skip to main content

Title: DeepHammer: Depleting the Intelligence of Deep Neural Networks through Targeted Chain of Bit Flips
Security of machine learning is increasingly becoming a major concern due to the ubiquitous deployment of deep learning in many security-sensitive domains. Many prior studies have shown external attacks such as adversarial examples that tamper the integrity of DNNs using maliciously crafted inputs. However, the security implication of internal threats (i.e., hardware vulnerabilities) to DNN models has not yet been well understood. In this paper, we demonstrate the first hardware-based attack on quantized deep neural networks–DeepHammer–that deterministically induces bit flips in model weights to compromise DNN inference by exploiting the rowhammer vulnerability. DeepHammer performs an aggressive bit search in the DNN model to identify the most vulnerable weight bits that are flippable under system constraints. To trigger deterministic bit flips across multiple pages within a reasonable amount of time, we develop novel system-level techniques that enable fast deployment of victim pages, memory-efficient rowhammering and precise flipping of targeted bits. DeepHammer can deliberately degrade the inference accuracy of the victim DNN system to a level that is only as good as random guess, thus completely depleting the intelligence of targeted DNN systems. We systematically demonstrate our attacks on real systems against 11 DNN architectures with 4 datasets corresponding to different application more » domains. Our evaluation shows that DeepHammer is able to successfully tamper DNN inference behavior at run-time within a few minutes. We further discuss several mitigation techniques from both algorithm and system levels to protect DNNs against such attacks. Our work highlights the need to incorporate security mechanisms in future deep learning systems to enhance the robustness against hardware-based deterministic fault injections. « less
; ;
Award ID(s):
1931871 2019548
Publication Date:
Journal Name:
29th {USENIX} Security Symposium ({USENIX} Security 20)
Page Range or eLocation-ID:
Sponsoring Org:
National Science Foundation
More Like this
  1. Security of modern Deep Neural Networks (DNNs) is under severe scrutiny as the deployment of these models become widespread in many intelligence-based applications. Most recently, DNNs are attacked through Trojan which can effectively infect the model during the training phase and get activated only through specific input patterns (i.e, trigger) during inference. In this work, for the first time, we propose a novel Targeted Bit Trojan(TBT) method, which can insert a targeted neural Trojan into a DNN through bit-flip attack. Our algorithm efficiently generates a trigger specifically designed to locate certain vulnerable bits of DNN weights stored in main memory (i.e., DRAM). The objective is that once the attacker flips these vulnerable bits, the network still operates with normal inference accuracy with benign input. However, when the attacker activates the trigger by embedding it with any input, the network is forced to classify all inputs to a certain target class. We demonstrate that flipping only several vulnerable bits identified by our method, using available bit-flip techniques (i.e, row-hammer), can transform a fully functional DNN model into a Trojan-infected model. We perform extensive experiments of CIFAR-10, SVHN and ImageNet datasets on both VGG-16 and Resnet-18 architectures. Our proposed TBT could classifymore »92 of test images to a target class with as little as 84 bit-flips out of 88 million weight bits on Resnet-18 for CIFAR10 dataset.« less
  2. We propose AccHashtag, the first framework for high-accuracy detection of fault-injection attacks on Deep Neural Networks (DNNs) with provable bounds on detection performance. Recent literature in fault-injection attacks shows the severe DNN accuracy degradation caused by bit flips. In this scenario, the attacker changes a few DNN weight bits during execution by injecting faults to the dynamic random-access memory (DRAM). To detect bit flips, AccHashtag extracts a unique signature from the benign DNN prior to deployment. The signature is used to validate the model’s integrity and verify the inference output on the fly. We propose a novel sensitivity analysis that identifies the most vulnerable DNN layers to the fault-injection attack. The DNN signature is constructed by encoding the weights in vulnerable layers using a low-collision hash function. During DNN inference, new hashes are extracted from the target layers and compared against the ground-truth signatures. AccHashtag incorporates a lightweight methodology that allows for real-time fault detection on embedded platforms. We devise a specialized compute core for AccHashtag on field-programmable gate arrays (FPGAs) to facilitate online hash generation in parallel to DNN execution. Extensive evaluations with the state-of-the-art bit-flip attack on various DNNs demonstrate the competitive advantage of AccHashtag in terms ofmore »both attack detection and execution overhead.« less
