Retrieval-augmented generation (RAG) has been extensively employed to mitigate hallucinations in large language models (LLMs). However, existing methods for multi-hop reasoning tasks often lack global planning, increasing the risk of falling into local reasoning impasses. Insufficient exploitation of retrieved content and the neglect of latent clues fail to ensure the accuracy of reasoning outcomes. To overcome these limitations, we propose **R**ecursive **E**valuation and **A**daptive **P**lanning (REAP), whose core idea is to explicitly maintain structured sub-tasks and facts related to the current task through the Sub-task Planner (SP) and Fact Extractor (FE) modules. SP maintains a global perspective, guiding the overall reasoning direction and evaluating the task state based on the outcomes of FE, enabling dynamic optimization of the task-solving trajectory. FE performs fine-grained analysis over retrieved content to extract reliable answers and clues. These two modules incrementally enrich a logically coherent representation of global knowledge, enhancing the reliability and the traceability of the reasoning process. Furthermore, we propose a unified task paradigm design that enables effective multi-task fine-tuning, significantly enhancing SP's performance on complex, data-scarce tasks. We conduct extensive experiments on multiple public multi-hop datasets, and the results demonstrate that our method significantly outperforms existing RAG methods in both in-domain and out-of-domain settings, validating its effectiveness in complex multi-hop reasoning tasks.

Aligning Large Language Models (LLMs) with diverse and potentially conflicting human values necessitates navigating complex multi-objective landscapes.However, existing prompt-conditioned approaches face a critical training-inference discrepancy: they rely on ground-truth scores during training while requiring manual user-specification at inference.We introduce prediction of implicit preferences to bridge this gap while reducing user burden.To this end, we propose Self-Guided Alignment (SGA), a framework that transforms passive reward dependency into an intrinsic adaptive sensing capability.It employs a dual-head architecture to unify preference internalization with conditional generation, enabling the model to learn a latent mapping between raw prompts and preference profiles.Through adaptive preference sensing, the model autonomously predicts the latent preference score to self-guide the generation, thereby eliminating the need for manual specification at inference.Extensive experiments across diverse model scales demonstrate that SGA often outperforms state-of-the-art baselines, achieving superior multi-objective trade-offs and improved preference alignment.

Oriented object detection (OOD) has rapidly advanced in recent years. However, the performance of existing methods is unsatisfactory when dealing with challenging scenarios, especially in scenes involving small-scale objects or objects with extreme aspect ratio. Inspired by recent advances in vision-language pre-training, we propose a novel Text-Guided Dual-Awareness Network (TG-DANet), which addresses these challenges from two complementary perspectives: robust feature interaction for multi-scale and long-range context modeling, and semantic-aware feature learning through textual guidance. Specifically, we design a Bi-Directional Feature Interaction Module (BDFIM) to capture horizontal and vertical contextual features via spatial interactions, which improves the representation of small and elongated objects. Additionally, a Text-Semantic Guidance Framework (TSGF) is supposed to align and fuse textual embeddings with visual features at multiple levels, which enhances model interpretability and discriminability for objects with ambiguous appearances or complex layouts. Extensive experiments on three benchmark datasets (DOTA, DIOR-R, and HRSC2016) show that TG-DANet achieves improvements of 3.05%, 3.49%, and 2.32% in mAP over baseline methods, respectively. These results demonstrate the effectiveness of our dual-perspective strategy in handling complex scenes with cluttered backgrounds and multi-scale objects, which highlights the promising potential of vision-language fusion in OOD.

To address the demand for fine-grained access control in Internet of Things (IoT) data sharing, attribute-based encryption (ABE) has emerged as a critical solution. Nevertheless, the computational overhead inherent in ABE poses a major challenge for its direct deployment on resource-constrained IoT devices. Data storage and sharing through centralized cloud may be interrupted due to cloud server failures. Additionally, if user attributes cannot be revoked, some users who should have had their data access rights canceled will still be capable of accessing the data. To solve these problems, we propose a multi-authority attribute-based data sharing scheme that combines blockchain to construct a secure and trusted data sharing environment, while supporting efficient attribute revocation. We reduce the computational burden during the online encryption phase through online/offline encryption, while introducing outsourced decryption to lessen the computational overhead of user decryption. Our scheme supports the tracing of key abusers and attribute revocation for users. Furthermore, by interplanetary file system (IPFS), we reduce on-chain storage burden, and by interplanetary name system (IPNS), we address the difficulty of logically deleting old ciphertexts when updating ciphertexts. Security analysis shows that our scheme exhibits static security, and experimental results demonstrate that our data sharing scheme exhibits excellent efficiency and practicality.

