Redundant Dependencies Research Articles

Adversarial-Causal Representation Learning Networks for Machine fault diagnosis under unseen conditions

To address the challenges of obtaining diverse data, domain generalization (DG) methods for fault diagnosis have been developed. Domain adversarial methods are currently the most popular, due to their ability to handle data from unknown domains without requiring target domain information. However, their capacity to extract domain-irrelevant features remains challenging, often resulting in accuracy below 90% in many DG scenarios. This limitation stems from their inability to fully capture global dependencies, causing feature entanglement and redundant dependencies. To address these issues, we proposed a novel intelligent fault diagnosis method called Adversarial-Causal Representation Learning Networks (ACRLN), which is based on causal learning. By spatial mask domain adversarial method, ACRLN can significantly enhance data utilization by fully capturing the global dependency that are often ignored by domain adversarial algorithms. At the same time, causal learning is integrated into the ACRLN to further accomplish feature decoupling and the reduction of redundant dependency. This is achieved through channel feature orthogonality method combined with a loss function rooted in correlation analysis. Moreover, it adeptly addresses the spill-over effect often encountered in causal learning. Finally, ACRLN achieves better results and proves its effectiveness by comparison with several state-of-the-art fault diagnosis and DG algorithms on multiple datasets.

Polyp-LVT: Polyp segmentation with lightweight vision transformers

Automatic segmentation of polyps in endoscopic images is crucial for early diagnosis and surgical planning of colorectal cancer. However, polyps closely resemble surrounding mucosal tissue in both texture and indistinct borders and vary in size, appearance, and location which possess great challenge to polyp segmentation. Although some recent attempts have been made to apply Vision Transformer (ViT) to polyp segmentation and achieved promising performance, their application in clinical scenarios is still limited by high computational complexity, large model size, redundant dependencies, and significant training costs. To address these limitations, we propose a novel ViT-based approach named Polyp-LVT, strategically replacing the attention layer in the encoder with a global max pooling layer, which significantly reduces the model’s parameter count and computational cost while keeping the performance undegraded. Furthermore, we introduce a network block, named Inter-block Feature Fusion Module (IFFM), into the decoder, aiming to offer a streamlined yet highly efficient feature extraction. We conduct extensive experiments on three public polyp image benchmarks to evaluate our method. The experimental results show that compared with the baseline models, our Polyp-LVT network achieves a nearly 44% reduction in model parameters while gaining comparable segmentation performance.

The Lighter the Better: Rethinking Transformers in Medical Image Segmentation Through Adaptive Pruning

Vision transformers have recently set off a new wave in the field of medical image analysis due to their remarkable performance on various computer vision tasks. However, recent hybrid-/transformer-based approaches mainly focus on the benefits of transformers in capturing long-range dependency while ignoring the issues of their daunting computational complexity, high training costs, and redundant dependency. In this paper, we propose to employ adaptive pruning to transformers for medical image segmentation and propose a lightweight and effective hybrid network APFormer. To our best knowledge, this is the first work on transformer pruning for medical image analysis tasks. The key features of APFormer are self-regularized self-attention (SSA) to improve the convergence of dependency establishment, Gaussian-prior relative position embedding (GRPE) to foster the learning of position information, and adaptive pruning to eliminate redundant computations and perception information. Specifically, SSA and GRPE consider the well-converged dependency distribution and the Gaussian heatmap distribution separately as the prior knowledge of self-attention and position embedding to ease the training of transformers and lay a solid foundation for the following pruning operation. Then, adaptive transformer pruning, both query-wise and dependency-wise, is performed by adjusting the gate control parameters for both complexity reduction and performance improvement. Extensive experiments on two widely-used datasets demonstrate the prominent segmentation performance of APFormer against the state-of-the-art methods with much fewer parameters and lower GFLOPs. More importantly, we prove, through ablation studies, that adaptive pruning can work as a plug-n-play module for performance improvement on other hybrid-/transformer-based methods. Code is available at https://github.com/xianlin7/APFormer.

