Framework for Exploration of Approximation and Reliability Trade-offs in DNN Acceleratorscited by 11
  12. 2021HASHTAG: Hash Signatures for Online Detection of Fault-Injection Attacks on Deep Neural Networkscited by 9
38 more
  1. 2021Towards a Safety Case for Hardware Fault Tolerance in Convolutional Neural Networks Using Activation Range Supervisioncited by 7
  2. 2022Hardware faults that matter: Understanding and Estimating the safety impact of hardware faults on object detection DNNscited by 7
  3. 2023Efficient Backdoor Attacks for Deep Neural Networks in Real-world Scenarioscited by 7
  4. 2021Exploring Fault-Energy Trade-offs in Approximate DNN Hardware Acceleratorscited by 6
  5. 2021FooBaR: Fault Fooling Backdoor Attack on Neural Network Trainingcited by 6
  6. 2021Don't Knock! Rowhammer at the Backdoor of DNN Modelscited by 6
  7. 2023A Low-cost Strategic Monitoring Approach for Scalable and Interpretable Error Detection in Deep Neural Networkscited by 6
  8. 2023A Proxy Attack-Free Strategy for Practically Improving the Poisoning Efficiency in Backdoor Attackscited by 5
  9. 2024Data Free Backdoor Attackscited by 5
  10. 2025Maximal Brain Damage Without Data or Optimization: Disrupting Neural Networks via Sign-Bit Flipscited by 4
  11. 2024DeepNcode: Encoding-Based Protection against Bit-Flip Attacks on Neural Networkscited by 3
  12. 2024Designing DNNs for a trade-off between robustness and processing performance in embedded devicescited by 3
  13. 2024PrisonBreak: Jailbreaking Large Language Models with at Most Twenty-Five Targeted Bit-flipscited by 3
  14. 2022Thales: Formulating and Estimating Architectural Vulnerability Factors for DNN Acceleratorscited by 2
  15. 2023DNN-Defender: A Victim-Focused In-DRAM Defense Mechanism for Taming Adversarial Weight Attack on DNNscited by 2
  16. 2023Compiled Models, Built-In Exploits: Uncovering Pervasive Bit-Flip Attack Surfaces in DNN Executablescited by 2
  17. 2024Harden Deep Neural Networks Against Fault Injections Through Weight Scalingcited by 2
  18. 2025Verification of Bit-Flip Attacks against Quantized Neural Networkscited by 2
  19. 2025DoCIA: An Online Document-Level Context Incorporation Agent for Speech Translationcited by 2
  20. 2023An Incremental Gray-box Physical Adversarial Attack on Neural Network Trainingcited by 1
  21. 2024ZOBNN: Zero-Overhead Dependable Design of Binary Neural Networks with Deliberately Quantized Parameterscited by 1
  22. 2024DeepBaR: Fault Backdoor Attack on Deep Neural Network Layerscited by 1
  23. 2025FL-PLAS: Federated Learning with Partial Layer Aggregation for Backdoor Defense Against High-Ratio Malicious Clientscited by 1
  24. 2025Bit-Flip Fault Attack: Crushing Graph Neural Networks via Gradual Bit Searchcited by 1
  25. 2026The Weight of a Bit: EMFI Sensitivity Analysis of Embedded Deep Learning Modelscited by 1
  26. 2021Dynamic Neural Network Architectural and Topological Adaptation and Related Methods -- A Surveycited by 0
  27. 2022FlatENN: Train Flat for Enhanced Fault Tolerance of Quantized Deep Neural Networkscited by 0
  28. 2023Evil from Within: Machine Learning Backdoors through Hardware Trojanscited by 0
  29. 2023Attacking Graph Neural Networks with Bit Flips: Weisfeiler and Lehman Go Indifferentcited by 0
  30. 2024Impactful Bit-Flip Search on Full-precision Modelscited by 0
  31. 2024Enhancing Neural Network Robustness Against Fault Injection Through Non-linear Weight Transformationscited by 0
  32. 2025Hessian-aware Training for Enhancing DNNs Resilience to Parameter Corruptionscited by 0
  33. 2025Threat Modeling for AI: The Case for an Asset-Centric Approachcited by 0
  34. 2025HAMLOCK: HArdware-Model LOgically Combined attacKcited by 0
  35. 2026MAED: Mathematical Activation Error Detection for Mitigating Physical Fault Attacks in DNN Inferencecited by 0
  36. 2026FLARE: One-Shot PE-Level Fault Localization in Systolic Arrays via Algebraic Test Vectorscited by 0
  37. 2026DP-SelFT: Differentially Private Selective Fine-Tuning for Large Language Modelscited by 0
  38. 2026Characterizing the Fault Response of the Intel Neural Compute Stick 2 Under Single-Pulse Electromagnetic Fault Injectioncited by 0

50 most cited of 55 All 55 (JSON)