Iterative Aggregation Research Articles

SBPA: Sybil-Based Backdoor Poisoning Attacks for Distributed Big Data in AIoT-Based Federated Learning System

Federated learning (FL) enables a great deal of distributed independent participants to collaborate in training without sharing data. Malicious adversary can poison the local model by backdoor poisoning attacks and utilize the characteristic that server cannot trace original data to make the poisoned model directly aggregated. Especially in the AIoT-FL network that generates large amounts of data in real-time, such an attack is more powerful. In this paper, we design a sybil-based backdoor poisoning attacks (SBPA) against the above vulnerability. Malicious participants inject backdoor triggers into distributed Big Data to covertly complete data poisoning. During subsequent iterative aggregation, the joint model activates the backdoor during testing to achieve misclassification for the backdoor images. Besides, malicious participants create some sybil nodes to join the aggregation by taking advantage of the vulnerability of system devices to be easily disconnected. They are committed to making the poisoned local models aggregated with higher probability. Their goal works on making the final global model misclassify backdoor images while keeping high classification accuracy on the other non-backdoor samples. We conduct extensive experiments on multiple datasets and exhibit more robust performance than the state-of-the-art in various metrics under both data distribution including i.i.d. and non-i.i.d. scenarios.

LAD: Layer-Wise Adaptive Distillation for BERT Model Compression.

Recent advances with large-scale pre-trained language models (e.g., BERT) have brought significant potential to natural language processing. However, the large model size hinders their use in IoT and edge devices. Several studies have utilized task-specific knowledge distillation to compress the pre-trained language models. However, to reduce the number of layers in a large model, a sound strategy for distilling knowledge to a student model with fewer layers than the teacher model is lacking. In this work, we present Layer-wise Adaptive Distillation (LAD), a task-specific distillation framework that can be used to reduce the model size of BERT. We design an iterative aggregation mechanism with multiple gate blocks in LAD to adaptively distill layer-wise internal knowledge from the teacher model to the student model. The proposed method enables an effective knowledge transfer process for a student model, without skipping any teacher layers. The experimental results show that both the six-layer and four-layer LAD student models outperform previous task-specific distillation approaches during GLUE tasks.

Multi-temporal Cloud Pixels Reconstruction Method for Optical Remote Sensing Satellite Images

Bachground: The existence of cloud pixels reduces the practicability of optical satellite remote sensing data. Existing problems: Existing cloud reconstruction methods generally cannot solve the following problems:(1)Large-scale thick cloud cannot be well reconstructed. (2)There are high requirements for reconstructed data. (3)Most data used to reconstructed are single temporal images. Ways: In order to overcome these problems, a new multi temporal weighted aggregation method is proposed. Specifically, we adopt a multi-temporal iterative aggregation method for cloud pixels to reconstruct and a multi-temporal weighted aggregation method for cloud shadow pixels to reconstruct. Result: Finally, the experiment proves that our method can quickly and accurately complete the cloud reconstruction, and under the effective uniform color strategy, a cloud- free image with accurate geometric position and uniform gray scale can be obtained. Conclusion: Experiments prove that the pixel reconstruction method proposed in this paper has achieved good cloud and cloud shadow pixel reconstruction effects in different types of ground objects.

ST‐SIGMA: Spatio‐temporal semantics and interaction graph aggregation for multi‐agent perception and trajectory forecasting

Scene perception and trajectory forecasting are two fundamental challenges that are crucial to a safe and reliable autonomous driving (AD) system. However, most proposed methods aim at addressing one of the two challenges mentioned above with a single model. To tackle this dilemma, this paper proposes spatio-temporal semantics and interaction graph aggregation for multi-agent perception and trajectory forecasting (ST-SIGMA), an efficient end-to-end method to jointly and accurately perceive the AD environment and forecast the trajectories of the surrounding traffic agents within a unified framework. ST-SIGMA adopts a trident encoder–decoder architecture to learn scene semantics and agent interaction information on bird’s-eye view (BEV) maps simultaneously. Specifically, an iterative aggregation network is first employed as the scene semantic encoder (SSE) to learn diverse scene information. To preserve dynamic interactions of traffic agents, ST-SIGMA further exploits a spatio-temporal graph network as the graph interaction encoder. Meanwhile, a simple yet efficient feature fusion method to fuse semantic and interaction features into a unified feature space as the input to a novel hierarchical aggregation decoder for downstream prediction tasks is designed. Extensive experiments on the nuScenes data set have demonstrated that the proposed ST-SIGMA achieves significant improvements compared to the state-of-the-art (SOTA) methods in terms of scene perception and trajectory forecasting, respectively. Therefore, the proposed approach outperforms SOTA in terms of model generalisation and robustness and is therefore more feasible for deployment in real-world AD scenarios.