GaitGCN++: Improving GCN-based gait recognition with part-wise attention and DropGraph

Gait recognition is becoming one of the promising methods for biometric authentication owing to its self-effacing nature. Contemporary approaches of joint position-based gait recognition generally model gait features using spatio-temporal graphs which are often prone to overfitting. To incorporate long-range relationships among joints, these methods utilize multi-scale operators. However, they fail to provide equal importance to all joint combinations resulting in an incomplete realization of long-range relationships between joints and important body parts. Furthermore, only considering joint coordinates may fail to capture discriminatory information provided by the bone structures and motion. In this work, a novel multi-scale graph convolution approach, namely ‘GaitGCN++’, is proposed, which utilizes joint and bone information from individual frames and joint-motion data from consecutive frames providing a comprehensive understanding of gait. An efficient hop-extraction technique is utilized to understand the relationship between closer and further joints while avoiding redundant dependencies. Additionally, traditional graph convolution is enhanced by leveraging the ‘DropGraph’ regularization technique to avoid overfitting and the ‘Part-wise Attention’ to identify the most important body parts over the gait sequence. On the benchmark gait recognition dataset CASIA-B and GREW, we outperform the state-of-the-art in diversified and challenging scenarios.

S-MAT: Semantic-Driven Masked Attention Transformer for Multi-Label Aerial Image Classification.

Multi-label aerial scene image classification is a long-standing and challenging research problem in the remote sensing field. As land cover objects usually co-exist in an aerial scene image, modeling label dependencies is a compelling approach to improve the performance. Previous methods generally directly model the label dependencies among all the categories in the target dataset. However, most of the semantic features extracted from an image are relevant to the existing objects, making the dependencies among the nonexistant categories unable to be effectively evaluated. These redundant label dependencies may bring noise and further decrease the performance of classification. To solve this problem, we propose S-MAT, a Semantic-driven Masked Attention Transformer for multi-label aerial scene image classification. S-MAT adopts a Masked Attention Transformer (MAT) to capture the correlations among the label embeddings constructed by a Semantic Disentanglement Module (SDM). Moreover, the proposed masked attention in MAT can filter out the redundant dependencies and enhance the robustness of the model. As a result, the proposed method can explicitly and accurately capture the label dependencies. Therefore, our method achieves CF1s of , , and on three multi-label aerial scene image classification benchmark datasets: UC-Merced Multi-label, AID Multi-label, and MLRSNet, respectively. In addition, extensive ablation studies and empirical analysis are provided to demonstrate the effectiveness of the essential components of our method under different factors.

SMART: Semantic-Aware Masked Attention Relational Transformer for Multi-label Image Recognition

As objects usually co-exist in an image, learning the label co-occurrence is a compelling approach to improving the performance of multi-label image recognition. However, the dependencies among the non-exist categories in an image cannot be effectively evaluated. These redundant label dependencies may bring noise and further decrease the performance of classification. Therefore, we proposed SMART, a Semantic-aware Masked Attention Relational Transformer for Multi-label Image Recognition for multi-label image recognition tasks. In addition to leveraging Transformer to model the inter-class dependencies, the proposed masked attention filters out the redundant dependencies among the non-exist categories. SMART is able to explicitly and accurately capture the label dependencies without extra word embeddings. Moreover, our method achieves new state-of-the-art results on two benchmarks for multi-label image recognition: MS-COCO 2014 and NUS-WIDE. In addition, extensive ablation studies and empirical analysis are provided to demonstrate the effectiveness of the essential components of our method under different factors.

High-Order Interdependencies in the Aging Brain.

