High Overhead Research Articles

HPD-Depth: High performance decoding network for self-supervised monocular depth estimation

Self-supervised monocular depth estimation methods have shown promising results by leveraging geometric relationships among image sequences for network supervision. However, existing methods often face challenges such as blurry depth edges, high computational overhead, and information redundancy. This paper analyzes and investigates technologies related to deep feature encoding, decoding, and regression, and proposes a novel depth estimation network termed HPD-Depth, optimized by three strategies: utilizing the Residual Channel Attention Transition (RCAT) module to bridge the semantic gap between encoding and decoding features while highlighting important features; adopting the Sub-pixel Refinement Upsampling (SPRU) module to obtain high-resolution feature maps with detailed features; and introducing the Adaptive Hybrid Convolutional Attention (AHCA) module to address issues of local depth confusion and depth boundary blurriness. HPD-Depth excels at extracting clear scene structures and capturing detailed local information while maintaining an effective balance between accuracy and parameter count. Comprehensive experiments demonstrate that HPD-Depth performs comparably to state-of-the-art algorithms on the KITTI benchmarks and exhibits significant potential when trained with high-resolution data. Compared with the baseline model, the average relative error and squared relative error are reduced by 6.09% and 12.62% in low-resolution experiments, respectively, and by 11.3% and 18.5% in high-resolution experiments, respectively. Moreover, HPD-Depth demonstrates excellent generalization performance on the Make3D dataset.

Design of an effective multiple objects tracking framework for dynamic video scenes

Nowadays, the applications corresponding to video surveillance systems are getting popular due to their wide range of deployment in various places such as schools, roads, and airports. Despite the continuous evolution and increasing deployment of object-tracking features in video surveillance applications, the loopholes still need to be solved due to the limited functionalities of video-tracking systems. The existing video surveillance systems pose high processing overhead due to the larger size of video files. However, the traditional literature report quite sophisticated schemes which might successfully retain higher object detection accuracy from the video scenes but needs more effectiveness regarding computational complexity under limited computing resources. The study thereby identifies the scope of enhancement in traditional object-tracking functions. Further, it introduces a novel, cost-effective tracking model based on Gaussian mixture model (GMM) and Kalman filter (KF) that can accurately identify numerous mobile objects from a dynamic video scene and ensures computing efficiency. The study's outcome shows that the proposed strategic modelling offers better tracking performance for dynamic objects with cost-effective computation compared to the popular baseline approaches.

MLP-GNAS: Meta-learning-based predictor-assisted Genetic Neural Architecture Search system

Convolutional neural networks achieve state-of-the-art results on many image classification tasks. However, according to the No Free Lunch Theorem, no single model performs optimally for all the datasets. Hence, the manual development of these models using the hit-and-trial approach requires high computational overhead. Transfer learning mitigates this issue to some extent by transferring knowledge learned from similar domain tasks to target tasks through model architecture and weights. Yet, the model remains designed for the source task, and using the same model on the target task may result in over-parameterization or sub-optimal results. Henceforth, the current work proposes a meta-learning-based predictor-assisted genetic Neural Architecture Search (MLP-GNAS) system to automate the model generation process for plant disease detection tasks to validate its efficacy in real-world scenarios. To this cause, the MLP-GNAS system employs a meta-learning component to recommend the top 3 suitable models for a target dataset, followed by a genetic algorithm-based NAS to fine-tune the recommended model. In addition, the proposed approach uses a CNN-based performance predictor designed to discard models unlikely to surpass the optimal performance recorded thus far. Further, extensive comparative experiments demonstrate that MLP-GNAS-generated models outperform 14 state-of-the-art models on two distinct plant disease datasets with an accuracy of 96.94% and 92.04%. Moreover, further comparison with the manually fine-tuned and NAS-generated models reveals that the proposed system outperforms both to obtain an accuracy of 98.35% and 99.95% on respective datasets.

