Text-Image-to-Video (TI2V) generation can be sensitive to compression distortions in the reference image. Small pixel-space artifacts may be amplified through iterative generation, leading to temporal inconsistency and semantic drift. One contributing factor is that conventional codecs are optimized for human-oriented pixel fidelity, without explicitly accounting for the distribution shift induced in the generator’s conditioning latents. Motivated by this mismatch, we propose a generation-centric latent-domain coding framework that compresses generator-facing deep latents instead of raw pixels, and jointly optimizes rate and latent-space distortion to preserve conditioning statistics. Experiments across four bitrate levels show that our approach mitigates distortion accumulation over iterations. It improves temporal coherence and content fidelity for both short and extended sequences, consistently reducing the pixel-level discrepancy relative to videos generated with lossless conditioning.

With the development of artificial intelligence, many visual detection methods have been applied to power systems. However, diverse tasks cause existing solutions to face low detection accuracy and resource constraints in edge power scenarios. In this paper, we propose a low-rank enhanced lightweight multimodal LLM framework for efficient edge power visual detection. The framework can effectively balance the model performance with the resource constraints of edge deployment by introducing multimodal LLM and fusing data augmentation strategies and low-rank optimization. First, to enhance the adaptability of diverse visual detection tasks, we design a diverse data augmentation strategy for multimodal LLM to solve the problem of insufficient power scene data. Then, we reduce the number of training parameters by a low-rank optimization technique to enable the model to run efficiently on resource-limited edge devices. Experimental results demonstrate that our proposed framework can improve the accuracy by more than 3% on average under different power visual detection tasks. It can also save 76.49% of graphics memory consumption and accelerate training by 81.88%.

Large language models have demonstrated impressive universal capabilities across a wide range of open-ended tasks and have extended their utility to encompass multi-modal conversations. However, existing methods encounter challenges in effectively handling both image and video understanding, particularly with limited visual tokens. In this work, we introduce Chat-UniVi, a Unified Vision-language model capable of comprehending and engaging in conver-sations involving images and videos through a unified visual representation. Specifically, we employ a set of dynamic visual tokens to uniformly represent images and videos. This representation framework empowers the model to ef-ficiently utilize a limited number of visual tokens to simul-taneously capture the spatial details necessary for images and the comprehensive temporal relationship required for videos. Moreover, we leverage a multi-scale representation, enabling the model to perceive both high-level seman-tic concepts and low-level visual details. Notably, Chat-UniVi is trained on a mixed dataset containing both images and videos, allowing direct application to tasks involving both mediums without requiring any modifications. Exten-sive experimental results demonstrate that Chat- UniVi con-sistently outperforms even existing methods exclusively de-signed for either images or videos. Code is available at https://github.com/PKu-Yuan Group/Chat-UniVi.

Many real-world applications require the prediction of long sequence time-series, such as electricity consumption planning. Long sequence time-series forecasting (LSTF) demands a high prediction capacity of the model, which is the ability to capture precise long-range dependency coupling between output and input efficiently. Recent studies have shown the potential of Transformer to increase the prediction capacity. However, there are several severe issues with Transformer that prevent it from being directly applicable to LSTF, including quadratic time complexity, high memory usage, and inherent limitation of the encoder-decoder architecture. To address these issues, we design an efficient transformer-based model for LSTF, named Informer, with three distinctive characteristics: (i) a ProbSparse self-attention mechanism, which achieves O(L log L) in time complexity and memory usage, and has comparable performance on sequences' dependency alignment. (ii) the self-attention distilling highlights dominating attention by halving cascading layer input, and efficiently handles extreme long input sequences. (iii) the generative style decoder, while conceptually simple, predicts the long time-series sequences at one forward operation rather than a step-by-step way, which drastically improves the inference speed of long-sequence predictions. Extensive experiments on four large-scale datasets demonstrate that Informer significantly outperforms existing methods and provides a new solution to the LSTF problem.

With the development of Service-Oriented technologies, the amount of Web services grows rapidly. QoS-Aware Web service recommendation can help service users to design more efficient service-oriented systems. However, existing methods assume the QoS information for service users are all known and accurate, but in real case, there are always many missing QoS values in history records, which increase the difficulty of the missing QoS value prediction. By considering the user-service-time three dimension context information, we study a Temporal QoS-Aware Web Service Prediction Framework which aims to recommend best candidates to service user's requirements and meanwhile improve the QoS prediction accuracy. One major challenge is that how to deal with the high dimension, sparse QoS value data. Tensor which is known as multi-way array provides a natural representation for such QoS value data. Therefore, we formalize this problem as a tensor factorization model and propose a Tucker Decomposition (TD) algorithm which is able to deal with the triadic relations of user-service-time model. Extensive experiments are conducted based on our real-world QoS dataset collected on Planet-Lab, comprised of service invocation response-time values from 408 users on 5,473 Web services at 56 time periods. Comprehensive empirical studies demonstrate that our approach is more accuracy than other approaches and achieves 100X to 1000X memory space reduction.

With the rapid growth of Web Service in the past decade, the issue of QoS-aware Web service recommendation is becoming more and more critical. Since the Web service QoS information collection work requires much time and effort, and is sometimes even impractical, the service QoS value is usually missing. There are some work to predict the missing QoS value using traditional collaborative filtering methods based on user-service static model. However, the QoS value is highly related to the invocation context (e.g., QoS value are various at different time). By considering the third dynamic context information, a Temporal QoS-aware Web Service Recommendation Framework is presented to predict missing QoS value under various temporal context. Further, we formalize this problem as a generalized tensor factorization model and propose a Non-negative Tensor Factorization (NTF) algorithm which is able to deal with the triadic relations of user-service-time model. Extensive experiments are conducted based on our real-world Web service QoS dataset collected on Planet-Lab, which is comprised of service invocation response-time and throughput value from 343 users on 5817 Web services at 32 time periods. The comprehensive experimental analysis shows that our approach achieves better prediction accuracy than other approaches.