Background: Brain interdependencies can be studied from either a structural/anatomical perspective ("structural connectivity") or by considering statistical interdependencies ("functional connectivity" [FC]). Interestingly, while structural connectivity is by definition pairwise (white-matter fibers project from one region to another), FC is not. However, most FC analyses only focus on pairwise statistics and they neglect higher order interactions. A promising tool to study high-order interdependencies is the recently proposed O-Information, which can quantify the intrinsic statistical synergy and the redundancy in groups of three or more interacting variables. Methods: We analyzed functional magnetic resonance imaging (fMRI) data obtained at rest from 164 healthy subjects with ages ranging in 10 to 80 years and used O-Information to investigate how high-order statistical interdependencies are affected by age. Results: Older participants (from 60 to 80 years old) exhibited a higher predominance of redundant dependencies compared with younger participants, an effect that seems to be pervasive as it is evident for all orders of interaction. In addition, while there is strong heterogeneity across brain regions, we found a "redundancy core" constituted by the prefrontal and motor cortices in which redundancy was evident at all the interaction orders studied. Discussion: High-order interdependencies in fMRI data reveal a dominant redundancy in functions such as working memory, executive, and motor functions. Our methodology can be used for a broad range of applications, and the corresponding code is freely available. Impact statement Past research has showcased multiple changes to the brain's structural and functional properties caused by aging. Here we expand prior work through recent advancements in multivariate information theory, which provide richer and more theoretically principled analyses than existing alternatives. We show that the brains of older participants contain more redundant information at multiple spatial scales-that is, activation in different brain regions is less diverse, compared with younger participants-and identify a "redundancy core" constituted by prefrontal and motor cortices, which might explained impaired performance in the old population in functions such as working memory and executive control.

Enhanced robustness of single-layer networks with redundant dependencies.

Dependency links in single-layer networks offer a convenient way of modeling nonlocal percolation effects in networked systems where certain pairs of nodes are only able to function together. We study the percolation properties of the weak variant of this model: Nodes with dependency neighbors may continue to function if at least one of their dependency neighbors is active. We show that this relaxation of the dependency rule allows for more robust structures and a rich variety of critical phenomena, as percolation is not determined strictly by finite dependency clusters. We study Erdős-Rényi and random scale-free networks with an underlying Erdős-Rényi network of dependency links. We identify a special "cusp" point above which the system is always stable, irrespective of the density of dependency links. We find continuous and discontinuous hybrid percolation transitions, separated by a tricritical point for Erdős-Rényi networks. For scale-free networks with a finite degree cutoff we observe the appearance of a critical point and corresponding double transitions in a certain range of the degree distribution exponent. We show that at a special point in the parameter space, where the critical point emerges, the giant viable cluster has the unusual critical singularity S-S_{c}∝(p-p_{c})^{1/4}. We study the robustness of networks where connectivity degrees and dependency degrees are correlated and find that scale-free networks are able to retain their high resilience for strong enough positive correlation, i.e., when hubs are protected by greater redundancy.

Оптимизация графов потока управления в промежуточных представлениях языка функционально-потокового параллельного программирования

Functional dataflow programming languages are intended for the development of architecture-independent parallel programs and support the control of computations based on data availability. Due to the fact that at present parallel computing systems are very widespread, and their programming in imperative languages is associated with portability problems, the development of architecturally independent parallel programming tools is an urgent task. When such a program is translated, intermediate representations are formed as the information graph and the corresponding control graph. During program execution, data readiness signals are transmitted along the arcs of the control graph. An explicit selection of the control graph allows us not only to change the computational control strategies and ensure the adaptation of the program to the architecture features, but also to apply specific methods for optimizing control dependencies. The paper proposes transformation methods that provide optimization of the control graph. When generating a control graph from an informational one, redundant arcs are introduced into it, the removal of which does not affect the result of the program, but leads to its more efficient execution. It is shown that in dataflow programs, in addition to control dependencies inherent in other programming languages, additional ones associated with the implementation features of deferred or conditional computations described by delayed lists arise. A formal description of redundant dependencies of various types is given, as well as an effective algorithm for their identification. The developed approach can be applied to such dataflow programming languages as PIFAGOR and Smile.