Rethinking prediction alignment in one-stage object detection

Owing to their excellent performance and efficiency, one-stage detectors have been widely used in multimedia tasks, such as temporal action detection, object tracking, and video detection. However, misalignment between classification and regression branches limits the accuracy of the detector. Most existing works add an auxiliary branch or adopt a specific sample assignment strategy to alleviate this problem, but with little effect. In this paper, we attribute this to incomplete branch interactions and propose a comprehensive Predictive Aligned Object Detector (PAOD), which can better correlate two subtasks. Specifically, our proposed PAOD achieves a better trade-off between prediction-interactive and prediction-specific by adopting an Iterative Aggregation Module (IAM) and a Mutual Constraint Module (MCM). We also design an aligned label assignment with an adaptive metric and re-weighting mechanism to further narrow the misalignment between prediction heads. With negligible additional overhead, PAOD achieves 50.4 AP at single-model single-scale testing on the MS-COCO branch, which demonstrates the effectiveness of our proposal. Notably, PAOD consistently outperforms previous sota such as ATSS (47.7 AP), BorderDet (48.0 AP) and GFL (48.2 AP) by a large margin on COCO test-dev dataset, and achieves better performance than various dense detectors on Pascal VOC and CrowdHuman datasets. Code is available at https://github.com/JunruiXiao/PAOD.

LLRHNet: Multiple Lesions Segmentation Using Local-Long Range Features.

The encoder-decoder-based deep convolutional neural networks (CNNs) have made great improvements in medical image segmentation tasks. However, due to the inherent locality of convolution, CNNs generally are demonstrated to have limitations in obtaining features across layers and long-range features from the medical image. In this study, we develop a local-long range hybrid features network (LLRHNet), which inherits the merits of the iterative aggregation mechanism and the transformer technology, as a medical image segmentation model. LLRHNet adopts encoder-decoder architecture as the backbone which iteratively aggregates the projection and up-sampling to fuse local low-high resolution features across isolated layers. The transformer adopts the multi-head self-attention mechanism to extract long-range features from the tokenized image patches and fuses these features with the local-range features extracted by down-sampling operation in the backbone network. These hybrid features are used to assist the cascaded up-sampling operations to local the position of the target tissues. LLRHNet is evaluated on two multiple lesions medical image data sets, including a public liver-related segmentation data set (3DIRCADb) and an in-house stroke and white matter hyperintensity (SWMH) segmentation data set. Experimental results denote that LLRHNet achieves state-of-the-art performance on both data sets.

Improved CenterNet for Accurate and Fast Fitting Object Detection

Accurate and fast detection of typical fittings is the prerequisite of condition monitoring and fault diagnosis. At present, most successful fitting detectors are anchor‐based, which are challenging to meet the requirements of edge deployment. In this paper, we propose a novel anchor‐free method called HRM‐CenterNet. Firstly, the lightweight MobileNetV3 is introduced into CenterNet to extract multi‐scale features of different layers. In addition, the lightweight receptive field enhancement module is proposed for the deep layer features to further enhance the characterization power of global features and generate more accurate heatmaps. Finally, the high‐resolution feature fusion network with iterative aggregation is designed to reduce the loss of spatial semantic information in subsampling and further improve the accuracy of small and occlusion objects. Experiments are carried out on the TFITS and PASCAL VOC datasets. The results show that the size of the network is more than 60% lower than that of CenterNet. Compared with other detectors, our method achieves comparable accuracy with all accurate models at a much faster speed and meets the performance requirements of real‐time detection.

OSLO: On-the-Sphere Learning for Omnidirectional Images and Its Application to 360-Degree Image Compression.