Encrypted fully model-free event-triggered HVAC control

This paper addresses the critical issue of privacy and high computational overhead in cloud-based HVAC control systems for building automation. Cloud computing platforms can infer sensitive information, such as occupancy, from sensor data, making privacy a major concern for users of intelligent buildings. Existing model-based control methods ensure privacy by introducing homomorphic encryption but they rely on approximating building dynamics and are resource-intensive, leading to increasing communication and computation costs. To solve this, we propose an encrypted fully model-free event-triggered control framework that guarantees privacy while minimizing both communication and computational costs without the need for approximating building dynamics. Our approach includes a model-free controller that regulates indoor temperature and CO2 levels, and an event-triggering mechanism that optimizes communication by only transmitting data when necessary. The numerical results from the simulations using the TRNSYS platform show that our method reduces communication between the system and cloud by 64% and decreases computation time by 75% compared to the latest encrypted HVAC control method. The primary contribution of this work is the novel model-free design that simplifies the control process modeling while significantly reducing resource demands, and provides a scalable and cost-effective solution for cloud-based HVAC control, enhancing both privacy and operational efficiency in smart buildings.

A mechanism model and data-driven fusion approach for rapid consequence prediction of explosion accidents in chemical clusters

Explosions are considered as a major hazard for chemical cluster safety because they can cause catastrophic disasters. The rapid explosion consequence prediction is critical for emergency response to prevent such disasters. Traditional modeling methods, such as experiment and numerical simulation, provide insights into the explosion mechanism but require high time overhead, thus challenging for real-time predictions. This paper proposes a mechanical model and data-driven fusion method for promptly predicting consequences of potential explosions in the specific chemical cluster. It exploits machine learning (ML) for explosion modeling of specific scenes. Numerical model validated by experiment is employed to provide high-quality benchmark data. Meanwhile, a zoning modeling and progressive-learning strategy is designed for enhancing data-driven model’s accuracy and efficiency. As a result, potential consequence of any explosion occurring in this specific scene can be quickly predicted by using the developed data-driven model. The proposed method is applied to vapor cloud explosion (VCE). Two data-driven models were developed to predict peak overpressure and time-varying overpressure, respectively. The results from numerical data show that our model exhibits high efficiency and strong generalization. Finally, a case study is presented to showcase the utilization of the proposed method. This research provides a reliable tool for achieving risk warning and effective response of explosions in chemical clusters.

Mask-Guided Spatial-Spectral MLP Network for High-Resolution Hyperspectral Image Reconstruction.

Hyperspectral image (HSI) reconstruction is a critical and indispensable step in spectral compressive imaging (CASSI) systems and directly affects our ability to capture high-quality images in dynamic environments. Recent research has increasingly focused on deep unfolding frameworks for HSI reconstruction, showing notable progress. However, these approaches have to break the optimization task into two sub-problems, solving them iteratively over multiple stages, which leads to large models and high computational overheads. This study presents a simple yet effective method that passes the degradation information (sensing mask) through a deep learning network to disentangle the degradation and the latent target's representations. Specifically, we design a lightweight MLP block to capture non-local similarities and long-range dependencies across both spatial and spectral domains, and investigate an attention-based mask modelling module to achieve the spatial-spectral-adaptive degradation representationthat is fed to the MLP-based network. To enhance the information flow between MLP blocks, we introduce a multi-level fusion module and apply reconstruction heads to different MLP features for deeper supervision. Additionally, we combine the projection loss from compressive measurements with reconstruction loss to create a dual-domain loss, ensuring consistent optical detection during HS reconstruction. Experiments on benchmark HS datasets show that our method outperforms state-of-the-art approaches in terms of both reconstruction accuracy and efficiency, reducing computational and memory costs.

