In modern urban environments, camera networks generate massive amounts of operational footage -- reaching petabytes each day -- making scalable video analytics essential for efficient processing. Many existing approaches adopt an SQL-based paradigm for querying such large-scale video databases; however, this constrains queries to rigid patterns with predefined semantic categories, significantly limiting analytical flexibility. In this work, we explore a language-driven video analytics paradigm aimed at enabling flexible and efficient querying of high-volume video data driven by natural language. Particularly, we build Lava, a system that accepts natural language queries and retrieves traffic targets across multiple levels of granularity and arbitrary categories. Lava comprises three main components: 1) a multi-armed bandit-based efficient sampling method for video segment-level localization; 2) a video-specific open-world detection module for object-level retrieval; and 3) a long-term object trajectory extraction scheme for temporal object association, yielding complete trajectories for object-of-interests. To support comprehensive evaluation, we further develop a novel benchmark by providing diverse, semantically rich natural language predicates and fine-grained annotations for multiple videos. Experiments on this benchmark demonstrate that Lava improves F1-scores for selection queries by 14% reduces MPAE for aggregation queries by 0.39, and achieves top-k precision of 86% while processing videos 9.6x faster than the most accurate baseline. Our code and dataset are available at https://github.com/yuyanrui/LAVA.

Reservoir characterization (identification and parameter estimation) is critical for gas hydrate exploration and development. While machine learning techniques excel at capturing complex relationships in well-logging reservoir characterization, existing research mainly focuses on end-to-end supervised learning approaches relying on costly labeled data and lacking multi-task learning capabilities. In this paper, we introduce a self-supervised learning (SSL) framework to learn general representations of large-scale unlabeled well-logging intervals from sandy, silty, and clayey hydrate reservoirs via masked autoencoders (WirMAE). By incorporating channel-based attention mechanisms and a masked reconstruction pre-training strategy for variable tokens, WirMAE effectively extracts intrinsic multivariate correlations within logging data, enabling the generation of various missing log curves. The model is validated on data from globally distributed hydrate with diverse accumulation patterns. Compared to supervised deep learning methods and classical machine learning models, Fine-turned WirMAE achieves superior reservoir identification accuracy (average F1-score: 0.864) using only 1% labeled data in complex geological settings. Combined with domain expertise, WirMAE yields precise estimation of key reservoir parameters such as hydrate saturation and permeability, outperforming conventional petrophysical methods. Additionally, embedding visualizations and attention analyses reveal the inner workings of the model and its consistency with expert-driven geological interpretations. Our findings highlight the potential of SSL for advancing more accurate and transparent intelligent reservoir characterization using well-log data, indicating that the application of WirMAE could be extended to broader hydrocarbon reservoirs in the future.

Distributed Acoustic Sensing (DAS) is an emerging vibration collection technology with advantages such as low cost, high-density sampling, and high sensitivity. It utilizes regional unlit fiber-optic telecommunication infrastructure (dark fiber) in the urban underground to record real-time environmental signals. How to identify vehicle signals, classify vehicle types, and estimate vehicle speeds from DAS signals has significant potential for the development of intelligent urban transportation. Addressing the challenges of high-noise environments and dense traffic, we develop an end-to-end two-stage deep learning process to identify and classify various vehicle signals in urban traffic rapidly. First, we propose the CarDenoiseNet network, based on Generative Adversarial Networks (GAN) and contrastive learning, to denoise and enhance the weak signal of DAS data. Then, the YOLOv8 segmentation model is employed to segment and classify the vehicle signal. Finally, the vehicle speeds are estimated using the segmented vehicle trajectory time and location information. We use the urban DAS field data recorded in the downtown area of Changchun to test the proposed workflow. The test result has good generalization and over 90% accuracy in identifying different vehicle types and speeds in high-density traffic environments. Moreover, transfer learning successfully applies the model to other datasets, proving its excellent generalization ability. Additionally, statistical analysis of traffic flow and speed trends provides technical references for alleviating urban traffic congestion, reducing traffic accidents, and enhancing the intelligence level of urban traffic management.

To improve the comfort of surgeons during the use of minimally invasive surgical robots and reduce occupational hazards, the precise puncture robot for vital organs was selected as the research subject. Based on the theory and methods of human-machine ergonomics, combined with JACK software, a simulation analysis was conducted on the comfort, working posture, and visibility of surgeons using existing related products. Design improvements were made based on the analysis results. By comparing and evaluating the simulation experimental data before and after the design changes, it was found that the improved solution at the human-machine ergonomics level helps enhance the comfort of users during the operation process and reduce the risk of occupational diseases among the medical community. This design can serve as a guide and reference for related fields.

Data aggregation based on machine learning (ML), in mobile edge computing, allows participants to send ephemeral parameter updates of local ML on their private data instead of the exact data to the untrusted aggregator. However, it still enables the untrusted aggregator to reconstruct participants’ private data, although parameter updates contain significantly less information than the private data. Existing work either incurs extremely high overhead or ignores malicious participants dropping out. The latest research deals with the dropouts with desirable cost, but it is vulnerable to malformed message attacks. To this end, we focus on the data aggregation based on ML in a practical setting where malicious participants may send malformed parameter updates to perturb the total parameter updates learned by the aggregator. Moreover, malicious participants may drop out and collude with other participants or the untrusted aggregator. In such a scenario, we propose a scheme named DAML , which to the best of our knowledge is the first attempt toward verifying participants’ submissions in data aggregation based on ML. The main idea is to validate participants’ submissions via SSVP, a novel secret-shared verification protocol, and then aggregate participants’ parameter updates using SDA, a secure data aggregation protocol. Simulation results demonstrate that DAML can protect participants’ data privacy with preferable overhead.

In this paper the millimeter wave (mmWave) single carrier transmission systems operating on wideband channels with frequency selectivity are considered. We propose two frequency selective hybrid precoding solutions, namely time delay compensation (TDC) based time domain equalization (TDE) and phase shift compensation (PSC) based frequency domain equalization (FDE). First, we implement a sectorized beamforming in RF domain to compensate the large path-loss at mmWave range and reduce the intra cluster delay spread. Here, the RF precoder is assumed to be frequency flat during one block transmission due to the independence of RF precoder design on the time delay spread. Subsequently, adopting TDC for TDE and PSC for FDE in digital baseband to reduce frequency selectivity, the baseband precoder is designed to maximize the achievable capacity. Finally, numerical results indicate that the proposed equalization algorithms can achieve a comparable performance with respect to the flat channel case.

The fading channel in mmWave communications has a very high chance to be frequency selective owing to the large transmission bandwidth. To overcome the frequency selectivity we propose a novel method, namely RF beamforming with time delay compensation. Simultaneously, baseband precoder at the transmitter side and the combiner at the receiver side are employed to maximize the capacity of the system. For effectively flattening the channel by beamforming on both sides, it is necessary to separate the individual rays. Motivated by achieving the best case beam separation, a signal-to-interference ratio (SIR) constrained capacity maximization algorithm is proposed in this paper. We also study the influence of the parameters such as the number of antennas at the transmitter side and the SIR threshold on the capacity and the system performance. Finally, the proposed method is validated by providing numerical examples by means of simulation.