We present unexpected findings from a large-scale benchmark study evaluating Conditional Average Treatment Effect (CATE) estimation algorithms. By running 16 modern CATE models across 43,200 datasets, we find that: (a) 62\% of CATE estimates have a higher Mean Squared Error (MSE) than a trivial zero-effect predictor, rendering them ineffective; (b) in datasets with at least one useful CATE estimate, 80\% still have higher MSE than a constant-effect model; and (c) Orthogonality-based models outperform other models only 30\% of the time, despite widespread optimism about their performance. These findings expose significant limitations in current CATE models and suggest ample opportunities for further research. Our findings stem from a novel application of \textit{observational sampling}, originally developed to evaluate Average Treatment Effect (ATE) estimates from observational methods with experiment data. To adapt observational sampling for CATE evaluation, we introduce a statistical parameter, $Q$, equal to MSE minus a constant and preserves the ranking of models by their MSE. We then derive a family of sample statistics, collectively called $\hat{Q}$, that can be computed from real-world data. We prove that $\hat{Q}$ is a consistent estimator of $Q$ under mild technical conditions. When used in observational sampling, $\hat{Q}$ is unbiased and asymptotically selects the model with the smallest MSE. To ensure the benchmark reflects real-world heterogeneity, we handpick datasets where outcomes come from field rather than simulation. By combining the new observational sampling method, new statistics, and real-world datasets, the benchmark provides a unique perspective on CATE estimator performance and uncover gaps in capturing real-world heterogeneity.

Large Language Models’ safety-aligned behaviors, such as refusing harmful queries, can be represented by linear directions in activation space. Previous research modeled safety behavior with a single direction, limiting mechanistic understanding to an isolated safety feature. In this work, we discover that safety-aligned behavior is jointly controlled by multi-dimensional directions. Namely, we study the vector space of representation shifts during safety fine-tuning on Llama 3 8B for refusing jailbreaks. By studying orthogonal directions in the space, we first find that a dominant direction governs the model’s refusal behavior, while multiple smaller directions represent distinct and interpretable features like hypothetical narrative and role-playing. We then measure how different directions promote or suppress the dominant direction, showing the important role of secondary directions in shaping the model’s refusal representation. Finally, we demonstrate that removing certain trigger tokens in harmful queries can mitigate these directions to bypass the learned safety capability, providing new insights on understanding safety alignment vulnerability from a multi-dimensional perspective.

Privacy-preserving online ride-hailing (ORH) services can offer riders and drivers a more enhanced travel experience without disclosing their location privacy. Group ride-sharing is specifically designed for riders undertaking long-distance trips, allowing a group of riders with similar travel plans to share a single taxi. However, the lack of integrated route planning in existing privacy-preserving schemes prevents them from effectively matching riders with the most optimal taxi. In this paper, we propose a privacy-preserving group ride-sharing matching scheme, PGRoute, based on leveled fully homomorphic encryption (LFHE). In PGRoute, we propose a privacy-preserving path planning method and design a fast ciphertext-based distance matrix computation protocol for the key time-consuming modules in it to effectively improve the efficiency. PGRoute can plan routes for groups of riders, determine the optimal boarding order, and match the most suitable taxi while protecting the location privacy of both riders and taxi drivers. Theoretical analysis and experimental results demonstrate that PGRoute is secure and efficient within ORH systems. Compared to previous works, PGRoute reduces the pickup time for a group of riders by a factor of 2.9-$7.5\times $, and achieves 2.7-$6.7\times $higher computational efficiency.

Federated Learning (FL) is proposed to address the challenge of data isolation, with federated machine learning algorithms continuously evolving. Vertical Federated Learning (VFL) is a specific FL setting where parties possessing different features but the same entities collaboratively train models. However, existing VFL neural network schemes lack sufficient interactivity between involved parties, which results in diminished data value as model complexity increases. This limitation can lead to reduced accuracy with more intricate model structures. In our proposed scheme, we introduce additional interactions to enhance the effective utilization of data owned by the parties during the training period. To maintain security, we design a novel protocol that utilizes dimensionality reduction methods, ensuring interactions occur without information leakage and excessive communication costs. Experimental comparisons among various schemes validate the algorithm’s efficiency considerably. Additionally, we assess the model’s robustness against data poisoning attacks.