Analysis of an k-out-of-n:G system with redundant dependency and repair equipment procurement

ABSTRACT This paper studies an k-out-of-n:G system with redundant dependency and repair equipment procurement lead time where the operating times and repair times of components follow exponential distributions and phase-type distributions, respectively. When one component breaks down, it is repaired by a repair equipment. The repair equipment may fail during the repair period and the following repair is not ‘as good as new’. After a number of repairs, it is replaced by a new one. The new spare repair equipment for replacement is only available by an order, and the procurement lead time for delivering follows a phase-type distribution. Moreover, in the multi-component system, the redundant dependency is taken into account. Applying the matrix-analytical method, the system availability, the rate of occurrence of failures of the system, the expected number of broken components, the availability and the rate of occurrence of failures of the repair equipment are derived. Finally, numerical examples are given to show these theoretical results.

Redundancy Dependence in the Context of Competing Risks and Dynamic Degradation

Parallel redundancy is a common approach to increase network availability. As a consequence of providing redundancy, the loading of that component changes, affecting its reliability. In this article, a holistic approach combining two models is introduced to assess both instantaneous and future effects on general failure rates in the context of competing risks and dynamic degradation. Concretely, both are designed to account for the duration of abnormal loading events. First, a per-unit-degradation method is introduced to map the different loading states onto a single normal operation sojourn time variable. Second, a time-inhomogeneous Markov chain is introduced to assess the irreversible degradation following a nondetrimental abnormal loading event. The numerical illustration shows that both models are necessary to correctly assess the effects of redundancy dependence if the underlying failure distribution has a shape parameter exceeding one.

Availability Analysis of a Repairable Series-Parallel System with Redundant Dependency

This paper investigates the steady-state availability of a repairable series-parallel system with redundant dependency. The different types of components and repairmen are taken into account, the failure rate of the operating component varies as the number of other failed components and the repair rate of the failed component is constant in each parallel redundant subsystem. To quantify the redundant dependency, a modified failure dependence function is introduced to determine the failure rate of the components in each subsystem. Markov theory and matrix analysis method are used to get the steady-state probability vector of each subsystem and the steady-state availability of the entire system. A numerical example is presented to illustrate the obtained results and to analyze the effect of redundant dependency class on the system availability.

RDE: A novel approach to improve the classification performance and expressivity of KDB.

Bayesian network classifiers (BNCs) have demonstrated competitive classification performance in a variety of real-world applications. A highly scalable BNC with high expressivity is extremely desirable. This paper proposes Redundant Dependence Elimination (RDE) for improving the classification performance and expressivity of k-dependence Bayesian classifier (KDB). To demonstrate the unique characteristics of each case, RDE identifies redundant conditional dependencies and then substitute/remove them. The learned personalized k-dependence Bayesian Classifier (PKDB) can achieve high-confidence conditional probabilities, and graphically interpret the dependency relationships between attributes. Two thyroid cancer datasets and four other cancer datasets from the UCI machine learning repository are selected for our experimental study. The experimental results prove the effectiveness of the proposed algorithm in terms of zero-one loss, bias, variance and AUC.

Multi-objective design optimisation of repairable k-out-of-n subsystems in series with redundant dependency

Several researches have been investigated on Multi-Objective Redundancy Allocation Problems (MORAPs), but none of them have considered the redundant dependency at the design stage. This latter which is a special kind of failure dependency can affect significantly the system performance. Due to this fact, this paper deals with the multi-objective system design optimisation with dependent components by focusing on two objectives: maximisation of system availability and minimisation of system cost with components choice and weight constraints. A system consisting of many k-out-of-n repairable subsystems connected in series is considered. The components of a subsystem are supposed to be identical and may be dependent. They are selected from a set of available component types. In addition to the redundancy level and the number of repair teams allocated to each subsystem, the choice of components type and the dependency level are also considered as decision variables. Since the described problem is NP hard, we propose three multi-objective meta-heuristic algorithms based on Non-dominated Sorting Genetic Algorithm (NSGA II) and Strength Pareto Evolutionary Algorithm (SPEA II) with different constraints handling. An exact method is also applied. To analyse their performances, numerical applications are provided and comparisons based on different well-known metrics are presented.

