Diffusion models have demonstrated impressive capabilities in synthesizing diverse content. However, despite their high-quality outputs, these models often perpetuate social biases, including those related to gender and race. These biases can potentially contribute to harmful real-world consequences, reinforcing stereotypes and exacerbating inequalities in various social contexts. While existing research on diffusion bias mitigation has predominantly focused on guiding content generation, it often neglects the intrinsic mechanisms within diffusion models that causally drive biased outputs. In this paper, we investigate the internal processes of diffusion models, identifying specific decision-making mechanisms, termed bias features, embedded within the model architecture. By directly manipulating these features, our method precisely isolates and adjusts the elements responsible for bias generation, permitting granular control over the bias levels in the generated content. Through experiments on both unconditional and conditional diffusion models across various social bias attributes, we demonstrate our method’s efficacy in managing generation distribution while preserving image quality. We also dissect the discovered model mechanism, revealing different intrinsic features controlling fine-grained aspects of generation, boosting further research on mechanistic interpretability of diffusion models. The project website is at https://foundation-model-research.github.io/difflens.

The integration of EtherCAT and TSN has been proposed to enhance performance of EtherCAT networks in industrial automation. EtherCAT over TSN transforms a traditional EtherCAT ring into multiple shorter rings interconnected via TSN switches, enabling concurrent data transmission across segments and reducing cycle time. However, the use of multiple segments introduces contention for the master in the return path, potentially leading to scheduling failures and an increase in cycle time. We study the impact of network topology, i.e., the number of segments and slave node distribution, on schedulability, and formulate the topology optimization problem for the converged network based on schedulability analysis. We evaluate our methodology and optimal solution using SMT and Integer Programming (LIP) solvers, respectively. Numerical results demonstrate the effectiveness of our method, and our solution outperforms baselines.

With the rapid growth of Industry 4.0, time-sensitive networking (TSN) has emerged as the new infrastructure for future industrial Internet of Things (IoT) communication. Ensuring the compatibility between TSN and legacy networks is inevitable. The ideal compatibility is to achieve deterministic interconnection and interoperability without changes in hardware and communication protocols, in other words, only using standard devices with software management. This paper targets the ideal compatibility of TSN and Profinet Isochronous Real Time (IRT). First, we propose an inter-domain Multiple Transmission Opportunity Mechanism (MTOM) to enable the asynchronous converged network of TSN and Profinet. The mechanism reserves multiple transmission time slots for cross-domain data flows to reduce their end-to-end delay and jitter. Second, we formulate an asynchronous scheduling model (ASM) based on MTOM to coschedule flows in inter-and-intra domains. Finally, a case study is performed on a typical industrial network. The experiment results demonstrate that the proposed MTOM can only use standard devices to achieve deterministic transmission of Profinet and TSN converged networks. Compared with previous asynchronous converged networks, the delay and jitter are reduced by 86% and 80% on average, respectively.

Microservice architecture has been widely adopted in large-scale applications. However, it also brings new challenges to ensuring reliable performance and maintenance due to the huge volume of data and complex dependencies of microservices. Existing approaches still suffer from the over-aggregation of data, interference from anomaly propagation, and ignoration of component differences. To solve these issues, this paper builds a root cause diagnosis framework at the operation granularity, named as MicroNet. Since operations are subfunctions of microservices, recorded as invocation purposes, we propose the operation-centric perspective, to realize fine-grained data aggregation and operation-level anomaly backtracking. We decompose the diagnosis task into four phases: dependency graph construction, anomaly detection, anomaly evaluation, and culprit location. To construct the invocation dependency accurately, we propose the concept of meta call, defined as the triple (caller, operation, callee), the smallest unit that can be aggregated. Based on the dependency graph, we quantify the operation’s abnormality by analyzing the operation execution process, to backtrack the propagated anomalies. Then, we customize a personalized PageRank algorithm to identify the root cause in which invocation latency and different invocation relationships are considered simultaneously. Our experimental evaluation on an open dataset shows that MicroNet can effectively locate root causes with 90% mean average precision, outperforming state-of-the-art methods.

Software Defined Network (SDN) has been considered as one of the most promising next-generation network solutions due to its network programmability. However, the upgrade from legacy IP network to pure SDN network is in general a gradual process. From a global perspective, a dynamic upgrade strategy should not only aim to pursue the local goal at each step, but also strive to optimize the final global solution when the upgrading process terminates. This raises three essential questions: which switches to upgrade, when to upgrade, and how to deploy controllers. Due to the interaction between the local goals at intermediate steps and the global goal at the final step, answering these questions altogether from a global perspective is challenging. In this paper, we study the dynamic SDN upgrade problem from a global perspective and answer these three questions altogether. We formulate the problem as a dynamic optimization problem that optimizes the global and local goals at the same time. We then propose two new formulations to combine the global and local goals, and two heuristic algorithms to solve them, respectively. We evaluate the proposed model and algorithms on realistic network topologies. The results show their feasibility and superiority.

