Machine learning systems increasingly rely on open-source artifacts such as datasets and models that are created or hosted by other parties. The reliance on external datasets and pre-trained models exposes the system to supply chain attacks where an artifact can be poisoned before it is delivered to the end-user. Such attacks are possible due to the lack of any authenticity verification in existing machine learning systems. Incorporating cryptographic solutions such as hashing and signing can mitigate the risk of supply chain attacks. However, existing frameworks for integrity verification based on cryptographic techniques can incur significant overhead when applied to state-of-the-art machine learning artifacts due to their scale, and are not compatible with GPU platforms. In this paper, we develop Sentry, a novel GPU-based framework that verifies the authenticity of machine learning artifacts by implementing cryptographic signing and verification for datasets and models. Sentry ties developer identities to signatures and performs authentication on the fly as artifacts are loaded on GPU memory, making it compatible with GPU data movement solutions such as NVIDIA GPUDirect that bypass the CPU. Sentry incorporates GPU acceleration of cryptographic hash constructions such as Merkle tree and lattice hashing, implementing memory optimizations and resource partitioning schemes for a high throughput performance. Our evaluations show that Sentry is a practical solution to bring authenticity to machine learning systems, achieving orders of magnitude speedup over a CPU-based baseline.

Federated learning enables users to collaboratively train a machine learning model over their private datasets. Secure aggregation protocols are employed to mitigate information leakage about the local datasets from user-submitted model updates. This setup, however, still leaks the user participation in training, which can also be sensitive. Protecting user anonymity is even more challenging in dynamic environments where users may (re)join or leave the training process at any point of time. This paper introduces AnoFel, the first framework to support private and anonymous dynamic participation in federated learning (FL). AnoFel leverages several cryptographic primitives, the concept of anonymity sets, differential privacy, and a public bulletin board to support anonymous user registration, as well as unlinkable and confidential model update submission. Our system allows dynamic participation, where users can join or leave at any time without needing any recovery protocol or interaction. To assess security, we formalize a notion for privacy and anonymity in FL, and formally prove that AnoFel satisfies this notion. To the best of our knowledge, our system is the first solution with provable anonymity guarantees. To assess efficiency, we provide a concrete implementation of AnoFel, and conduct experiments showing its ability to support learning applications scaling to a large number of clients. For a TinyImageNet classification task with 512 clients, the client setup to join is less than 3 sec, and the client runtime for each training iteration takes a total of 8 sec, where the added overhead of AnoFel is 46% of the total runtime. We also compare our system with prior work and demonstrate its practicality. AnoFel client runtime is up to 5x faster than Truex et al., despite the added anonymity guarantee and dynamic user joining in AnoFel. Compared to Bonawitz et al., AnoFel is only 2x slower for added support for privacy in output, dynamic user joining, and anonymity.

In two-party machine learning prediction services, the client’s goal is to query a remote server’s trained machine learning model to perform neural network inference in some application domain. However, sensitive information can be obtained during this process by either the client or the server, leading to potential collection, unauthorized secondary use, and inappropriate access to personal information. These security concerns have given rise to Private Inference (PI), in which both the client’s personal data and the server’s trained model are kept confidential. State-of-the-art PI protocols consist of a pre-processing or offline phase and an online phase that combine several cryptographic primitives: Homomorphic Encryption (HE), Secret Sharing (SS), Garbled Circuits (GC), and Oblivious Transfer (OT). Despite the need and recent performance improvements, PI remains largely arcane today and is too slow for practical use.

Training contemporary AI models requires investment in procuring learning data and computing resources, making the models intellectual property of the owners. Popular model watermarking solutions rely on key input triggers for detection; the keys have to be kept private to prevent discovery, forging, and removal of the hidden signatures. We present ZKROWNN, the first automated end-to-end framework utilizing Zero-Knowledge Proofs (ZKP) that enable an entity to validate their ownership of a model, while preserving the privacy of the watermarks. ZKROWNN permits a third party client to verify model ownership in less than a second, requiring as little as a few KBs of communication.

