Non-Uniform Memory Access (NUMA) architectures bring new opportunities and challenges to bridge the gap between computing power and memory performance. Their complex memory hierarchies feature non-uniform access performance, known as NUMA locality, indicating data placement and access without NUMA-awareness significantly impact performance. Existing NUMA-aware solutions often prioritize fast local access but at the cost of heavy replication overhead, suffering a read-write performance tradeoff and limited scalability. To overcome these limitations, this paper presents Ladapa, a scalable and high-performance locality-aware data placement strategy. The key insight is decoupling data into metadata and data layers, allowing independent management with adaptive asynchronous replication for lower overhead. Additionally, Ladapa employs multi-level metadata management leveraging fast caches for efficient data location, further boosting performance. Experimental results show that Ladapa outperforms typical replication techniques by up to 27.37× in write performance and 1.63× in read performance.

The credibility and practicality of a reconstructed hand-object interaction sequence depend largely on its physical plausibility. However, due to high occlusions during hand-object interaction, physical plausibility remains a challenging criterion for purely vision-based tracking methods. To address this issue and enhance the results of existing hand trackers, this paper proposes a novel physically-aware hand motion de-noising method. Specifically, we introduce two learned loss terms that explicitly capture two crucial aspects of physical plausibility: grasp credibility and manipulation feasibility. These terms are used to train a physically-aware de-noising network. Qualitative and quantitative experiments demonstrate that our approach significantly improves both fine-grained physical plausibility and overall pose accuracy, surpassing current state-of-the-art de-noising methods.

Multi-robot global localization (MR-GL) with unknown initial positions in a large scale environment is a challenging task. The key point is the data association between different robots’ viewpoints. It also makes traditional Appearance-based localization methods unusable. Recently, researchers have utilized the object’s semantic invariance to generate a semantic graph to address this issue. However, previous works lack robustness and are sensitive to overlap rate of maps, resulting in unpredictable performance in real-world environments. In this paper, we propose a data association algorithm based on neighbor constraints to improve the robustness of the system. We demonstrate the effectiveness of our method on three different datasets, indicating a significant improvement in robustness compared to previous works.

Land use planning seeks to outline the future location and type of development activity. The planning process should reconcile development with environmental conservation and other concerns pertaining to sustainability; hence multi-objective spatial optimization is considered an effective tool to serve this purpose. However, as the number of social, economic, and environmental objectives increases, especially when numerous spatial units exist, the curse of dimensionality becomes a serious problem, making previous methods unsuitable. In this paper, we formulate a probabilistic framework based on the gradient descent algorithm (GDA) to search for Pareto optimal solutions more effectively and efficiently. Under this framework, land use as decision parameter(s) in each cell is represented as a probability vector instead of an integer value. Thus, the objectives can be designed as differentiable functions such that the GDA can be used for multi-objective optimization. An initial experiment is conducted using simulation data to compare the GDA with the genetic algorithm, with the results showing that the GDA outperforms the genetic algorithm, especially for large-scale problems. Furthermore, the outcomes in a real-world case study of Shenzhen demonstrate that the proposed framework is capable of generating effective optimal scenarios more efficiently, rendering it a pragmatic tool for planning practices.

Although deep learning has achieved great success in the image-classification tasks, its performance is subject to the quantity and quality of the training samples. For the classification of the polarimetric synthetic aperture radar (PolSAR) images, it is nearly impossible to annotate the images from visual interpretation. Therefore, it is urgent for remote-sensing scientists to develop new techniques for PolSAR image classification under the condition of very few training samples. In this letter, we take the advantage of active learning and propose active ensemble deep learning (AEDL) for PolSAR image classification. We first show that only 35% of the predicted labels of the deep-learning model's snapshots near its convergence were exactly the same. The disagreement between the snapshots is nonnegligible. From the perspective of multiview learning, the snapshots together serve as a good committee to evaluate the importance of the unlabeled instances. Using the snapshot committee to give out the informativeness of the unlabeled data, the proposed AEDL achieved better performance on two real PolSAR images than the standard active learning strategies. It achieved the same classification accuracy with only 86% and 55% of the training samples compared to the breaking tie active learning and random selection for the Flevoland data set.

In this paper, we propose a wide contextual residual network (WCRN) with active learning (AL) for remote sensing image (RSI) classification. Although ResNets have achieved great success in various applications (e.g. RSI classification), its performance is limited by the requirement of abundant labeled samples. As it is very difficult and expensive to obtain class labels in real world, we integrate the proposed WCRN with AL to improve its generalization by using the most informative training samples. Specifically, we first design a wide contextual residual network for RSI classification. We then integrate it with AL to achieve good machine generalization with limited number of training sampling. Experimental results on the University of Pavia and Flevoland datasets demonstrate that the proposed WCRN with AL can significantly reduce the needs of samples.

A novel early termination algorithm (ETA) is proposed for the serial low-complexity chase (LCC) Reed-Solomon (RS) decoder. It can terminate the decoding procedure after the reencoding step conditionally, which results in the improvement of the serial LCC decoders' speed. The ETA will not degrade the decoding performance of RS codes; on the contrary, it can even improve that of short RS codes remarkably. The simulations of the ETA for (3125) and (255,239) RS codes over the additive white Gaussian noise (AWGN) channel and a (458,410) RS code over the EPR4 channel with 100% AWGN show that it is effective. Moreover, the hardware implementation of the early terminated LCC decoder for (458,410) and (255,239) RS codes is provided. The analysis shows that it can work faster than the previous serial LCC decoder with only a few extra hardware costs. It also shows great advantages in average throughput performance than the common pipelined LCC decoders, especially for relatively high Eb/N0 values, by an overall consideration of the area, speed, and decoding performance.

Instruction-level parallelism (ILP) processors have been widely used to improve speed for several decades. However, the requirement of parallelism changes between applications, even within an application. Fixed high parallelism could result in poor utilization and extra leakage energy. Designing energy-efficient ILP processors to trade off power/speed has been a critical issue in current research. In this paper, a compiler-guided parallelism adaption based on an application partition algorithm is proposed to implement parallelism adaption with applications running on ILP processors. The aim is to minimize energy consumption without degrading the execution time. The main idea is described as follows: 1) partition the application into several power gating regions (PGRs); 2) assign adapted parallelism for each region by analyzing the requirements of resources and energy efficiency; and 3) reschedule each region with its own parallelism and insert power-gating instructions into the application to control hardware ON/OFF. The experimental results of evaluation with the CoreMarkPro benchmark suits show the expected savings of leakage energy. Our algorithm could reduce the leakage energy in register files by 30.46% and 64.06% for applications with high variance on software-inherent parallelism. Furthermore, the overhead energy originated from state transition is much lower than Tabkhi's algorithm.