Microservice architecture has been widely adopted for large-scale applications because of its benefits of scalability, flexibility, and reliability. However, due to the heterogeneity of system architecture and variable invocations between services, it is difficult to accurately diagnose the root causes of performance degradation in time. This paper proposes WinG, a system to pinpoint root causes. Firstly, since the characteristic of a system element is difficult to capture due to variable invocations between elements, WinG characterizes an element's status by a feature vector that includes all its invocation relationships. Secondly, WinG adopts a warping procedure to assess an element's anomaly severity based on its status deviation, to mitigate the interference of variable invocations. Thirdly, WinG groups heterogeneous elements with similar invocation characteristics to avoid the interference of diverse elements types. Finally, false alarms are filtered by the anomaly duration and frequency. Experimental evaluation results on the public dataset show that, with the above four methods, WinG can locate root causes with 87% precision, outperforming baseline methods. On average of 78 test cases, it achieves 34% precision improvement over the champion method of the competition.

The explosive development of short video applications has brought severe pressure on radio resources at hotspot areas. The features of short video recommendations-and-pushing techniques provide us an opportunity to relieve the radio resource pressure via wireless multicast: An edge server can be deployed at the base station, which receives short video lists recommended by remote video server and then pushes such mobile video services to local users through wireless multicast. In this paper, we study how to reshuffle the video lists received from remote server so as to facilitate wireless multicast to maximally reduce the required wireless resource while considering the fact that a user client can only buffer one short video for watching based on off-the-shelf short video APPs. We formulate the problem of video list reshuffling for minimizing the total wireless resources consumption as an integer programming problem. We design a Minimum degree of Freedom based Maximum Filling video reshuffling algorithm (MFMF) to address this problem. MFMF moves videos from the original video lists into same sized but reshuffled video lists in a greedy manner, once for a video, whose moving can satisfy the most reshuffled video lists, and if multiple such choices exist, selects the one having the least position options. This process continues until all the videos are moved. We deduce the computation complexity of MFMF. Numerical results demonstrate the significantly high performance of MFMF.

This paper investigates a new lane reservation problem with task merging that consists of optimally determining which lanes in a transportation network have to be reserved and designing reserved lane-based routes in the network for time-crucial transport tasks. Part of the tasks whose destinations are geographically close is merged to reduce the number of vehicles and transport costs. Reserved lanes can reduce the travel time of task vehicles passing through them, while they will generate negative impact on normal traffic, such as traffic delay to the vehicles on adjacent non-reserved lanes. The objective is to minimize the total negative impact of all reserved lanes. For this problem, two new integer linear programming (ILP) models are first developed. The complexity of the problem is proved to be NP-hard. Since commercial solver (like CPLEX) is time-consuming for solving it when the problem size increases, a fast and effective improved differential evolution algorithm (IDEA) is developed based on explored problem properties. Extensive experimental results for a real-life case and benchmark instances of up to 500 nodes in the network and 30 transport tasks show the favorable performance of the IDEA, as compared to CPLEX, differential evolution algorithm and genetic algorithm. Management insights are also drawn to support practical decision-making.

How to effectively organize various heterogeneous clients for effective model training has been a critical issue in federated learning. Existing algorithms in this aspect are all for single model training and are not suitable for parallel multi-model training due to the inefficient utilization of resources at the powerful clients. In this paper, we study the issue of multi-model training in federated learning. The objective is to effectively utilize the heterogeneous resources at clients for parallel multi-model training and therefore maximize the overall training efficiency while ensuring a certain fairness among individual models. For this purpose, we introduce a logarithmic function to characterize the relationship between the model training accuracy and the number of clients involved in the training based on measurement results. We accordingly formulate the multi-model training as an optimization problem to find an assignment to maximize the overall training efficiency while ensuring a log fairness among individual models. We design a Logarithmic Fairness based Multi-model Balancing algorithm (LFMB), which iteratively replaces the already assigned models with a not-assigned model at each client for improving the training efficiency, until no such improvement can be found. Numerical results demonstrate the significantly high performance of LFMB in terms of overall training efficiency and fairness.