Recently, machine learning-embedded large-scale multiobjective evolutionary algorithms (LMOEAs) have shown great promise in solving large-scale multiobjective optimization problems (LMOPs). However, the fast convergence of the population to the true Pareto-optimal front (POF) and even distribution of the obtained Pareto-optimal solutions (POSs) on the POF are not adequately considered when tackling an LMOP. Besides, existing LMOEAs typically pair solutions with a matching rule and employ a network to learn the evolution pattern among the obtained solution pairs. It is difficult to learn various evolution patterns through a simple network, which hinders the collaboration of different patterns for enhancing the search capability. Facing such difficulties, this article proposes an LMOEA with multipattern learning and collaboration (LMOEA-MLC), where a single-hidden-layer multioutput network (SMN) is established to learn inductive and hybrid evolution patterns. Specifically, two inductive ones can be learned with the solution pairs built by two matching rules toward fast convergence and even distribution, respectively. Moreover, the solution pairs considering the fusion of the two inductive ones are collected, enabling SMN to learn a hybrid one and thus making a tradeoff between fast convergence and even distribution. Besides, the learned evolution patterns collaborate to enhance the search capability due to the distinct patterns. To enhance learning speed, SMN’s parameters are updated by an incremental random vector functional link (IRVFL). In our experiments, comprehensive comparisons with eight state-of-the-art LMOEAs demonstrate the significant performance improvement of LMOEA-MLC in handling LMOPs.

Oriented object detection (OOD) in remote sensing imagery has rapidly advanced in recent years. However, the complex characteristics of both objects and backgrounds still pose significant challenges, resulting in miss detections and false detections. Specifically, small objects and extreme aspect ratio objects are easily assimilated into the background. To solve these problems, we propose a High-Quality Sample Guidance Network (HQSGNet). Inspired by the human tendency to improve themselves through learning from outstanding individuals, we design a High-Quality Sample Guidance (HQSG) branch. This branch selects samples with high classification confidence and accurate regression to guide low-quality samples to optimize feature learning. Furthermore, to obtain more representative high-quality samples, we also propose a Discriminative Feature Refinement (DFR) module. This module provides a robust foundation for the selection of high-quality samples by reinforcing key feature information while suppressing irrelevant noise. Extensive experiments on three challenging remote sensing datasets (DOTA, DIOR-R and HRSC2016) achieve mAPs of 77.12%, 65.25% and 96.81% respectively, thereby validating the effectiveness of high-quality sample guidance mechanism in oriented object detection.

Compiler optimization is essential for improving program performance, yet modern compilers still depend on manually crafted transformation rules over intermediate representations (IRs). As compilers grow in complexity, maintaining these rule-based optimizations becomes increasingly labor-intensive and difficult to scale. Recent advances in large language models (LLMs) offer a promising alternative, but their effectiveness in compiler optimization remains limited—primarily due to the lack of IR-oriented datasets that expose models to diverse transformation samples in real-world scenarios (optimization-sensitive samples), hindering LLMs from learning rich and generalizable optimization strategies.In this paper, we introduce IR-OptSet, the first public optimization-sensitive dataset for advancing LLM-based IR optimizers. It comprises 170K LLVM IR samples from open-source repositories across 8 representative optimization domains. IR-OptSet defines two core tasks: Code Analysis and Optimized Code Generation, and provides tools for correctness verification, performance evaluation, and dataset expansion. In our experiments, fine-tuning three representative LLMs on IR-OptSet leads to significant accuracy improvements across both tasks. Moreover, the LLM fine-tuned with IR-OptSet outperforms traditional compiler with the -O3 option in 64 test cases in terms of performance. Further analysis reveals that IR-OptSet provides greater transformation diversity and representativeness than three widely used IR-oriented datasets, highlighting its potential to drive model-based IR optimization. IR-OptSet is publicly available at https://huggingface.co/datasets/YangziResearch/IR-OptSet.