Distributed GNN training on contemporary massive and densely connected graphs requires information aggregation from all neighboring nodes, which leads to an explosion of inter-server communications. This paper proposes AdaGL, a highly scalable end-to-end framework for rapid distributed GNN training. AdaGL novelty lies upon our adaptive-learning based graph-allocation engine as well as utilizing multi-resolution coarse representation of dense graphs. As a result, AdaGL achieves an unprecedented level of balanced server computation while minimizing the communication overhead. Extensive proof-of-concept evaluations on billion-scale graphs show AdaGL attains ∼30−40% faster convergence compared with prior arts.

Privacy-preserving federated learning allows multiple users to jointly train a model with coordination of a central server. The server only learns the final aggregation result, thereby preventing leakage of the users’ (private) training data from the individual model updates. However, keeping the individual updates private allows malicious users to degrade the model accuracy without being detected, also known as Byzantine attacks. Best existing defenses against Byzantine workers rely on robust rank-based statistics, e.g., setting robust bounds via the median of updates, to find malicious updates. However, implementing privacy-preserving rank-based statistics, especially median-based, is nontrivial and unscalable in the secure domain, as it requires sorting of all individual updates. We establish the first private robustness check that uses high break point rank-based statistics on aggregated model updates. By exploiting randomized clustering, we significantly improve the scalability of our defense without compromising privacy. We leverage the derived statistical bounds in zero-knowledge proofs to detect and remove malicious updates without revealing the private user updates. Our novel framework, zPROBE, enables Byzantine resilient and secure federated learning. We show the effectiveness of zPROBE on several computer vision benchmarks. Empirical evaluations demonstrate that zPROBE provides a low overhead solution to defend against state-of-the-art Byzantine attacks while preserving privacy.

The recent rise of privacy concerns has led researchers to devise methods for private neural inference—where inferences are made directly on encrypted data, never seeing inputs. The primary challenge facing private inference is that computing on encrypted data levies an impractically-high latency penalty, stemming mostly from non-linear operators like ReLU. Enabling practical and private inference requires new optimization methods that minimize network ReLU counts while preserving accuracy. This paper proposes DeepReDuce: a set of optimizations for the judicious removal of ReLUs to reduce private inference latency. The key insight is that not all ReLUs contribute equally to accuracy. We leverage this insight to drop, or remove, ReLUs from classic networks to significantly reduce inference latency and maintain high accuracy. Given a network architecture, DeepReDuce outputs a Pareto frontier of networks that tradeoff the number of ReLUs and accuracy. Compared to the state-of-the-art for private inference DeepReDuce improves accuracy and reduces ReLU count by up to 3.5% (iso-ReLU count) and 3.5x (iso-accuracy), respectively.

Extracting interesting scenarios from real-world data as well as generating failure cases is important for the development and testing of autonomous systems. We propose efficient mechanisms to both characterize and generate testing scenarios using a state-of-the-art driving simulator. For any scenario, our method generates a set of possible driving paths and identifies all the possible safe driving trajectories that can be taken starting at different times, to compute metrics that quantify the complexity of the scenario. We use our method to characterize real driving data from the Next Generation Simulation (NGSIM) project, as well as adversarial scenarios generated in simulation. We rank the scenarios by defining metrics based on the complexity of avoiding accidents and provide insights into how the AV could have minimized the probability of incurring an accident. We demonstrate a strong correlation between the proposed metrics and human intuition.

The simultaneous rise of machine learning as a service and concerns over user privacy have increasingly motivated the need for private inference (PI). While recent work demonstrates PI is possible using cryptographic primitives, the computational overheads render it impractical. State-of-art deep networks are inadequate in this context because the source of slowdown in PI stems from the ReLU operations whereas optimizations for plaintext inference focus on reducing FLOPs. In this paper we re-think ReLU computations and propose optimizations for PI tailored to properties of neural networks. Specifically, we reformulate ReLU as an approximate sign test and introduce a novel truncation method for the sign test that significantly reduces the cost per ReLU. These optimizations result in a specific type of stochastic ReLU. The key observation is that the stochastic fault behavior is well suited for the fault-tolerant properties of neural network inference. Thus, we provide significant savings without impacting accuracy. We collectively call the optimizations Circa and demonstrate improvements of up to 4.7$\times$ storage and 3$\times$ runtime over baseline implementations; we further show that Circa can be used on top of recent PI optimizations to obtain 1.8$\times$ additional speedup.

