Recent years have witnessed growing scholarly interest in binary post-training quantization (PTQ) techniques for large language models (LLMs). While state-of-the-art (SOTA) binary quantization methods significantly reduce memory footprint and computational demands, they introduce additional memory overhead beyond binary weight tensors to mitigate performance degradation. Moreover, binary LLMs still suffer from substantial accuracy loss. To address these limitations, we propose MemeBQ, a novel binary PTQ framework for LLMs that reduces the memory overhead of auxiliary flag bitmaps in existing binary quantization methods. Specifically, we first design a greedy row clustering method, which leverages the similarity between the row vectors of weights to partition the weight rows into different groups. By sharing the common flag bitmap within each row group, we significantly mitigate the memory overhead associated with flag bitmaps. Besides, to improve the performance of binary LLMs, we propose a novel weight splitting method for each row group of weights, which determines the flag bitmap's values in a fine-grained way. Extensive experiments on OPT, Llama-2, and Llama-3 models demonstrate that MemeBQ reduces 50% extra memory demand while achieving comparable accuracy compared with current SOTA methods. Alternatively, MemeBQ outperforms SOTA binary quantization methods up to 7% with the same extra bits on reasoning benchmarks.

Uncertainty has always been a significant yet unresolved issue in geography. Uncertainty in spatial optimization models, particularly in coverage maximization problems, stems from the limitations of binary coverage metrics for areal-based demand units. Traditional models classify units as either fully covered or not, ignoring partial coverage scenarios. This binary approach creates discrepancies between modeled and actual service demand, as even marginally covered units may count as fully serviced. The propagation of such errors is amplified by variations in the unit size and coverage thresholds, leading to suboptimal solutions in urban planning tasks such as urban greenway planning. To address these limitations, we proposed a deep reinforcement learning (DRL)-driven optimization framework that quantifies the optimality gap caused by spatially induced uncertainty and provides superior solutions. By employing a three-module structure that incorporates a graph neural network (GNN)-driven agent for data representation, a policy and a value network for decision-making, and a stochastic environment for simulating individual-level demand distributions, our model dynamically adapts to spatial uncertainty and generates robust solutions. We tested our framework in solving the maximal covering location problem for lines (MCLP-Line) for urban greenway route planning. The experimental results demonstrated that traditional mixed integer linear programming (MILP) methods yielded highly variable outcomes with different coverage metrics, whereas our proposed model consistently outperformed MILP across diverse scenarios. Our findings underscore the potential of modern AI-driven approaches in mitigating spatial uncertainty and achieving optimized solutions in geographic applications.

Although a couple of robust cooperative controllers have been proposed to achieve the multi-tug towing of unactuated offshore platform with uncertainties and unknown disturbances, the convergence performance of towing control system is unsatisfactory. To surmount this challenge, this research aims to improve the convergence performance of multi-tug towing system based on fixed-time stability theory. Initially, a fixed-time extended state observer (FxESO) is designed to estimate the uncertainties and disturbances. Based on the FxESO, a fixed-time virtual controller is proposed to acquire desired drag force for the offshore platform to track the expected trajectory. Subsequently, the desired drag force is allocated to the desired towline tensions of tugs by the quadratic programming algorithm, and the corresponding desired towline length is calculated through the towline catenary model. Then, the desired position of each tug is obtained based on the desired towline length and the current position information of the platform and tug. Based on the desired positions of tugs, a fixed-time distributed cooperative controller is designed to achieve the multi-tug towing of offshore platform within the fixed time. Finally, simulations and comparisons have verified the progressiveness and effectiveness of the proposed method.

The appearance information of the target has been used as the only tracking cue for most trackers to locate the target in the video. However, when the surrounding environment changes drastically or there are similar interference targets, it usually causes target drift. We propose a tracker named MRTrack, a transformer-based spatiotemporal information de coupling network architecture to enhance the target tracking capability. We designed two training schemes to explore the effective integration of motion cues derived from optical flow with representation information. The first approach integrates the target's motion and representation information for training. The second scheme is a step-by-step training, where the target motion information is first learned, and the learned model is used for representation learning. We compare the two training methods on five generic tracking datasets. The experiment results indicate that the first training approach can better integrate motion and representation information, leading to more precise tracking results for MRTrack compared to solely relying on the appearance model. In addition, optical flow cues are used only in the training phase to guide the tracker in understanding motion information, and no additional cost is incurred during tracking inference.

Siamese trackers usually use the target in the first frame as a fixed template, but the static template cannot adapt to target changes. The existing updater is challenging to deal with target deformation and update noise, and there is an excellent risk of updating with an inaccurate updater. In our research, a dynamic template updating strategy based on spatial-temporal information is proposed to improve the tracking accuracy of the Siamese tracker. Furthermore, Tracking Confidence Network (TCNet) is proposed to judge whether to update, which ensures that high-quality target features are used to update and reduce the noise caused by adding unreliable targets. In experiments, the proposed method is embedded into two baseline trackers: SiamRPN and SiamFC++, and tested on five popular benchmarks. The experimental results show that the proposed method can improve the performance of the Siamese trackers while maintaining real-time speed.

In the article, the research of sub-pixel snow mapping was conducted using moderate resolution data of remote sensing for obtaining high-accuracy data of snow cover in Qinghai-Tibet Plateau. But the end-member database is very large in the research, so the amount of calculation is too great if all the end-members will be used in the unmixing of pixel. In light of the characteristics of end-member database, the method of combining use of typical and neighboring end-members was established for the pixel unmixing of MODIS data in Qinghai-Tibet Plateau. Through the method, the percentage data of snow cover have been obtained in the plateau. Based on ASTER data, the validation has been conducted for the unmixing result in the research, which is quite accurate and reliable.