In the Text-to-Image (T2I) generation task, precisely constructed prompts are crucial for fully leveraging the generative model’s capabilities and obtaining preferable image outputs. However, well-performed prompts are usually model-specific, requiring customization for each T2I model. Existing prompt optimization methods typically rely on inefficient searches, heuristic algorithms, or model-specific training, which are computationally expensive and inflexible. To address these challenges, BanditRewriter is proposed, which leverages large language model to adaptively optimize prompts for different T2I models. By incorporating upper confidence bounds and beam search, BanditRewriter achieves an effective exploration-exploitation trade-off to efficiently develop model-preferred prompt rewriting strategies without model training. Experiments on Stable Diffusion 2.1 and DALL·E 3 demonstrate that BanditRewriter can enhance text-image alignment with an 8.76% improvement in Image Reward while boosting generated image quality through a 3.56% increase in Aesthetics score. Code available at https://github.com/Cosmos-FFF8E7/BanditRewriter.

Prompt learning has been proven to be quite an effective technique for adapting large visual-language models (LVLMs) to downstream tasks via few-shot learning. Early methods often rely on a single prompt, which is insufficient for comprehensively representing a class. Subsequent efforts have explored multiple prompts to further enhance the adaptability and performance of LVLMs. However, these methods primarily focus on learning a set of more discriminative prompts, overlooking their generalizability. To learn prompts that are more balanced in both generalization and discrimination, we propose a novel multi-prompt learning approach, Masked Multi-Prompt Learning with Knowledge Mixing (dubbed TriMPL), which contains two pivotal mechanisms: (1) knowledge mixing to enhance the generalization of each individual prompt and (2) prompt masking to boost the prompt set's overall robustness. With respect to knowledge mixing, it progressively injects the general knowledge of handcrafted prompts into each learnable prompt at different Transformer encoding stages. While for prompt masking, of which the critical insight is that an optimal set of prompts should exhibit independence, allowing accurate predictions with just a subset of prompts. During training, TriMPL randomly masks some prompts to enhance the overall robustness of the learned prompts for image classification. We evaluate the effectiveness of TriMPL under three settings: (1) base-to-new generalization, (2) cross-dataset transfer, and (3) domain generalization. Extensive experiments demonstrate that TriMPL is capable of learning a set of effective prompts, achieving superior performance to quite a few state-of-the-art competitors.

A critical challenge to image-text retrieval is how to learn accurate correspondences between images and texts. Most existing methods mainly focus on coarse-grained correspondences based on co-occurrences of semantic objects, while failing to distinguish the fine-grained local correspondences. In this paper, we propose a novel Scene Graph based Fusion Network (dubbed SGFN), which enhances the images’/texts’ features through intra-and cross-modal fusion for image-text retrieval. To be specific, we design an intra-modal hierarchical attention fusion to incorporate semantic contexts, such as objects, attributes, and relationships, into images’/texts’ feature vectors via scene graphs, and a cross-modal attention fusion to combine the contextual semantics and local fusion via contextual vectors. Extensive experiments on public datasets Flickr30K and MSCOCO show that our SGFN performs better than quite a few SOTA image-text retrieval methods.

Inter destination video synchronisation (IDVS) is a key technology in emerging interactive multimedia applications. It is essential to ensure the synchronous experiences of users in such applications. However, one inevitable barrier for IDVS is the packet transfer delay differences (PTDDs) among different destinations. Existing researches have tried to eliminate the side effects of PTDD passively and schedule the video packets in a ‘back‐to‐back’ fashion. In this paper, the authors propose to proactively leverage such differences to design a transmission scheme that enhances video quality while ensuring the synchronous arrival of packets. Based on the network measurements from a real multimedia conferencing system and the Planetlab, they find that the PTDD is between the ranges of 100–250 ms. Motivated by this observation, they propose a scheme named asynchronous departure for synchronous arrival (ADSA), which inserts intervals between consecutive packets according to the synchronisation reference. To prove the superiority of ADSA, they carry out analysis based on Gilbert loss model and continuous time Markov chain. They conduct performance evaluation through emulations in Exata and experimental results show that ADSA improves the video peak signal‐to‐noise ratio by up to 9.1 and 6.9 dB compared with existing latest and earliest schemes, respectively.