Alzheimer's disease (AD) is one of the key diseases that seriously threatens the health of the elderly population and significantly increases the burden of social elderly care. The widespread application of artificial intelligence technology in the medical field has brought new ideas for the early diagnosis and intervention of AD. To this end, we propose a novel Multi-Branch Adaptive Deep Fusion Network (MADF-Net) to improve the accuracy of AD diagnosis. The MADF-Net consists of three main parts: the Supervised Autoencoder Alignment Module (S-AEAM), the Multi-Branch Adaptive Fusion Module (MAFusion), and the Dynamic Gated Deep Fusion Decision Module (DGD-Fusion). The S-AEAM is used to align and enhance the features of different modalities. The MAFusion aims to retain the main information of the current modality while integrating features from other modalities. Finally, the DGD-Fusion is employed to deeply fuse the features from all modalities and obtain the final output. Extensive experiments conducted on theCookie Theftcorpus from DementiaBank demonstrate that our proposed MADF-Net outperforms state-of-the-art (SOTA) models, achieving an accuracy of 88.91% and an F1 score of 90.48%.

Two-stage detectors are widely used in oriented object detection and have achieved high detection accuracy. However, some inconsistency issues in two-stage detectors limit their further improvement. To overcome these issues, this work proposes an effective two-stage center-symmetry representation-based high-quality localization detector (CR-HLDet) for oriented objects in remote sensing images. It improves the three main components of a typical two-stage detector: neck, region proposal network (RPN), and head. Specifically, in the neck, we introduce a feature enhancement block (FEB) to extract multilevel semantic information to solve the inconsistency in feature fusion. And we use a novel center-symmetry representation to generate high-quality oriented proposals in the RPN. Instead of the prediction of the rotated angle, it utilizes the symmetry of the center point to locate any oriented bounding box (OBB) by three points, which gets rid of the inconsistency between proposals and oriented objects. Finally, we design a dual classification detection head (DCDH) to obtain more robust and reasonable classification confidence. It solves the inconsistency between classification confidence and the regression quality by establishing the connection between the classification branch and the regression branch. The effective combination of the above three modules enables our CR-HLDet to accurately locate oriented objects in remote sensing images. Extensive and comprehensive experiments on three challenging remote sensing datasets (DOTA, HRSC2016, and DIOR-R) demonstrate that the proposed method achieves state-of-the-art performance.

The growing adoption of electric vehicles (EVs) has placed significant demands on power grids, necessitating coordination between EV charging and power dispatching. This paper proposes a novel EV charging system using Multi-Agent Large Language Models (LLMs) to enhance recommendations, optimize decision-making, and dynamically adapt to user behaviors and grid conditions. The system includes a User Agent and an EV Charging Station (EVCS) Agent, connected through a Negotiation Platform for secure data sharing. The User Agent provides personalized recommendations based on historical data, while the EVCS Agent adjusts prices in real time using fine-tuned LLMs. A Conditional Generative Adversarial Network (CGAN) model is used to generate user behavior and pricing data to fine-tune the LLMs. The proposed system effectively adapts to dynamic user behaviors and grid conditions by combining Multi-Agent coordination with fine-tuned LLMs and CGAN-generated data. A case study demonstrates the system’s ability to balance user preferences with power dispatching, offering scalable, efficient, and intelligent solutions for modern EV ecosystems.

Commit messages provide important information for comprehending the code changes, and a number of researchers try to generate commit messages by using an automatic way. These research on commit message generation has profited from the code tokens or code structures such as AST. Since the edit sequence of code change is also important for capturing the code change intent, we propose a new commit message generation method called ESGen, which extracts AST edit sequences of code changes as model input. Specifically, we employ an O(ND) difference algorithm to extract the edit sequence from AST by comparing the ASTs before and after applying the code changes. Then, we construct a Bi-Encoder, which encodes the textual information and the AST edit sequence information of code change. The experimental results show that ESGen outperforms other baseline models, improving the BLEU-4 to 15.14. Also, when applying the edit sequence to 7 baseline models, they improve the BLEU-4 scores of these models by an average of 8.5%. Additionally, a human evaluation confirmed the effectiveness of ESGen in generating commit messages.