State-of-the-art 2D image compression schemes rely on the power of convolutional neural networks (CNNs). Although CNNs offer promising perspectives for 2D image compression, extending such models to omnidirectional images is not straightforward. First, omnidirectional images have specific spatial and statistical properties that can not be fully captured by current CNN models. Second, basic mathematical operations composing a CNN architecture, e.g., translation and sampling, are not well-defined on the sphere. In this paper, we study the learning of representation models for omnidirectional images and propose to use the properties of HEALPix uniform sampling of the sphere to redefine the mathematical tools used in deep learning models for omnidirectional images. In particular, we: i) propose the definition of a new convolution operation on the sphere that keeps the high expressiveness and the low complexity of a classical 2D convolution; ii) adapt standard CNN techniques such as stride, iterative aggregation, and pixel shuffling to the spherical domain; and then iii) apply our new framework to the task of omnidirectional image compression. Our experiments show that our proposed on-the-sphere solution leads to a better compression gain that can save 13.7% of the bit rate compared to similar learned models applied to equirectangular images. Also, compared to learning models based on graph convolutional networks, our solution supports more expressive filters that can preserve high frequencies and provide a better perceptual quality of the compressed images. Such results demonstrate the efficiency of the proposed framework, which opens new research venues for other omnidirectional vision tasks to be effectively implemented on the sphere manifold.

Large-Window Curvature Computations for High-Resolution Digital Elevation Models

With the increasing availability of high-resolution digital elevation model (DEM) data, a need has emerged for new processing techniques. Topographic variables such as slope and curvature are relevant on length scales far larger than the pixel resolution of modern DEM data sets. We propose an approach for computing slope and curvature that uses standard regression coefficients over large windows while generating output on the full resolution of the original data, without adding substantially to the computation time. In our approach, to which we will refer as Iterative Aggregation of Regression Terms (IART), aggregates for fitting a quadratic function are computed iteratively from the DEM data in a process that scales logarithmically with the window size. We show that the IART algorithm produces results of much higher-quality than the two-step process of applying neighborhood operations like focal statistics followed by small-window topographic computations, at comparable computational cost.

Gan-Based Iterative Aggregation Graph Representation Learning

Gan-Based Iterative Aggregation Graph Representation Learning

Memory and communication efficient algorithm for decentralized counting of nodes in networks.

Node counting on a graph is subject to some fundamental theoretical limitations, yet a solution to such problems is necessary in many applications of graph theory to real-world systems, such as collective robotics and distributed sensor networks. Thus several stochastic and naïve deterministic algorithms for distributed graph size estimation or calculation have been provided. Here we present a deterministic and distributed algorithm that allows every node of a connected graph to determine the graph size in finite time, if an upper bound on the graph size is provided. The algorithm consists in the iterative aggregation of information in local hubs which then broadcast it throughout the whole graph. The proposed node-counting algorithm is on average more efficient in terms of node memory and communication cost than its previous deterministic counterpart for node counting, and appears comparable or more efficient in terms of average-case time complexity. As well as node counting, the algorithm is more broadly applicable to problems such as summation over graphs, quorum sensing, and spontaneous hierarchy creation.

MIA-UNet: Multi-Scale Iterative Aggregation U-Network for Retinal Vessel Segmentation

As an important part of the new generation of information technology, the Internet of Things (IoT) has been widely concerned and regarded as an enabling technology of the next generation of health care system. The fundus photography equipment is connected to the cloud platform through the IoT, so as to realize the real-time uploading of fundus images and the rapid issuance of diagnostic suggestions by artificial intelligence. At the same time, important security and privacy issues have emerged. The data uploaded to the cloud platform involves more personal attributes, health status and medical application data of patients. Once leaked, abused or improperly disclosed, personal information security will be violated. Therefore, it is important to address the security and privacy issues of massive medical and healthcare equipment connecting to the infrastructure of IoT healthcare and health systems. To meet this challenge, we propose MIA-UNet, a multi-scale iterative aggregation U-network, which aims to achieve accurate and efficient retinal vessel segmentation for ophthalmic auxiliary diagnosis while ensuring that the network has low computational complexity to adapt to mobile terminals. In this way, users do not need to upload the data to the cloud platform, and can analyze and process the fundus images on their own mobile terminals, thus eliminating the leakage of personal information. Specifically, the interconnection between encoder and decoder, as well as the internal connection between decoder sub-networks in classic U-Net are redefined and redesigned. Furthermore, we propose a hybrid loss function to smooth the gradient and deal with the imbalance between foreground and background. Compared with the U-Net, the segmentation performance of the proposed network is significantly improved on the premise that the number of parameters is only increased by 2%. When applied to three publicly available datasets: DRIVE, STARE and CHASE_DB1, the proposed network achieves the accuracy/F1-score of 96.33%/84.34%, 97.12%/83.17% and 97.06%/84.10%, respectively. The experimental results show that the MIA-UNet is superior to the state-of-the-art methods.