Identifying superword level parallelism with directed graph reachability}{Identifying superword level parallelism with directed graph reachability

SLP (superword level parallelism) is an efficient solution to exploit the parallelism between statements in the basic blocks of SIMD (single instruction multiple data). It has been implemented in almost all the mainstream vectorizing compilers. However, its vectorization ability is limited due to the conservativeness of the parallelism identification process. To solve this problem, this paper proposes an edg (extended dependence graph) approach to identify SLP. First, we extend the adg (array dependence graph) and sdg (statement dependence graph) to construct the edg, which includes both the dependences between each array pair and those between each statement pair. When a statement is represented as an SCC (strong connected component), all of its array references are also constructed in this SCC. We then eliminate the redundant dependence edges between the SCCs from the edg. The dependence information of each statement pair and its SLP vectorization are thus determined by analyzing the reachability between each node pair from the corresponding SCC. We implement this approach to optimize the Open64-5.0 compiler, which improves the compilers identification ability. The evaluation tests on the gcc-vect benchmarks show that the optimized Open64-5.0 compiler can identify more SLP vectorizable loops than the GCC4.9, and that the number of vectorizable loops is comparable to that of ICC14.0. The performance of our generated codes is better than the state-of-the-art for most practical applications.

Abstract B44: Emerging targets from Cancer Dependency Map v0.1

Abstract Precision Cancer Medicine requires the identification of vulnerabilities linked to genetic features of tumors. Recent studies utilizing highly annotated small molecule collections to assess dependencies across hundreds of genomically annotated cell lines have demonstrated the potential for such large-scale preclinical “Dependency Map” projects. We have undertaken a complementary approach using genetic perturbation tools (RNAi and CRISPR-Cas9 based loss-of-function viability screens), to systematically catalog preferential genetic dependencies and markers that predict response. These efforts are providing a foundation for the discovery of novel targets poised for early therapeutic discovery projects together with patient populations that may be enriched for responders to such therapies. Here, we present results from our initial Cancer Dependency Map consisting of RNAi loss-of-function screens across 503 cell lines, including both solid and hematopoietic tumors. We discovered 43 genes whose mutation or copy number creates a cancer dependency (oncogene addiction) including a novel dependency on the small GTPase, GNAI2 in Diffuse large B-cell Lymphoma. We discovered 142 genes in which elevated levels of expression create a dependency (gene addiction), a group of genes highly enriched for master regulator transcription factors such as SOX10, SPDEF, PAX8 and HNF1B. We discovered 474 genes for which hemizygous copy number creates a dependency (CYCLOPS genes), a group of genes highly enriched for members of macromolecular protein complexes including the spliceosome and proteasome. Finally, we discovered 171 genes that become a dependency when a redundant functional paralog is lost in cancer cells (redundant essentials). We demonstrate the mechanistic basis behind one such redundant essential dependency relationship in which promoter methylation of the UBB ubiquitin gene eliminates a compensatory mechanism leading to a novel vulnerability on the suppression of the UBC ubiquitin gene. These observations begin to provide an initial census, categorization and prioritization of robust cancer dependencies and support the potential impact for expanding early efforts to develop dependency maps of cancer. Citation Format: Francisca Vazquez, Aviad Tsherniak, Barbara Weir, Phil Montgomery, Glenn Cowley, Stanley Gill, Gregory Kryukov, Sasha Pantel, Will Harrington, Mike Burger, Robin Meyers, Levi Ali, Amy Goodale, Yenarae Lee, Levi Garraway, Jesse Boehm, David Root, Todd Golub, William Hahn. Emerging targets from Cancer Dependency Map v0.1. [abstract]. In: Proceedings of the AACR Precision Medicine Series: Targeting the Vulnerabilities of Cancer; May 16-19, 2016; Miami, FL. Philadelphia (PA): AACR; Clin Cancer Res 2017;23(1_Suppl):Abstract nr B44.