  3. Recent advancements in Deep Neural Networks (DNNs) have enabled widespread deployment in multiple security-sensitive domains. The need for resource-intensive training and the use of valuable domain-specific training data have made these models the top intellectual property (IP) for model owners. One of the major threats to DNN privacy is model extraction attacks where adversaries attempt to steal sensitive information in DNN models. In this work, we propose an advanced model extraction framework DeepSteal that steals DNN weights remotely for the first time with the aid of a memory side-channel attack. Our proposed DeepSteal comprises two key stages. Firstly, we develop a new weight bit information extraction method, called HammerLeak, through adopting the rowhammer-based fault technique as the information leakage vector. HammerLeak leverages several novel system-level techniques tailored for DNN applications to enable fast and efficient weight stealing. Secondly, we propose a novel substitute model training algorithm with Mean Clustering weight penalty, which leverages the partial leaked bit information effectively and generates a substitute prototype of the target victim model. We evaluate the proposed model extraction framework on three popular image datasets (e.g., CIFAR-10/100/GTSRB) and four DNN architectures (e.g., ResNet-18/34/Wide-ResNetNGG-11). The extracted substitute model has successfully achieved more than 90% testmore »accuracy on deep residual networks for the CIFAR-10 dataset. Moreover, our extracted substitute model could also generate effective adversarial input samples to fool the victim model. Notably, it achieves similar performance (i.e., ~1-2% test accuracy under attack) as white-box adversarial input attack (e.g., PGD/Trades).« less
  4. Motivated by the rise of quantum computers, existing public-key cryptosystems are expected to be replaced by post-quantum schemes in the next decade in billions of devices. To facilitate the transition, NIST is running a standardization process which is currently in its final Round. Only three digital signature schemes are left in the competition, among which Dilithium and Falcon are the ones based on lattices. Besides security and performance, significant attention has been given to resistance against implementation attacks that target side-channel leakage or fault injection response. Classical fault attacks on signature schemes make use of pairs of faulty and correct signatures to recover the secret key which only works on deterministic schemes. To counter such attacks, Dilithium offers a randomized version which makes each signature unique, even when signing identical messages. In this work, we introduce a novel Signature Correction Attack which not only applies to the deterministic version but also to the randomized version of Dilithium and is effective even on constant-time implementations using AVX2 instructions. The Signature Correction Attack exploits the mathematical structure of Dilithium to recover the secret key bits by using faulty signatures and the public-key. It can work for any fault mechanism which can inducemore »single bit-flips. For demonstration, we are using Rowhammer induced faults. Thus, our attack does not require any physical access or special privileges, and hence could be also implemented on shared cloud servers. Using Rowhammer attack, we inject bit flips into the secret key s1 of Dilithium, which results in incorrect signatures being generated by the signing algorithm. Since we can find the correct signature using our Signature Correction algorithm, we can use the difference between the correct and incorrect signatures to infer the location and value of the flipped bit without needing a correct and faulty pair. To quantify the reduction in the security level, we perform a thorough classical and quantum security analysis of Dilithium and successfully recover 1,851 bits out of 3,072 bits of secret key $s_{1}$ for security level 2. Fully recovered bits are used to reduce the dimension of the lattice whereas partially recovered coefficients are used to to reduce the norm of the secret key coefficients. Further analysis for both primal and dual attacks shows that the lattice strength against quantum attackers is reduced from 2128 to 281 while the strength against classical attackers is reduced from 2141 to 289. Hence, the Signature Correction Attack may be employed to achieve a practical attack on Dilithium (security level 2) as proposed in Round 3 of the NIST post-quantum standardization process.« less
  5. Traditional Deep Neural Network (DNN) security is mostly related to the well-known adversarial input example attack.Recently, another dimension of adversarial attack, namely, attack on DNN weight parameters, has been shown to be very powerful. Asa representative one, the Bit-Flip based adversarial weight Attack (BFA) injects an extremely small amount of faults into weight parameters to hijack the executing DNN function. Prior works of BFA focus on un-targeted attacks that can hack all inputs into a random output class by flipping a very small number of weight bits stored in computer memory. This paper proposes the first work oftargetedBFA based (T-BFA) adversarial weight attack on DNNs, which can intentionally mislead selected inputs to a target output class. The objective is achieved by identifying the weight bits that are highly associated with classification of a targeted output through a class-dependent weight bit searching algorithm. Our proposed T-BFA performance is successfully demonstrated on multiple DNN architectures for image classification tasks. For example, by merely flipping 27 out of 88 million weight bits of ResNet-18, our T-BFA can misclassify all the images from Hen class into Goose class (i.e., 100% attack success rate) in ImageNet dataset, while maintaining 59.35% validation accuracy.