Text-to-image generation applications based on diffusion models face substantial challenges in computational efficiency and latency, particularly in time-sensitive scenarios, due to the inherently iterative denoising process. Although approximate caching techniques can reduce denoising iterations by reusing intermediate states of diffusion models, existing approaches fail to adequately capture user request behaviors. It is observed that users tend to issue a large number of prompt requests within short time intervals (bursty patterns), and that prompts from the same user often exhibit high similarity over short time periods (temporal locality). In this work, we define and formalize the Intermediate State Selection (ISS) problem to minimize denoising iterations. We further prove the NP-hardness of the ISS problem via a polynomial-time reduction from the Dominating Set problem. We then exploit both characteristics of prompt requests and present EdgeDiffusion, a novel edge-cloud cooperative framework in which the cloud retains image generation, while prompts caching and intermediate state selection are offloaded to edge servers. Specifically, we design an ISS algorithm that optimizes state reuse by leveraging temporal locality and an adaptive caching strategy tailored to bursty patterns. Experimental results on real-world datasets demonstrate that EdgeDiffusion achieves 18.3%-77.3% computational savings over baseline strategies (NIRVANA, qLRU-AC, LRU and LFU), while maintaining 98% quality of images.

The proliferation of Internet of Things applications has driven rapid Mobile Edge Computing (MEC) systems development by various Edge Service Providers (ESPs), creating a competitive computing market. Handling all received tasks from each ESP individually can significantly degrade the service performance of the MEC system. To enhance flexibility in network workload and service coverage, unmanned aerial vehicles (UAVs) have been employed in the MEC system. This paper proposes a potential game based trustful task offloading scheme for the multiple EPSs scenario in UAV-assisted MEC. Specifically, we formulate the Multiple ESPs Task Offloading (METO) problem into a potential game and prove the existence of Nash Equilibrium (NE). To guarantee the security of resource trading among different EPSs, we propose a blockchain-based sharing mechanism that converges to NE. Additionally, a reputation smart contract assesses ESPs' Quality of Service (QoS), influencing task allocation. Extensive simulations show our approach outperforms traditional baselines in maximizing each ESP's utility.

Backscatter communication, which enables a pas-sive backscatter device to transmit information to a reader using incident radio-frequency signals, is a promising technology for low-power Internet of Things. To improve the backscatter communication range, we consider a multistatic backscatter communication system in which multiple dislocated readers illuminate a tag simultaneously, thus strengthening the received signals. However, this system suffers from strong direct co-channel interference between different readers. To tackle this challenge, we jointly design the tag's transmit waveform and the readers' optimal detectors to cancel out the interference first and then recover the tag information. Simulations results verify that the proposed transceiver achieves better bit-error-rate performance without error floor than the benchmark.

Convolutional neural networks (CNNs) have met great success in abstractive summarization, but they cannot effectively generate summaries of desired lengths.Because generated summaries are used in difference scenarios which may have space or length constraints, the ability to control the summary length in abstractive summarization is an important problem.In this paper, we propose an approach to constrain the summary length by extending a convolutional sequence to sequence model.The results show that this approach generates high-quality summaries with user defined length, and outperforms the baselines consistently in terms of ROUGE score, length variations and semantic similarity.

Many existing systems for analyzing and summarizing customer reviews about products or service are based on a number of prominent review aspects.Conventionally, the prominent review aspects of a product type are determined manually.This costly approach cannot scale to large and cross-domain services such as Amazon.com,Taobao.com or Yelp.comwhere there are a large number of product types and new products emerge almost everyday.In this paper, we propose a novel framework, for extracting the most prominent aspects of a given product type from textual reviews.The proposed framework, ExtRA, extracts K most prominent aspect terms or phrases which do not overlap semantically automatically without supervision.Extensive experiments show that ExtRA is effective and achieves the state-of-the-art performance on a dataset consisting of different product types.