Online Ride Hailing (ORH) services gain remarkable development in the past decade, which enable riders and drivers to establish optimized rides via mobile device. To guarantee user safety, ORH service providers often monitor the ride trajectory and report the abnormal behavior once a trajectory deviation occurs. Along with the advantage of safety monitoring raises some vital privacy concerns on user location information leakage. In this paper, we propose a privacy-preserving safety monitoring scheme for ORH services, called pSafety. It enables an ORH service provider to detect user’s trajectory deviation without learning anything about users’ locations. In pSafety, we propose two secure trajectory similarity computation algorithms by using somewhat homomorphic encryption, which are used to plan an agreed path and measure trajectory deviation, respectively. Furthermore, we also design a ciphertext compression algorithm and a secure comparison protocol to improve efficiency. Theoretical analysis and experimental evaluations show that pSafety is secure, accurate and efficient.

Federated learning (FL) is an emerging paradigm of privacy-preserving distributed machine learning that effectively deals with the privacy leakage problem by utilizing cryptographic primitives. However, how to prevent poisoning attacks in distributed situations has recently become a major FL concern. Indeed, an adversary can manipulate multiple edge nodes and submit malicious gradients to disturb the global model’s availability. Currently, most existing works rely on an Independently Identical Distribution (IID) situation and identify malicious gradients using plaintext. However, we demonstrates that current works cannot handle the data heterogeneity scenario challenges and that publishing unencrypted gradients imposes significant privacy leakage problems. Therefore, we develop APFed, a layered privacy-preserving defense mechanism that significantly mitigates the effects of poisoning attacks in data heterogeneity scenarios. Specifically, we exploit HE as the underlying technique and employ the median coordinate as the benchmark. Subsequently, we propose a secure cosine similarity scheme to identify poisonous gradients, and we innovatively use clustering as part of the defense mechanism and develop a hierarchical aggregation that enhances our scheme’s robustness in IID and non-IID scenarios. Extensive evaluations on two benchmark datasets demonstrate that APFed outperforms existing defense strategies while reducing the communication overhead by replacing the expensive remote communication method with inexpensive intra-cluster communication.

Mobile Edge Computing (MEC) is a promising and fast-developing paradigm that provides cloud services at the edge of the network. MEC enables IoT devices to offload and execute their real-time applications at the proximity of these devices with low latency. Such applications include efficient manufacture inspection, virtual/augmented reality, image recognition, Internet of Vehicles (IoV), and e-Health. However, task offloading experiences security and privacy attacks such as data tampering, private data leakage, data replication, etc. To this end, in this paper, we propose a new blockchain-based framework for secure task offloading in MEC systems with guaranteed performance in terms of execution delay and energy consumption. First, blockchain technology is introduced as a platform to achieve data confidentiality, integrity, authentication, and privacy of task offloading in MEC. Second, we formulate an integration model of resource allocation and task offloading for a multi-user with multi-task MEC systems to optimize the energy and time cost. This is an NP-hard problem because of the curse-of-dimensionality and dynamic characteristics challenges of the considered scenario. Therefore, a deep reinforcement learning-based algorithm is developed to derive the close-optimal task offloading decision efficiently. Theoretical analysis and experimental results demonstrate that the proposed framework is resilient to several task offloading security attacks and it can save about 22.2% and 19.4% of system consumption with respect to the local and edge execution scenarios. Moreover, the benchmark analysis proves that the framework consumes few resources in terms of memory and disk usage, CPU utilization, and transaction throughput.

Online Ride Hailing (ORH) services enable a rider to request a taxi via a smartphone app in real time. When using ORH services, users (including riders and taxis) have to submit their locations to the ORH server. With received locations, the ORH server makes online ride matching between riders and taxis. There are serious privacy concerns for users to reveal location information to ORH servers. In this article, we propose an efficient and privacy-preserving ride matching scheme for ORH services, named EPRide. EPRide can find the taxi with the minimum road distance to serve an incoming rider, while protecting the location information of both taxis and riders against ORH servers or other curious servers. In EPRide, we propose an efficient exact shortest road distance computation approach over encrypted data, which converts road distance computation into Hamming distance computation over packed ciphertexts by using road network hypercube embedding and somewhat homomorphic encryption. Meanwhile, we design a secure comparison protocol, which efficiently compares encrypted distances in parallel by using ciphertexts blinding and packing, without leaking any distance. Theoretical analysis and experimental evaluations show that EPRide is secure, accurate and efficient.