Machine learning as a service has given raise to privacy concerns surrounding clients' data and providers' models and has catalyzed research in private inference (PI): methods to process inferences without disclosing inputs. Recently, researchers have adapted cryptographic techniques to show PI is possible, however all solutions increase inference latency beyond practical limits. This paper makes the observation that existing models are ill-suited for PI and proposes a novel NAS method, named CryptoNAS, for finding and tailoring models to the needs of PI. The key insight is that in PI operator latency cost are inverted: non-linear operations (e.g., ReLU) dominate latency, while linear layers become effectively free. We develop the idea of a ReLU budget as a proxy for inference latency and use CryptoNAS to build models that maximize accuracy within a given budget. CryptoNAS improves accuracy by 3.4% and latency by 2.4x over the state-of-the-art.

We present Safe-TPU, a framework for secure computations of Deep Neural Networks (DNNs) in untrusted hardware corrupted by Trojans or fault injection attacks. This work leverages previous advances on interactive proof (IP) systems for verifying, at run-time, the correctness of a neural network’s computations, and makes three new contributions: (1) We present a Trojan resilient DNN hardware accelerator based on interactive proofs; (2) We introduce new protocol enhancements that significantly reduce the space and time required to generate proofs; and (3) we propose an implementation of Safe-TPU with high parallelism and reuse of existing resources already deployed in the baseline DNN accelerator. We prototype Safe-TPU on an FPGA and analyze its security guarantees. Experimentally, we show that Safe-TPU’s area overhead is small (28%) over the baseline DNN accelerator and is 3.15× faster than state-of-the-art, while at the same time, Safe-TPU guarantees to catch, with high probability, any incorrect computations.

Hardware accelerators are being increasingly deployed to boost the performance and energy efficiency of deep neural network (DNN) inference. In this paper we propose Thundervolt, a new framework that enables aggressive voltage underscaling of high-performance DNN accelerators without compromising classification accuracy even in the presence of high timing error rates. Using post-synthesis timing simulations of a DNN accelerator modeled on the Google TPU, we show that Thundervolt enables between 34%-57% energy savings on state-of-the-art speech and image recognition benchmarks with less than 1% loss in classification accuracy and no performance loss. Further, we show that Thundervolt is synergistic with and can further increase the energy efficiency of commonly used run-time DNN pruning techniques like Zero-Skip.

Energy harvesting is a promising solution to power Internet of Things (IoT) devices. Due to the intermittent nature of these energy sources, one cannot guarantee forward progress of program execution. Prior work has advocated for checkpointing the intermediate state to off-chip non-volatile memory (NVM). Encrypting checkpoints addresses the security concern, but significantly increases the checkpointing overheads. In this paper, we propose a new online checkpointing policy that judiciously determines when to checkpoint so as to minimize application time to completion while guaranteeing security. Compared to state-of-the-art checkpointing schemes that do not account for the overheads of encrypted checkpoints we improve execution time up to 1.4 x.

Inference using deep neural networks is often outsourced to the cloud since it is a computationally demanding task. However, this raises a fundamental issue of trust. How can a client be sure that the cloud has performed inference correctly? A lazy cloud provider might use a simpler but less accurate model to reduce its own computational load, or worse, maliciously modify the inference results sent to the client. We propose SafetyNets, a framework that enables an untrusted server (the cloud) to provide a client with a short mathematical proof of the correctness of inference tasks that they perform on behalf of the client. Specifically, SafetyNets develops and implements a specialized interactive proof (IP) protocol for verifiable execution of a class of deep neural networks, i.e., those that can be represented as arithmetic circuits. Our empirical results on three- and four-layer deep neural networks demonstrate the run-time costs of SafetyNets for both the client and server are low. SafetyNets detects any incorrect computations of the neural network by the untrusted server with high probability, while achieving state-of-the-art accuracy on the MNIST digit recognition (99.4%) and TIMIT speech recognition tasks (75.22%).