Decomposition-based real-time control of multi-stage transfer lines with residence time constraints

It is commonly observed in the food industry, battery production, automotive paint shop, and semiconductor manufacturing that an intermediate product’s residence time in the buffer within a production line is controlled by a time window to guarantee product quality. There is typically a minimum time limit reflected by a part’s travel time or process requirement. Meanwhile, these intermediate parts are prevented from staying in the buffer for too long by an upper time limit, exceeding which a part will be scrapped or need additional treatment. To increase production throughput and reduce scrap, one needs to control machines’ working mode according to real-time system information in the stochastic production environment, which is a difficult problem to solve, due to the system’s complexity. In this article, we propose a novel decomposition-based control approach by decomposing a production system into small-scale subsystems based on domain knowledge and their structural relationship. An iterative aggregation procedure is then used to generate a production control policy with convergence guarantee. Numerical studies suggest that the decomposition-based control approach outperforms general-purpose reinforcement learning method by delivering significant system performance improvement and substantial reduction on computation overhead.

Optimization of planned solutions based on network models for large-scale problems

This article covers the development of the project planning algorithm based on network models for solving large-scale problems. The planning algorithm which brings the problem of plan optimization based on a network model down to a general problem of non-linear programming with the possibility of flexible formation of the goal function and constraints is completed by the iterative aggregation method for problems of non-linear programming which allows increasing the efficiency of solving large-scale problems.

Workforce Allocation in Motorcycle Transmission Assembly Lines: A Case Study on Modeling, Analysis, and Improvement

This letter presents a case study of workforce allocation in a gear assembly line at a motorcycle powertrain manufacturing plant. Through structural modeling of operators' work flow and part flow, assembly system models with different configurations are developed. Using iterative aggregation techniques, the performances of such systems are analyzed, and the associated workforce allocation strategies are investigated. By identifying and mitigating system bottlenecks, improvement recommendations are obtained. Such a study provides an effective method for operator assignment in assembly systems.

Transient Analysis of Multiproduct Bernoulli Serial Lines With Setups

Performance analysis of flexible manufacturing systems has received substantial research attention. Most of the studies focus on steady-state behavior. The transient performance, which is critical for system operation, control, and improvement, is less investigated. No analytical method is available to evaluate the transients in flexible lines with unreliable machines, finite buffers, and setup times. In this article, an analytical model is introduced to study the transient performance of multiproduct serial lines with Bernoulli reliability machines and nonnegligible setups during changeovers. To reduce the state-space dimension, an approximation method is introduced to evaluate the performance in two-machine lines, where the sequence of multiple part types in the inventory is approximated by the ratio of product mix. Such an approximation method leads to high precision of production rate estimation. Then, for longer lines, the analysis approach in two-machine lines is used as a building block in an iterative aggregation procedure. Numerical experiments show that the iteration procedure is convergent and results in an acceptable accuracy in performance evaluation. In addition, system properties, such as monotonicity and reversibility, are investigated.

Flexible preventative maintenance for serial production lines with multi-stage degrading machines and finite buffers

In production systems, machines are typically subject to degradation, which is a gradual and accumulating process that can influence the performance of the production systems. In this work, we focus on the flexible preventative maintenance problem on serial production lines with multi-stage degrading machines and finite buffers. Condition-based maintenance decisions are first investigated for a two-machine-one-buffer system, considering machine degradation stages and the buffer level. The optimal maintenance policy is obtained using Markov decision models. For longer lines, approximation methods are developed based on the results from the two machine case. Specifically, an iterative state and machine aggregation approach is developed to find the optimal preventative maintenance policy for each machine in large systems. Numerical experiments show that the proposed method outperforms the state-of-the-art.

