Due to its reliance on fewer precise annotations, weakly supervised semantic segmentation (WSSS) techniques are in high demand in the field of remote sensing (RS) image processing. Despite mainstream WSSS approaches achieve remarkable dense prediction accuracies, they still face challenges such as insufficient pre-trained and ambiguous segment predictions. To this end, we propose to improve the accuracy of weakly supervised semantic segmentation by unifying the vision language foundation model and the Segment Anything Model (SAM). Specifically, we leverage a remote sensing vision language foundational model, RemoteCLIP, to provide sufficient pre-trained knowledge. Subsequently, we employ a decoder to transform the high-level feature representations extracted by RemoteCLIP into the segmentation predictions. Then, we introduce a multi-prompt fusion (MPF) approach via the Segment Anything Model (SAM) to obtain high-quality segment results with well-defined boundaries. To the best of our knowledge, this is the first study to apply a unified framework of foundation model and Segment Anything Model for RS WSSS. Experimental results demonstrate that our method achieves remarkable performance across three remote sensing datasets.

Temporal Knowledge Graph (TKG) reasoning aims to predict missing facts using historical TKG data. Existing methods often struggle to effectively capture crucial semantics within long-term historical data and typically overlook dynamic semantic associations among relations, thereby limiting reasoning accuracy. To address these challenges, we propose the TKG Reasoning with Long- and Short-Term Dependencies and Relational Semantics Model (LSTR). LSTR employs an encoder-decoder architecture with a dual-layer entity encoding structure comprising short-term and long-term entity encodings. The short-term entity encoding adaptively highlights essential subgraphs through a subgraph-aware attention mechanism. The long-term entity encoding incorporates a query association layer and a long-term historical dependency layer, constructing multi-hop association graphs and long-term temporal-aware graphs, respectively, to capture entities’ evolving trends effectively. Furthermore, we introduce a relation association graph that explicitly models dynamic semantic associations among relations via relational graph convolution networks, significantly enhancing the representational power of relation embeddings. Extensive experiments on four benchmark datasets demonstrate the superior performance and effectiveness of LSTR.

In this study, we tackle the path-tracking problem of a nonlinear four-wheel steering vehicle dynamics model subject to model mismatches and propose a model predictive control (MPC) algorithm based on an event-triggered mechanism (ET -MPC). The goal is to maintain closed-loop control performance while reducing the computational and communication burdens of traditional MPC. We introduce an ET -MPC framework utilizing a model-free reinforcement learning agent with proximal policy optimization (PPO). This agent interacts with the MPC system, progressively learning to determine the optimal event-triggered mechanism. To enhance exploration and training efficiency, we incorporate the Long Short-Term Memory (LSTM) technique into PPO. Experimental results show that the proposed ET -MPC framework, combined with reinforcement learning for reward optimization, demonstrates superior overall performance in path-tracking control of four-wheel steering vehicles.

Knowledge Graph Completion (KGC) endeavors to use existing knowledge graph data for predicting missing elements in triples. Recently, due to the efficiency of graph neural networks (GNNs) in capturing topological structures and the effectiveness of text descriptions in supplementing semantic information, numerous models integrating graph structures and entity descriptions have emerged. However, these approaches typically focus on aggregating neighboring feature information and overlook mining hidden structural information. Furthermore, they often append textual descriptions to entities independently without considering the semantics within specific relations. Hence, we propose the SRE-KGC model to address these challenges. First, while aggregating neighbors, we analyze and mine the hidden structure in the neighborhood from the perspective of entities and relations; then we introduced a dual-layer attention mechanism to extract the most pertinent textual information towards relations from both the relational semantic level and the neighbor semantic level respectively; finally, the two learned features are fused and sent to the decoder for scoring. Experiments demonstrate that our model delivers superior performance.

Multimodal sentiment analysis integrates various modalities of information to collectively inform decision-making processes. Previous studies often treat different modal features equally or emphasize textual information as the primary consideration. However, when the modalities in the sample contain different sentiment information, these methods may not be able to effectively deal with this situation. To solve this problem, we propose a multimodal sentiment analysis model focusing on each modality (MFM). In this paper, we separately integrate each modality as a primary modality interacting with other secondary modal information so that each modality can play a leading role. In addition, we use shared mask in modal interaction to capture important information in the secondary modality related to the primary modality, and improve the effectiveness of the information interaction process. The model is evaluated against baseline models using the MOSI and MOSEI multimodal sentiment analysis datasets. The experimental results show that the model achieves better performance, thereby validating its effectiveness in multimodal sentiment analysis tasks.

Cleanliness of city streets has an important impact on city environment and public health. Conventional street cleaning methods involve street sweepers going to many spots and manually confirming if the street needs to be clean. However, this method takes a substantial amount of manual operations for detection and assessment of street’s cleanliness which leads to a high cost for cities. Using pervasive mobile devices and AI technology, it is now possible to develop smart edge-based service system for monitoring and detecting the cleanliness of streets at scale. This paper explores an important aspect of cities — how to automatically analyze street imagery to understand the level of street litter. A vehicle (i.e. trash truck) equipped with smart edge station and cameras is used to collect and process street images in real time. A deep learning model is developed to detect, classify and analyze the diverse types of street litters such as tree branches, leaves, bottles and so on. In addition, two case studies are reported to show its strong potential and effectiveness in smart city systems.

During its life cycle, data has to go through different stages, from generation, storage, query, various processing, to deletion or archiving. Meanwhile, all these evolutions can be recorded by data provenance, which can be used for data deduction and credibility verification. Starting from the status of data application and processing, current problems exist in data management have been raised in this paper. Thus, in order to solve the problems, provenance information on space and time was discussed. Then, based on Open Provenance Model (Referred to as OPM), analysis and research were conducted, a more complete description model has been given, which contains time and space provenance information. With this model, it will be more convenient for data query and credibility verification in the following data management.

The existing research of distributed provenance storage is at the preliminary stage. Most of the researchers are endeavor to improve the query efficiency of storage solution, whereas, very few people pay attention to the matching problem of provenance and data and the waste of storage space. Two aspects on the problem of provenance storage were proposed in this paper. First one is the relationship between provenance and the data itself and the second is internal storage of provenance. Based on these two aspects, a two level index storage model of provenance was put forward here. In the first level, according to the current situation that provenance does not match the object it described, a dictionary table index method was proposed. In the second level, aiming at the internal storage of provenance, three dimensional indexes were established to store provenance. Coding was designed at the same time, variable long integer coding method was used to solve the problem of the storage space waste. Optimization of provenance storage was implemented in this model, therefore, requirements of different user's query were met.