As one of the most well-known programmer Q&A websites, Stack Overflow (i.e., SO) is serving tens of thousands of developers every day. Previous work has shown that many developers reuse the code snippets on SO when they find an answer (from SO) that functionally matches the programming problem they encounter in their development activities. To study how programmers reuse code on SO during project development, we conduct a comprehensive empirical study. First, to capture the development activities of programmers, we collect 342,148 modified code snippets in commits from 793 open-source Java projects, and these modified code can reflect the programming problems encountered during development. We also collect the code snippets from 1,355,617 posts on SO. Then, we employ CCFinder to detect the code clone between the modified code from commits and the code from SO, and further analyze the code reuse when programmer solves a programming problem during development. We count the code reuse ratios of the modified code snippets in the commits of each project in different years, the results show that the average code reuse ratio is 6.32%, and the maximum is 8.38%. The code reuse ratio in project commits has increased year by year, and the proportion of code reuse in the newly established project is higher than that of old projects. We also find that some projects reuse the code snippets from many years ago. Additionally, we find that experienced developers seem to be more likely to reuse the knowledge on SO. Moreover, we find that the code reuse ratio in bug-related commits (6.67%) is slightly higher than that of in non-bug-related commits (6.59%). Furthermore, we also find that the code reuse ratio (14.44%) in Java class files that have undergone multiple modifications is more than double the overall code reuse ratio (6.32%).

While the influence maximization problem (IMP) which studies how to trigger a large cascade in Online Social Networks (OSNs) by properly selecting seed nodes has been extensively studied in the past decade, one important and practical issue on how these seed nodes or celebrities will get paid for promoting cascades is seldom addressed. In order to get selected by the advertising platform and to maximize his/her own utility, it is natural for each celebrity to determine his/her price of promoting cascades based on other celebrities' decisions and his/her power of influence. In this paper, we formulate the problem of determining prices by celebrities as a pricing game, with celebrities as players. We show that celebrity selection by the advertising platform is NP-hard, and assume that the advertising platform will adopt the simple greedy algorithm that is widely used in IMP. Under this assumption, we study the pure Nash equilibrium of the pricing game among celebrities. In particular, we prove that while equilibrium exists and is unique when there are only two or three players, the equilibrium is not guaranteed to exist in cases of four or more players.

Failures in data center networks sometimes can lead to user-perceived service interruptions. Automated failure detection is needed to maintain the reliability of data centers. However, researches rarely identify quality of service (QoS) multiplicity for failure detection in data center networks.

Multicast denotes an idea of sending data to numbers of receivers from one source in one transmission. It has been widely applied in group communication (e.g., media streaming, multi-point video conferencing). Multicast routing tree (MRT) is usually built to keep the right paths to transmit data, where data copies are created in parent nodes and then forwarded to child nodes. However, constructing an MRT is usually difficult for a given network topology; finding an optimal multicast routing tree with the minimal cost is a proven NP-complete problem. Moreover, multicast applications usually run in local or small networks due to the limitations in flexibility, scalability, and security.

Although FPGA's power and performance advantages were recognized widely, designing applications on FPGA-based systems is traditionally a task undertaken by hardware experts. It is significant to allow application-level programmers with less system-level but more algorithm knowledge to realize their applications conveniently on FPGAs. In this paper, an embedded FPGA operating system is proposed to facilitate application-level programmers to use FPGAs. Firstly, it builds specific I/Os and optimizes bus interconnection among I/Os, DDR memory, user IPs etc within the FPGA for vision computing. Secondly, it manages resources of the FPGA such as I/Os, DDR memory, communication etc, frees users from low-level details. Thirdly, it schedules tasks (IPs) executed on the FPGA dynamically in runtime, which makes the FPGA multiplexed when necessary. After porting the FPGA operating system to different FPGA platforms and implementing vision algorithms based on that, it shows the FPGA operating system is able to simplify algorithm development on FPGA platforms and improve portability of user applications. Furthermore, implementation results of several popular vision algorithms show the FPGA operating system is efficient and effective for vision computing. Finally, experimental results shows that for multiple algorithms requiring more FPGA resources, runtime task scheduling of multiple IPs is more efficient than a fixed IP when the SoC of FPGA is considered.