BTIP: Branch Triggered Instruction Prefetcher Ensuring Timeliness

In CPU microarchitecture, caches store frequently accessed instructions and data by exploiting their locality, reducing memory access latency and improving application performance. However, contemporary applications with large code footprints often experience frequent Icache misses, which significantly degrade performance. Although Fetch-Directed Instruction Prefetching (FDIP) has been widely adopted in commercial processors to reduce Icache misses, our analysis reveals that FDIP still suffers from Icache misses caused by branch mispredictions and late prefetch, leaving considerable opportunity for performance optimization. Priority-Directed Instruction Prefetching (PDIP) has been proposed to reduce Icache misses caused by branch mispredictions in FDIP. However, it neglects Icache misses due to late prefetch and suffers from high storage overhead. In this paper, we proposed a branch-triggered instruction prefetcher (BTIP), which aims to prefetch Icache lines that FDIP cannot efficiently handle, including the Icache misses due to branch misprediction and late prefetch. We also introduce a novel Branch Target Buffer (BTB) organization, BTIP BTB, which stores prefetch metadata and reuses information from existing BTB entries, effectively reducing storage overhead. We implemented BTIP on the Champsim simulator and evaluated BTIP in detail using traces from the 1st Instruction Prefetching Championship (IPC-1). Our evaluation shows that BTIP outperforms both FDIP and PDIP. Specifically, BTIP reduces Icache misses by 38.0% and improves performance by 5.1% compared to FDIP. Additionally, BTIP outperforms PDIP by 1.6% while using only 41.9% of the storage space required by PDIP.

Optimization of frequent item set mining parallelization algorithm based on spark platform

In this paper, we propose a new method that combines the parallelism of the Spark-based platform with fast frequent mining, called STB_Apriori. Previous research has shown that traditional frequent itemset mining algorithms have high overhead when faced with large datasets and high-dimensional data computation, and generate a large number of candidate itemsets; at the same time, when faced with diverse user requirements, they often generate very sparse and diverse data. In order to solve the problem of fast mining of massive data, our idea originates from the capability of Spark distributed computing and the common optimisation ideas in Apriori mining, by using the efficient operator BitSet to achieve transaction compression, bit storage and data manipulation by Boolean matrices, and at the same time by parallelising the processing and optimising the algorithmic logic to achieve fast and frequent mining. In experiments on real-world datasets, our model consistently outperforms five widely used methods by a significant margin on very large data and maintains its excellence in the remaining cases, proving its effectiveness on real-world tasks, while further analysis shows that increasing the number of distributed nodes also incrementally and continuously improves performance.

An Intelligent Mobile Prediction method with Mini-batch HTIA-based Seq2Seq Networks

The prediction of individual mobility has been shown to hold significant commercial and social value in traffic planning and location advertising. However, the prediction of individual mobility is prone to substantial errors and high overhead time because of online learning or long input sequences. We propose an innovative sequence-to-sequence model with mini-batch hierarchical temporal incidence attention (HTIA) to address this issue. This model effectively captures long-term and short-term dependencies underlying individual mobility patterns. We perform mini-batch training via sequence padding in HTIA to increase the model's efficiency while maintaining interpretability. Through extensive experiments conducted on three public datasets exhibiting different degrees of uncertainty, we demonstrated that our proposed approach outperforms state-of-the-art competing schemes, reducing the best mean relative error results by more than 70.8%, 60.8%, and 69.9%.

OASR-WFBP: An overlapping aware start-up sharing gradient merging strategy for efficient communication in distributed deep learning

Wait-Free-Back-Propagation (WFBP) is a practical method for distributed deep-learning, but it suffers from a high communication overhead. To address this issue, the communication overhead can be reduced by overlapping gradient communication and computation, and sharing the startup time among multiple gradient communication phases. However, existing optimizations choose to share the startup time greedily and fail to coordinately exploit the overlapping opportunity between computation and communication. We propose an overlapping aware startup sharing Wait-Free-Back-Propagation (OASR-WFBP). An analytic model is designed to guide the sharing procedure. Evaluations show that OASR-WFBP achieves a 5%-16% optimization in iteration time over the state-of-the-art WFBP algorithm.