Unconventional analogs of single-parametric method of iterational aggregation

When we solve practical problems that arise, for example, in mathematical economics, in the theory of Markov processes, it is often necessary to use the decomposition of operator equations using methods of iterative aggregation. In the studies of these methods for the linear equation $x=Ax+b$ the most frequent are the conditions of positiveness of the operator $A$, constant $b$ and the aggregation functions, and also the implementation of the inequality $\rho(A) <1$ for the spectral radius $\rho(A)$ of the operator $A$.
 In this article for an approximate solution of a system composed of the equation $x = Ax + b$ represented in the form $ x = A_1x + A_2x + b, $ where $ b \in E, $ $ E $ is a Banach space, $ A_1, A_2 $ are linear continuous operators that act from $ E $ to $ E $ and the auxiliary equation $ y = \lambda y - (\varphi, A_2x) - (\varphi, b) $ with a real variable $ y $, where $ (\varphi, x) $ is the value of the linear functional $ \varphi \in E ^ * $ on the elements $ x \in E $, $ E^* $ is conjugation with space $ E $, an iterative process is constructed and investigated \begin{equation*} \begin{split} x^{(n+1)}&=Ax^{(n)}+b+\frac{\sum\limits_{i=1}^{m}A^i_1x^{(n)}}{(\varphi, x^{(n)})\sum\limits_{i=0}^{m}\lambda^i}(y^{(n)}-y^{(n+1)}) \quad (m<\infty),\\ y^{(n+1)}&=\lambda y^{(n+1)}-(\varphi,A_2x^{(n)})-(\varphi,b). \end{split} \end{equation*}
 The conditions are established under which the sequences $ {x ^ {(n)}}, {y ^ {(n)}}$, constructed with the help of these formulas, converge to $ x ^ *, y ^ * $ as a component of solving the system constructed from equations $ x = A_1 x + A_2 x + b $ and the equation $ y = \lambda y - (\varphi, A_2 x) - (\varphi, b) $ not slower than the rate of convergence of the geometric progression with the denominator less than $1$. In this case, it is required that the operator $ A $ be a compressive and constant by sign, and that the space $ E $ is semi-ordered. The application of the proposed algorithm to systems of linear algebraic equations is also shown.

Stereoview to Multiview Conversion Architecture for Auto-Stereoscopic 3D Displays

In this paper, a stereoview to multiview conversion system, which includes stereo matching and depth image-based rendering (DIBR) hardware designs, is proposed. To achieve an efficient architecture, the proposed stereo matching algorithm simply generates the raw matching costs and aggregates cost based on 1D iterative aggregation schemes. For the DIBR architecture, an inpainting-based method is used to find the most similar patch from the background, according to depth information. The simulation results show that the designed architecture achieves an averaged peak signal-to-noise ratio of 30.2 dB and structure similarity of 0.94 for the tested images. The hardware design for the proposed 2D to 3D conversion system operates at a maximum clock frequency of 160.2 MHz for outputting 1080p ( $1920 \times 1080$ ) video at 60 frames per second.

Boundary crash data assignment in zonal safety analysis: An iterative approach based on data augmentation and Bayesian spatial model

Boundary effect refers to the issue of ambiguous allocation of crashes occurred on or near the boundaries of neighboring zones in zonal safety analysis. It results in bias estimates for associate measure between crash occurrence and possible zonal factors. It is a fundamental problem to compensate for the boundary effect and enhance the model predictive performance. Compared to conventional approaches, it might be more reasonable to assign the boundary crashes according to the crash predisposing agents, since the crash occurrence is generally correlated to multiple sources of risk factors. In this study, we proposed a novel iterative aggregation approach to assign the boundary crashes, according to the ratio of model-based expected crash number in adjacent zones. To verify the proposed method, a case study using a dataset of 738 Traffic Analysis Zones (TAZs) from the county of Hillsborough in Florida was conducted. Using Bayesian spatial models (BSMs), the proposed approach demonstrated the capability in reasonably compensating for the boundary effect with better model estimation and predictive performance, as compared to three conventional approaches (i.e., half and half ratio method, one to one ratio method, and exposure ratio method). Results revealed that several factors including the number of intersections, road segment length with 35 mph speed limit, road segment length with 65 mph speed limit and median household income, were sensitive to the boundary effect.