Relaying is a promising technique in cellular networks for improving system capacity and coverage. To facilitate the deployment of relays in remote areas without ready access to the electrical grid, energy-harvesting relays may be deployed. Energy-harvesting has been studied extensively for sensor networks, but it is still an open problem for cellular networks. In this paper, we study the problem of the joint allocation of orthogonal frequency division multiplexing access resource blocks and transmission power to users in a cellular network with energy-harvesting relays. The energy-harvesting process is stochastically described by a time-varying Poisson process. We propose a new metric called survival probability as the selection criteria for an energy-harvesting relay to support data transmissions. We propose a survival probability-based resource allocation (SPRA) algorithm. The algorithm solves the joint problem of resource block allocation, power control, and associating relays to users in a cellular network. We show the achievable data rates of SPRA for different energy harvesting rates.

High-quality algorithms for dense optical flow computation are computationally intensive. To compute them with high speed and low power is vital to make optical flow computation applicable in real-world applications. In contrast to only the Horn-Schunck model being studied on FPGA-based systems today, one of the best linear variational methods for dense optical flow computation, Combine-Brightness-Gradient, is implemented on FPGA-accelerated heterogeneous platforms in this paper. C instead of HDLs is employed and optimizing techniques based on the algorithmic parallelism and hardware architecture are introduced. Experimental results show that 30-110x improvement of the computing efficiency over CPUs was achieved. The FPGA-accelerated version is able to process 640 × 480 image at 12 fps with 0.38 J per frame, while it is 0.8 fps and around 40 J on CPUs. Through demonstrating high performance and low power of dense optical flow algorithm on FPGA-based heterogeneous platforms implemented in C, this paper shows that the off-the-shelf commodity FPGAs coupled with High-Level-Synthesis (HLS) tools could provide an available option when computational efficiency together with development speed are required.

We investigate bandwidth allocation and scheduling in non-cooperative wireless networks as a mixed integer programming problem. Fast Vickrey-Clarke-Groves (VCG) auction-based bandwidth allocation (FABA), incorporating relaxation-based greedy algorithm (RGA) and split-flow-based algorithm (SFA), is proposed by modifying the traditional VCG auction to make it computationally feasible. With incentives provided by FABA, the dominant strategy of any selfish node in the network is to be cooperative so that the system cost is minimized. We implement FABA via a batching-based mechanism which allocates bandwidth for all call routing requests arriving in a certain batching period simultaneously. Our simulation evaluates the performance in terms of system cost, payment-cost ratio, and setup time.

We study multi path routing with traffic assignment in selfish networks. Based on the Vickrey-Clarke-Groves (VCG) auction, an optimal and strategy-proof scheme, known as optimal auction-based multipath routing (OAMR), is developed. However, OAMR is computationally expensive and cannot run in real time when the network size is large. Therefore, we propose sequential auction-based multi path routing (SAMR). SAMR handles routing requests sequentially using some greedy strategies. In particular, with reference to the Ausubel auction, we develop a water-draining algorithm to assign the traffic of a request among its available paths and determine the payment of the transmission in approximately constant time. Our simulation results show that SAMR can rapidly compute the allocations and payments of requests with small sacrifice on the system cost. Moreover, various sequencing strategies for sequential auction are also investigated.

Grid technology has being widely accepted in distributed resources sharing and high performance computing cross multi administrative domains. In this paper, we analysis the security issues in grid computing environments, and propose a security framework for VegaGOS which is a service oriented architecture middleware developed for the China National Grid. We address mutual authentication using certificate with digital signature. We address authorization through combining VO level access control decision and resource level enforcement. Communication security is guaranteed by TLS/SSL at transport level and WS-security at message level. This security framework has been implemented in VegaGOS and deployed in China National Grid Environment.

Virtual organizations (VO) are widely accepted in the grid and other distributed computing environments. However, there are few effective VO implementations. This paper presents a layered architecture to construct Agora, an implementation of VO. Agora manages users, resources, and agora instances, provides policies to support a DAC/MAC-hybrid cross-domain access control mechanism, and maintains the context of operations. The Agora architecture consists of three layers. At the bottom is the physical layer containing external resources, then an abstraction RController is introduced to manipulate external resources. Above the physical layer, all the involved entities, including users, resources, and agoras, are abstracted as GNodes, and a naming layer is introduced to manage these GNodes. At the top, the logic layer implements all the Agora functionalities. This architecture has been implemented in Vega GOS and applied in the China National Grid and other grid platforms. The evaluation shows that the architecture provides minimal but sufficient VO functionalities while keeping decentralization, flexibility, simplicity, and effectiveness.