As an application of the Internet of Things in the automotive field, Vehicular Ad-hoc NETworks (VANETs) are developed to facilitate traffic safety and traffic flow optimization. VANET consists of the communication network and the underlay road network. Due to the characteristics of open access, the communication network is vulnerable to various attacks, especially, the information pollution attack which is highly risky and concealed. Such an attack can lead the vehicles to react to the false messages from attackers, even cause failure cascades. However, the impact of the information pollution attack on the road network has not gained enough attention. To assess such impact, we build a traffic model with polluted information and assess the vulnerability of the road network under different pollution scenarios in terms of attack percentage, attack types, and road network topologies. Experimental results demonstrate that VANETs are vulnerable to the information pollution attack, and transportation performance drops by half when only 15% of edges are attacked in the worst case.

With the increase of transport demands, more pressure and challenges are being imparted into efficient transportation. As a conventional and direct congestion alleviation strategy, constructing new roads and lanes are increasingly restricted by limited land resources and high costs. Thus, making full use of existing transport network via appropriate management is critical to realize the sustainable development of transportation systems. As a flexible management strategy, lane reservation strategy has been widely adopted in real life. The reserved lanes can improve the efficiency of special transports, while they bring negative impact such as travel delay for general-purpose transports. In addition, the setting and operating of reserved lanes require a certain amount of cost. This paper proposes a new dual-objective integer linear programming model for optimally determining reserved lanes on a network for time-guaranteed special transports in order to simultaneously maximize the benefits and minimize the negative impact brought by reserved lanes, which incorporates road residual capacity and limited budget to the actual decision. Moreover, an iterative weighted sum-based method is proposed to solve it, in which a new relax-and-optimize algorithm is developed to exactly solve the single-objective optimization problems. Results of extensive numerical experiments show the effectiveness and efficiency of the proposed model and approach.

Free-viewpoint television enables users to view a scenario from arbitrary viewpoint as if they are physically in the scenario and can watch the scene freely. Thus, Free-viewpoint television can provide immersive experience of physical event broadcast, especially for large-scale live vocal concert broadcasts. However, huge bandwidth demand is a major challenge faced by free-viewpoint television broadcast since many streaming with different view angles are needed for users' selection. In this paper, we propose a scheme named P2P transcoder to realize free-viewpoint video streaming transmissions. It selects a subset of users to work as transcoders and these transcoders will produce video rates/angles that other remaining users request. Thus the total amount of traffic to deliver is greatly reduced and more saved bandwidth can be used to improve the video quality. We further build a model for achieving optimal bandwidth allocation using this scheme. Numerical results show that the proposed scheme can significantly improve the video quality as compared with existing work.

The explosive demand of online video watching brings huge bandwidth pressure to cellular networks. Efficient video caching is critical for providing high‐quality streaming Video‐on‐Demand (VoD) services to satisfy the rapid increasing demands of online video watching from mobile users. Traditional caching algorithms typically treat individual video files separately and they tend to keep the most popular video files in cache. However, in reality, one video typically corresponds to multiple different files (versions) with different sizes and also different video resolutions. Thus, caching of such files for one video leads to a lot of redundancy since one version of a video can be utilized to produce other versions of the video by using certain video coding techniques. Recently, fog computing pushes computing power to edge of network to reduce distance between service provider and users. In this paper, we take advantage of fog computing and deploy cache system at network edge. Specifically, we study transcoding based video caching in cellular networks where cache servers are deployed at the edge of cellular network for providing improved quality of online VoD services to mobile users. By using transcoding, a cached video can be used to convert to different low‐quality versions of the video as needed by different users in real time. We first formulate the transcoding based caching problem as integer linear programming problem. Then we propose a Transcoding based Caching Algorithm (TCA), which iteratively finds the placement leading to the maximal delay gain among all possible choices. We deduce the computational complexity of TCA. Simulation results demonstrate that TCA significantly outperforms traditional greedy caching algorithm with a decrease of up to 40% in terms of average delivery delay.

Video service has become a killer application for mobile terminals. For providing such services, most of the traffic is carried by the Dynamic Adaptive Streaming over HTTP (DASH) technique. The key to improve video quality perceived by users,i.e., Quality of Experience (QoE), is to effectively characterize it by using measured data. There have been many literatures that studied this issue. Some existing solutions use probe mechanism at client/server, which, however, are not applicable to network operator. Some other solutions, which aimed to predict QoE by deep packet parsing, cannot work properly as more and more video traffic is encrypted. In this paper, we propose a fog‐assisted real‐time QoE prediction scheme, which can predict the QoE of DASH‐supported video streaming using fog nodes. Neither client/server participations nor deep packet parsing at network equipment is needed, which makes this scheme easy to deploy. Experimental results show that this scheme can accurately detect QoE with high accuracy even when the video traffic is encrypted.