The massive proliferation of wireless infrastructures with complementary characteristics prompts the bandwidth aggregation for Concurrent Multipath Transfer (CMT) over heterogeneous access networks. Stream Control Transmission Protocol (SCTP) is the standard transport-layer solution to enable CMT in multihomed communication environments. However, delivering high-quality streaming video with the existing CMT solutions still remains problematic due to the stringent quality of service (QoS) requirements and path asymmetry in heterogeneous wireless networks. In this paper, we advance the state of the art by introducing video distortion into the decision process of multipath data transfer. The proposed distortion-aware concurrent multipath transfer (CMT-DA) solution includes three phases: 1) per-path status estimation and congestion control; 2) quality-optimal video flow rate allocation; 3) delay and loss controlled data retransmission. The term `flow rate allocation' indicates dynamically picking appropriate access networks and assigning the transmission rates. We analytically formulate the data distribution over multiple communication paths to minimize the end-to-end video distortion and derive the solution based on the utility maximization theory. The performance of the proposed CMT-DA is evaluated through extensive semi-physical emulations in Exata involving H.264 video streaming. Experimental results show that CMT-DA outperforms the reference schemes in terms of video peak signal-to-noise ratio (PSNR), goodput, and inter-packet delay.

Load distribution is a key research issue in deploying the limited network resources available to support traffic transmissions. Developing an effective solution is critical for enhancing traffic performance and network utilization. In this paper, we investigate the problem of load distribution for real-time traffic over multipath networks. Due to the path diversity and unreliability in heterogeneous overlay networks, large end-to-end delay and consecutive packet losses can significantly degrade the traffic flow's goodput, whereas existing studies mainlyfocus on the delay or throughput performance. To address the challenging problems, we propose a Goodput-Aware Load distribuTiON (GALTON) model that includes three phases: (1) path status estimation to accurately sense the quality of each transport link, (2) flow rate assignment to optimize the aggregate goodput of input traffic, and (3) deadline-constrained packet interleaving to mitigate consecutive losses. We present a mathematical formulation for multipath load distribution and derive the solution based on utility theory. The performance of the proposed model is evaluated through semi-physical emulations in Exata involving both real Internet traffic traces and H.264 video streaming. Experimental results show that GALTON outperforms existing traffic distribution models in terms of goodput, video Peak Signal-to-Noise Ratio (PSNR), end-to-end delay, and aggregate loss rate.

The technology of Cloud Computing has shown its great power in consolidating and integrating compute resources for higher utilizing efficiency. With the cloud computing, the organizations or companies can own their hardware and software infrastructures easily. Because of the diversity of the operating systems and applications, it is very difficult or even impossible for administrator to deploy a large number of virtual machines within a short time manually. In this paper, we propose an automatic deployment mechanism on Open Stack, a popular Cloud Computing platform. This proposed system supports the automatic deployment service at both operating system level and application level. We also develop a dashboard to facilitate users operations. Without professional knowledge of cloud, Users also can deploy their systems and the applications expediently.

Service platforms have disadvantages such as they have long construction periods, low resource utilizations and isolated constructions. Migrating service platforms into clouds can solve these problems. The scalability is an important characteristic of service clouds. With the scalability, the service cloud can offer on-demand capacities to different services. In order to achieve the scalability, we need to know when and how to scale virtual resources assigned to different services. In this paper, a linear regression model is used to predict the workload. Based on this predicted workload, an auto-scaling mechanism is proposed to scale virtual resources at different resource levels in service clouds. The automatic scaling mechanism combines the real-time scaling and the pre-scaling. Finally experimental results are provided to demonstrate that our approach can satisfy the user SLA while keeping scaling costs low.

As cloud-based services become more numerous and dynamic, service deployment becomes more and more challenging. For the reason of economic, the service deployment strategy is crucial for maximize the system utility and profit gained. In this paper, considered both the cost and quality, the service deployment in cloud is modeled as a congestion game, where every service has several components is considered as a player and his strategy is to select the subset of resources to deploy his components. Based on the congestion game, a game theoretic method is proposed to optimize both the overall cost and quality. To solve the congestion game, efficient algorithms are proposed to achieve the equilibrium in polynomial time.

With the rapid development of Service-Oriented Architecture (SOA), more and more enterprise applications are built on the Enterprise Services Bus (ESB). However, ESB would not be able to deal with the exchange of data among multiple ESBs. In this paper, we propose an Enterprise Cloud Bus (ECB) framework that combines the WS-Notification with cloud computing. On the basis of WS-Notification specification and cloud queue model, we design and implement the Subscribe Request Queue cloud service model that supports real-time communication among distributed ESB, and supports running non-Service components or composition services deployed on distributed ESB or in different business process. On the basis of cloud platform, our framework shows better scalability and reliability.

Spectrum sensing is an essential mechanism in a cognitive radio network. However, the detection performance will be greatly degraded when few untrusted secondary users exist which can be identified as 'Always Yes' users and 'Always No' users. In this paper, we propose an energy detector with double thresholds combined with revised data fusion rules to find these untrusted users and counteract their malicious effects. Probabilities of detection and false alarm of three kinds of revised data fusion rules are derived. Compared with the conventional cooperative methods, our method can get a better detection performance.