An online ride hailing (ORH) service creates a typical supply-and-demand two-sided market, which enables riders and drivers to establish optimized rides conveniently via mobile applications. Group ridesharing is a novel form of ridesharing, which allows a group of riders to share a vehicle that holds the minimum aggregate distance to the whole group. Accompanied by the advantage of ORH services, there comes some vital privacy concerns. In this article, we propose a privacy-preserving online group ridesharing matching scheme for ORH services, called PGRide. PGRide can select the nearest driver to serve a group of riders, without leaking the location privacy of both riders and drivers. In PGRide, we propose an encrypted aggregate distance computation approach by using somewhat homomorphic encryption with ciphertexts packing, which efficiently computes the aggregate distances from a group of riders to large-scale dynamic drivers in encrypted form. Meanwhile, we design a secure minimum selection protocol by using ciphertexts packing and blinding, which efficiently finds the minimum element from a set of encrypted integers without leaking any actual element value. Theoretical analysis and performance evaluations prove that PGRide is secure, accurate, and efficient.

Online Ride Hailing (ORH) has extensively made our trip more convenient. With mobile devices, riders can request taxis through ORH systems in a short time. However, to enjoy ORH services, users need to submit their location information to ORH systems, which raises serious privacy concerns. In this paper, we study the privacy leakage of online ridesharing matching, a more complex and economy ORH service that allows riders to share rides with others, and propose a privacy-preserving shared ride matching scheme, called PSRide. PSRide can find the taxi with the minimum additional travel time to serve a new rider based on its existing schedule, while protecting the location privacy of both riders and taxis. In PSRide, we propose a zone-based minimum road travel time estimation approach and a secure comparison protocol to efficiently optimize the schedules of taxis for a new rider over encrypted data. We implement PSRide and analyze it thoroughly. Theoretical analysis and experimental evaluations show that PSRide is secure and efficient for ORH systems.

Online ride hailing (ORH) services enable a rider to request a driver to take him wherever he wants through a smartphone app on short notice. To use ORH services, users have to submit their ride information to the ORH service provider to make ride matching, such as pick-up/drop-off location. However, the submission of ride information may lead to the leakages of users’ privacy. In this paper, we focus on the issue of protecting the location information of both riders and drivers during ride matching and propose a privacy-preserving online ride matching scheme, called pRMatch. It enables an ORH service provider to find the closest available driver for an incoming rider over a city-scale road network, while protecting the location privacy of both riders and drivers against the ORH service provider and other unauthorized participants. In pRMatch, we compute the shortest road distance over encrypted data by using road network embedding and partially homomorphic encryption and further efficiently compare encrypted distances by using ciphertext packing and shuffling. The theoretical analysis and experimental results demonstrate that pRMatch is accurate and efficient, yet preserving users’ location privacy.

Structure references are commonly-used at the core of applications in a multitude of domains such as image processing, signal processing, especially the scientific and engineering applications. SIMD instruction sets, as SSE, AVX, AltiVec and 3DNow, provide a promising and widely available avenue for enhancing performance on modern processors. However existing memory accessing shackles limit the achieved performance for structure reference on modern SIMD architectures. In this paper, we propose a novel data layout transformation technology that addresses the accessing obstacles, along with a static analysis technique for detecting the legal loops in where this transformation is suitable. And this approach is implemented in the Optimizing Compiler Open64. The experimental results show that the proposed method can translate application with structure access into vectorizable codes, thereby advancing the execution efficiency adequately.

Due to the unattended operation nature, nodes in Unattended Wireless Sensor Networks (UWSNs) are susceptible to physical attacks. Once a sensor is compromised, the adversary will be able to learn all its secrets. While some previous works tried to address the node self-healing issue in UWSNs, little effort has been devoted to ensure the sustainability of node self-healing. In this paper, we present a novel sustainable node self-healing protocol for UWSNs. We generate unpredictable random data for key update and thus the node self-healing capability doesn't decrease when the number of attack rounds increases. We show both analytically and through simulation experiments that our protocol provides efficient and sustainable node self-healing capabilities with small overheads.

A base station (BS) is the controller and the data receiving center of a wireless sensor network. Hence, a reliable and secure BS is critical to the network. Once an attacker locates the BS, he can do a lot of damages to the network. In this paper, we study the BS location protection issue. First, we present a new attack on BS: the Parent-based Attack Scheme (PAS). The PAS can locate a BS within one radio (wireless transmission) range of sensors. Different from existing methods, the PAS determines the BS location based on parent-child relationship of sensor nodes. The PAS cannot be defended by existing BS protection schemes. To defend against the PAS, we design a new parent-free (PF) secure routing protocol for sensor networks. Our simulation results show that the PF protocol has small communication and computation costs, while ensuring the security of the BS.