Redundant Design in Interdependent Networks

Modern infrastructure networks are often coupled together and thus could be modeled as interdependent networks. Overload and interdependent effect make interdependent networks more fragile when suffering from attacks. Existing research has primarily concentrated on the cascading failure process of interdependent networks without load, or the robustness of isolated network with load. Only limited research has been done on the cascading failure process caused by overload in interdependent networks. Redundant design is a primary approach to enhance the reliability and robustness of the system. In this paper, we propose two redundant methods, node back-up and dependency redundancy, and the experiment results indicate that two measures are effective and costless. Two detailed models about redundant design are introduced based on the non-linear load-capacity model. Based on the attributes and historical failure distribution of nodes, we introduce three static selecting strategies-Random-based, Degree-based, Initial load-based and a dynamic strategy-HFD (historical failure distribution) to identify which nodes could have a back-up with priority. In addition, we consider the cost and efficiency of different redundant proportions to determine the best proportion with maximal enhancement and minimal cost. Experiments on interdependent networks demonstrate that the combination of HFD and dependency redundancy is an effective and preferred measure to implement redundant design on interdependent networks. The results suggest that the redundant design proposed in this paper can permit construction of highly robust interactive networked systems.

Design optimization and redundant dependency study of series k — out — of — n : G repairable systems

This paper proposes a new optimization approach to find the best redundancy allocation of dependent repairable series k—out—of—n : G system using ant colony optimization. A model for the failure redundant dependency between the system components is formulated. It is considered in the system performance evaluation. An optimization model is then derived to be solved using the software LINGO and the proposed ACO algorithm. The objective consists of minimizing the system cost while satisfying a required level of system availability. The best system configuration in terms of number of components and number of repair teams in each subsystem is to be determined. The components are characterized by their failure rate, repair rate, unit acquisition cost and unit repair cost. In order to validate the proposed approach, an illustrative example is applied. A comparison between LINGO and ACO results is facilitated. The results show that the redundant dependency may positively affect the design configuration and the system cost. Furthermore, the proposed ACO outperforms LINGO for large scale systems in term of solution quality.

Transient Analysis of ak-out-of-n:GSystem withN-Policy, Repairmen’s Multiple Vacations, and Redundant Dependency

This paper analyzes ak-out-of-n:Grepairable system withN-policy, repairmen’s multiple synchronous vacations, and redundant dependency. When there is no failed component in the system, the repairmen leave for a vacation, the duration of which follows a phase type distribution. Upon returning from vacation, they should take another vacation if there are less thanNfailed components waiting in the system. This pattern continues until at leastNfailed components are waiting. Moreover, the redundant dependency which is a special kind of failure dependency is taken into account in the multicomponent system. Under such assumptions, the availability, the rate of occurrence of failures, and the reliability of the system are derived in transient regime by applying the quasi-birth-and-death process. Furthermore, the Runge-Kutta method is carried out to numerically discuss the time-dependent behavior of the system reliability measures. Finally, a special case of the system is presented to show the validity of our model.

Reliability Evaluation for Multi-State Markov Repairable Systems with Redundant Dependencies

In multi-state systems with load-sharing, the load will be redistributed among working units while some units go to failure. Hence dependencies exit among units. To model the evolution of these systems, a multi-state Markov repairable system with redundant dependencies is introduced in this paper. More than two states are allowed for each unit of the system, including perfect working, deterioration and complete failure. The state transition rate of each working unit is a function of the number of functioning units and is quantified by a redundant dependence function. A two-dimensional vector, whose elements denote respectively the number of units that are in perfect and degraded operating states, is presented to describe accurately the performance of the system. The state space of the system is divided into distinct state sets according to the value of the vector. The visit probability to a specified state set (the acceptable, excellent, operating and warning state sets), the time to the first system failure and the steady-state distribution of sojourn time in the acceptable state set are discussed. Markov theory and aggregated stochastic process theory are used to get these reliability indices. A numerical example is presented to illustrate the results obtained in the paper. The impact of redundant dependencies in the system is also considered.