Aggregating multiple access interfaces of a client is a promising way to satisfy the high bandwidth demand by high-definition video streaming services. However, how to efficiently use such aggregated bandwidth to improve the reliability of timely fetching video contents needs to be further studied. In this paper, we propose a partial overlapping chunk based dual-path transmission scheme, which uses partial redundancy based transmissions to optimize the playback performance at the client side. Specifically, we schedule the transmissions of different sized chunks with partial overlapping according to the delivery capabilities of different paths. We then build an optimal model to compute the optimal overlapping ratio between the transmitted chunks to maximize the probability of timely fetching of video contents. Numerical results demonstrate that, our scheme can improve the probability of timely fetching video contents, by up to 19.3%, compared to the traditional scheme without chunk overlapping, while the incurred average transmission redundancy is below 14.4%.

Real-time video communications has been incorporated into many instant communication tools such as ichat, Skype, QQ, etc. Real-time video communications has low delivery delay requirement, which imposes great challenge to the provisioning of such services. To address this problem, in this paper, we propose a selective redundant transmission mechanism to support real-time streaming on multi-interface wireless terminals. This mechanism selectively duplicates some video frames according to the tightness of their lifetimes and further schedule their transmissions (or some of them) via neighbors' assistance. We build a model to select the optimal encoding rate and also the optimal per-frame copy number in order to maximize the peak signal noise ratio (PSNR) of video streaming service when maximal allowable total traffic rate is given. Numerical results show that the proposed mechanism can significantly improve the PSNR of real-time video streaming as compared with existing work.

Multiparty video conferencing (MPVC) facilitates real-time group interaction between users. While P2P is a natural delivery solution for MPVC, a peer often does not have enough bandwidth to deliver her video to all other peers in the conference. Recently, we have witnessed the popularity of one-view MPVC, where each user only watches full video of another user. One-view MPVC opens up the design space for P2P delivery. In this paper, we explore the feasibility of a pure P2P solution for one-view MPVC. We characterize the video source rate region achievable through video relays between peers. For both homogeneous and heterogeneous MPVC systems, we establish tight universal video rate lower bounds that are independent of the number of peers, the number of video sources, and the specific viewing relations between peers. We further propose, P2P video relay designs to approach the maximal video rate region. Through numerical simulations, we verified that the derived lower bounds are indeed tight bounds, and the proposed bandwidth allocation algorithm can achieve a close-to-optimal peer upload bandwidth utilization. Our results demonstrate that P2P is a promising solution for one-view MPVC. Insights obtained from our study can be used to guide the design of P2P MPVC systems.

A novel active queue management scheme for Internet DiffServ named class-guided random early discarding (CGRED) is proposed in this paper. The key enhancement of CGRED to standard RED is to introduce per-class states at routers and compute probability of discarding packets of different traffic classes adaptively based on the periodical measurements of actually consumed bandwidth of each classes, CGRED can maintain the assigned priority and pre-allocated bandwidth for each service class, eliminate the starvation of lower-priority classes, and ensure that each individual flows of a higher-priority class to obtain averaged rate not less than that of lower classes. Simulation results are presented to show that CGRED is simple, robust and scalable to be implemented.

The increasing demand of multimedia applications has spurred recent interest in end-to-end congestion control for real-time applications. However, all these mechanisms proposed so far try to provide smooth sending rate and be friendly to TCP. They have paid little attention to the characteristics that distinguish a real-time application from a non-real-time service. The works on equal and unequal protection multimedia transmission through FEC (forward error correction) encoding show that, there are many different structures in an encoded multimedia stream with different importance in the media restoration. In order to protect these different structures of different importance in the transport layer instead of in application layer through FEC a coupon based AIMD (CAIMD, coupon based additive-increase multiplicative-decrease) algorithm is proposed in this paper. In this algorithm, a coupon is attached to each sending packet to indicate the importance of the message carried in this packet. The network services packets differently according to the coupon value attached on them. In order to make CAIMD TCP friendly, the "linear" increase slope in CAIMD is dynamically changed with the importance of the message carried in a sending packet. Theoretical analysis and simulations carried in this paper show that CAIMD is capable of providing different protection in the packet level with a fine granularity, which opens a new means to optimize the QoS of media streams in network transportation. In additional, CAIMD is friendly to TCP flow as well.