Provenance-based APT campaigns detection via masked graph representation learning

Advanced Persistent Threats (APTs) are well-planned, persistent, and highly stealthy cyberattacks designed to steal confidential information or disrupt specific target systems. Recent studies have used system audit logs to construct provenance graphs that describe system interactions to detect potentially malicious activities. Although they are effective, they still suffer from problems such as the need for a priori knowledge, lack of attack data, and high computational overhead that limit their application. In this paper, we propose a self-supervised learning-based APT detection model, APT-MGL, which learns the embedded representations of nodes through a graph mask self-encoder and transforms the detection problem into an outlier detection problem for malicious nodes. APT-MGL characterizes the behavior of nodes based on node type, action, and interaction frequency, and fuses the features through a multi-head self-attention mechanism. Then the node embedding is obtained by combining graph features and structural information using masked graph representation learning. Finally, the unsupervised outlier detection method is used to analyze the computed embeddings and obtain the final detection results. The experimental results show that APT-MGL outperforms existing monitoring models and achieves a small overhead.

MIDAS: Multi-layered attack detection architecture with decision optimisation

The proliferation of cyber attacks has led to the use of data-driven detection countermeasures, in an effort to mitigate this threat. Machine learning techniques, such as the use of neural networks, have become mainstream and proven effective in attack detection. However, these data-driven solutions are limited by: a) high computational overhead associated with data pre-processing and inference cost, b) inability to scale beyond a centralised deployment to cope with environmental variances, and c) requirement to use multiple bespoke detection models for effective attack detection coverage across the cyber kill chain. In this context, this paper introduces MIDAS, a cost-effective framework for attack detection, which introduces a dynamic decision boundary that is used in a multi-layered detection architecture. This is achieved by modelling the decision confidence of the participating detection models and judging its benefits using a novel reward policy. Specifically, a reward is assigned to a set of available actions, corresponding to a decision boundary, based on its cost-to-performance, where an overall cost-saving is prioritised. We evaluate our approach on two widely used datasets representing two of the most common threats today, i.e., phishing and malware. MIDAS shows that it effectively reduces the expenditure on detection inference and processing costs by controlling the frequency of expensive detection operations. This is achieved without significant sacrifice of attack detection performance.

Practical and lightweight defense against website fingerprinting

Website fingerprinting is a passive network traffic analysis technique that enables an adversary to identify the website visited by a user despite encryption and the use of privacy services such as Tor. Several website fingerprinting defenses built on top of Tor have been proposed to guarantee a user’s privacy by concealing trace features that are important to classification. However, some of the best defenses incur a high bandwidth and/or latency overhead. To combat this, new defenses have sought to be both lightweight — i.e., introduce a small amount of bandwidth overhead — and zero-delay to real network traffic. This work introduces a novel zero-delay and lightweight website fingerprinting defense, called BRO, which conceals the feature-rich beginning of a trace while still enabling the obfuscation of features deeper into the trace without spreading the padding budget thin. BRO schedules padding with a randomized beta distribution that can skew to both the extreme left and right, keeping the applied padding clustered to a finite portion of a trace. This work specifically targets deep learning attacks, which continue to be among the most accurate website fingerprinting attacks. Results show that BRO outperforms other well-known website fingerprinting defenses, such as FRONT, with similar bandwidth overhead.

Deep Learning-Based Approximated Observation Sparse SAR Imaging via Complex-Valued Convolutional Neural Network

Sparse synthetic aperture radar (SAR) imaging has demonstrated excellent potential in image quality improvement and data compression. However, conventional observation matrix-based methods suffer from high computational overhead, which is hard to use for real data processing. The approximated observation sparse SAR imaging method relieves the computation pressure, but it needs to manually set the parameters to solve the optimization problem. Thus, several deep learning (DL) SAR imaging methods have been used for scene recovery, but many of them employ dual-path networks. To better leverage the complex-valued characteristics of echo data, in this paper, we present a novel complex-valued convolutional neural network (CNN)-based approximated observation sparse SAR imaging method, which is a single-path DL network. Firstly, we present the approximated observation-based model via the chirp-scaling algorithm (CSA). Next, we map the process of the iterative soft thresholding (IST) algorithm into the deep network form, and design the symmetric complex-valued CNN block to achieve the sparse recovery of large-scale scenes. In comparison to matched filtering (MF), the approximated observation sparse imaging method, and the existing DL SAR imaging methods, our complex-valued network model shows excellent performance in image quality improvement especially when the used data are down-sampled.

Tensor Based Semi-Blind Channel Estimation for Reconfigurable Intelligent Surface-Aided Multiple-Input Multiple-Output Communication Systems

Reconfigurable intelligent surfaces (RISs) are a promising technology for sixth-generation (6G) wireless networks. However, a fully passive RIS cannot independently process signals. Wireless systems equipped with it often encounter the challenge of large channel matrix dimensions when acquiring channel state information using pilot-assisted algorithms, resulting in high pilot overhead. To address this issue, this article proposes a semi-blind joint channel and symbol estimation receiver without a pilot training stage for RIS-aided multiple-input multiple-output (MIMO) (including massive MIMO) communication systems. In a semi-blind system, a transmission symbol matrix and two channel matrices are coupled within the received signals at the base station (BS). We decouple them by building two parallel factor (PARAFAC) tensor models. Leveraging PARAFAC tensor decomposition, we transform the joint channel and symbol estimation problem into least square (LS) problems, which can be solved by Alternating Least Squares (ALSs). Our proposed scheme allows duplex communication. Compared to recently proposed pilot-based methods and semi-blind receivers, our results demonstrate the superior performance of our proposed algorithm in estimation accuracy and speed.

Secured Fog-Body-Torrent : A Hybrid Symmetric Cryptography with Multi-layer Feed Forward Networks Tuned Chaotic Maps for Physiological Data Transmission in Fog-BAN Environment

Recently, the Wireless Body Area Networks (WBAN) have become a promising and practical option in the tele-care medicine information system that aids for the better clinical monitoring and diagnosis. The trend of using Internet of Things (IoT) has propelled the WBAN technology to new dimension in terms of its network characteristics and efficient data transmission. However, these networks demand the strong authentication protocol to enhance the confidentiality, integrity, recoverability and dependability against the emerging cyber-physical attacks owing to the exposure of the IoT ecosystem and the confidentiality of biometric data. Hence this study proposes the Fog based WBAN infrastructure which incorporates the hybrid symmetric cryptography schemes with the chaotic maps and feed forward networks to achieve the physiological data info security without consuming the characteristics of power hungry WBAN devices. In the proposed model, scroll chaotic maps are iterated to produce the high dynamic keys streams for the real time applications and feed-forward layers are leveraged to align the complex input-output associations of cipher data for subsequent mathematical tasks. The feed forward layers are constructed which relies on the principle of Adaptive Extreme Learning Machines (AELM) thereby increasing randomness in the cipher keys thereby increasing its defensive nature against the different cyber-physical attacks and ensuring the high secured encrypted-decrypted data communication between the users and fog nodes. The real time analysis is conducted during live scenarios. BAN-IoT test beds interfaced with the heterogeneous healthcare sensors and various security metrics are analysed and compared with the various residing cryptographic algorithms. Results demonstrates that the recommended methodology has exhibited the high randomness characteristics and low computational overhead compared with the other traditional BAN oriented cryptography protocol schemes

PQSF: post-quantum secure privacy-preserving federated learning

In federated learning, secret sharing is a key technology to maintain the privacy of participants’ local models. Moreover, with the rapid development of quantum computers, existing federated learning privacy protection schemes based on secret sharing will no longer be able to guarantee the data security of participants in the post-quantum era. In addition, existing privacy protection methods have the problem of high communication and computational overhead. Although the multi-stage secret sharing scheme proposed by Pilaram et al. is one of the effective solutions to the above problems, existing studies have proven the privacy leakage risk of this scheme. This paper firstly designs a new lattice-based multi-stage secret sharing scheme Improved-Pilaram to solve the security problem, which allows participants to use public vectors to reconstruct different secret values without changing the secret sharing. Based on Improved-Pilaram, this article proposes a post-quantum secure federated learning scheme PQSF. PQSF uses double masking technology to encrypt model parameters and achieves mask reconstruction through secret sharing. Since Improved-Pilaram is multi-stage, participants do not need to update their local secret shares frequently during training. Analysis and experimental results show that the PQSF proposed in this paper reduces the communication complexity between participants and reduces the computational overhead by about 20% compared with existing solutions.

Secure and Efficient Data Hiding in the Cloud with Quantum-Resistant Encryption

Traditional steganographic methods in multi-cloud environments face significant challenges in cloud security, including complex key management, high computational overhead, and vulnerabilities to various attacks regaring secure data concealment. This paper presents a novel steganographic mechanism tailored for single cloud environments that address these issues. The proposed mechanism leverages a combination of homomorphic encryption, elliptic curve cryptography (ECC), and lattice-based cryptographic techniques to ensure robust security against classical and quantum attacks. Confining operations to a single cloud simplifies credential and key management, reducing operational complexity and enhancing system efficiency. The proposed approach mitigates the risks associated with multi-cloud environments and offers strong resistance to brute-force, differential, and steganalysis attacks. The implementation and validation of the mechanism demonstrate its efficacy in maintaining data integrity, confidentiality, and availability while minimizing computational resources. The results indicate that this approach significantly improves the security and performance of steganographic operations in cloud storage environments, setting a new standard for secure data concealment in cloud computing.

HardTaint: Production-Run Dynamic Taint Analysis via Selective Hardware Tracing

Dynamic taint analysis (DTA), as a fundamental analysis technique, is widely used in security, privacy, and diagnosis, etc. As DTA demands to collect and analyze massive taint data online, it suffers extremely high runtime overhead. Over the past decades, numerous attempts have been made to lower the overhead of DTA. Unfortunately, the reductions they achieved are marginal, causing DTA only applicable to the debugging/testing scenarios. In this paper, we propose and implement HardTaint, a system that can realize production-run dynamic taint tracking. HardTaint adopts a hybrid and systematic design which combines static analysis, selective hardware tracing and parallel graph processing techniques. The comprehensive evaluations demonstrate that HardTaint introduces only around 8% runtime overhead which is an order of magnitude lower than the state-of-the-arts, while without sacrificing any taint detection capability.

Sensor-Aware Data Imputation for Time-Series Machine Learning on Low-Power Wearable Devices

Wearable devices that have low-power sensors, processors, and communication capabilities are gaining wide adoption in several health applications. The machine learning algorithms on these devices assume that data from all sensors are available during runtime. However, data from one or more sensors may be unavailable due to energy or communication challenges. This loss of sensor data can result in accuracy degradation of the application. Prior approaches to handle missing data, such as generative models or training multiple classifiers for each combination of missing sensors are not suitable for low-energy wearable devices due to their high overhead at runtime. In contrast to prior approaches, we present an energy-efficient approach, referred to as Sensor-Aware iMputation (SAM), to accurately impute missing data at runtime and recover application accuracy. SAM first uses unsupervised clustering to obtain clusters of similar sensor data patterns. Next, it learns inter-relationship between clusters to obtain imputation patterns for each combination of clusters using a principled sensor-aware search algorithm. Using sensor data for clustering before choosing imputation patterns ensures that the imputation is aware of sensor data observations. Experiments on seven diverse wearable sensor based time-series datasets demonstrate that SAM is able to maintain accuracy within 5% of the baseline with no missing data when one sensor is missing. We also compare SAM against generative adversarial imputation networks (GAIN), transformers, and k-nearest neighbor methods. Results show that SAM outperforms all three approaches on average by more than 25% when two sensors are missing with negligible overhead compared to the baseline.