Software engineering (SE) is increasingly collaborative, with developers working together on shared complex codebases. Effective collaboration in environments requires participants -- whether humans or AI agents to stay the same page as their environment evolves. When a collaborator's understanding diverges from current state what we term out-of-sync challenge actions may fail, leading integration issues. In this work, introduce SyncMind, framework that systematically defines problem faced...

Mega-constellation network traffic forecasting provides key information for routing and resource allocation, which is of great significance to the performance satellite networks. However, due self-similarity long-range dependence (LRD) mega-constellation traffic, traditional linear/non-linear models cannot achieve sufficient accuracy. In order resolve this problem, a model based on EMD (empirical mode decomposition)-ARIMA (autoregressive integrated moving average) IGWO (improved grey wolf...

The expansion of megaconstellation networks (MCNs) represents a promising solution for achieving global Internet coverage. To meet the growing demand satellite services, multipath routing allows simultaneous establishment multiple transmission paths, enabling flows in parallel. Nevertheless, mobility satellites and time-varying link states presents challenge discovery optimal paths traffic scheduling routing. Given inflexibility traditional static deep reinforcement learning (DRL)-based...

Large language Models (LLMs) have achieved promising performance on arithmetic reasoning tasks by incorporating step-by-step chain-of-thought (CoT) prompting. However, LLMs face challenges in maintaining factual consistency during reasoning, exhibiting tendencies to condition overlooking, question misinterpretation, and hallucination over given problems. Existing methods use coarse-grained feedback (e.g., whether the answer is correct) improve consistency. In this work, we propose RCoT...

This paper investigates the single agile optical satellite scheduling problem, which has received increasing attention due to rapid growth in earth observation requirements. Owing complicated constraints and considerable solution space of this conventional exact methods heuristic methods, are sensitive problem scale, demand high computational expenses. Thus, an efficient approach is demanded solve proposes a deep reinforcement learning algorithm with local mechanism. A mathematical model...

As one of the most important part weapon system systems (WSoS), quantitative evaluation reconnaissance satellite (RSS) is indispensable during its construction and application. Aiming at problem nonlinear effectiveness under small sample conditions, we propose an method based on support vector regression (SVR) to effectively address defects traditional methods. Considering performance SVR influenced by penalty factor, kernel type, other parameters deeply, improved grey wolf optimizer (IGWO)...

Today's large language models (LLMs) typically train on short text segments (e.g., <4K tokens) due to the quadratic complexity of their Transformer architectures. As a result, performance suffers drastically inputs longer than those encountered during training, substantially limiting applications in real-world tasks involving long contexts such as encoding scientific articles, code repositories, or dialogues. Through theoretical analysis and empirical investigation, this work identifies...

With the expansion of user scale in LEO satellite networks, unbalanced regional load and bursty network traffic lead to problem disequilibrium. A distributed hops-based back-pressure (DHBP) routing is proposed. DHBP theoretically derives a fast solution for minimum end-to-end propagation hops between nodes inclined-orbit networks; hence, link weights are determined based on remaining next hop destination satellites. In order control number available retransmission paths, permitted region...

Due to the openness of inter-satellite links (ISLs) in mega-constellations, threat posed by jamming from non-cooperative constellations is becoming increasingly significant. Most existing approaches focus on up/down link capacity between satellites and ground stations, which differs greatly situation whereby ISLs are subjected jamming. Therefore, this work investigates transmission rates under mega-constellations. Based this, a novel satellite network calculation method proposed evaluate...

Position bias has proven to be a prevalent issue of modern language models (LMs), where the prioritize content based on its position within given context. This often leads unexpected model failures and hurts performance, robustness, reliability across various applications. Our mechanistic analysis attributes two components employed in nearly all state-of-the-art LMs: causal attention relative positional encodings. Specifically, we find that generally causes favor distant content, while...

With the rapid growth in space-imaging demands, scheduling problem of multiple agile optical satellites has become a crucial field on-orbit satellite applications. Because considerable solution space and complicated constraints, existing methods suffer from huge computation burden low quality. This paper establishes mathematical model this problem, which aims to maximize observation profit rate realize load balance, proposes multi-pointer network solve adopts attention layers as pointers...

Drones provide a versatile platform for remote sensing and atmospheric studies. However, strict payload mass limits intense vibrations have proven obstacles to adoption astronomy. We present concept system-level testing of long-baseline CubeSat space interferometer using drones, taking advantage their cm-level xyz station-keeping, 6-dof freedom movement, large operational environment, access guide stars end-to-end optical train control algorithms, comparable power requirements. purchased two...

Building a human-like system that continuously interacts with complex environments -- whether simulated digital worlds or human society presents several key challenges. Central to this is enabling continuous, high-frequency interactions, where the interactions are termed experiences. We refer envisioned as LifeSpan Cognitive System (LSCS). A critical feature of LSCS its ability engage in incremental and rapid updates while retaining accurately recalling past identify two major challenges...

Complex news events, such as natural disasters and socio-political conflicts, require swift responses from the government society. Relying on historical events to project future is insufficient are sparse do not cover all possible conditions nuanced situations. Simulation of these complex can help better prepare reduce negative impact. We develop a controllable event simulator guided by both schema representing domain knowledge about scenario user-provided assumptions case-specific...

This paper investigates the agile optical satellite scheduling problem, which aims to arrange an observation sequence and actions for tasks. Existing research mainly maximize number of completed tasks or total priorities but ignores influence on imaging quality. Besides, conventional exact methods heuristic can hardly obtain a high-quality solution in short time due complicated constraints considerable space this problem. Thus, proposes two-stage framework with deep reinforcement learning...

The rapid advancements in large language models (LLMs) have demonstrated their potential to accelerate scientific discovery, particularly automating the process of research ideation. LLM-based systems shown promise generating hypotheses and ideas. However, current approaches predominantly rely on prompting-based pre-trained models, limiting ability optimize generated content effectively. Moreover, they also lack capability deal with complex interdependence inherent restrictions among...

Satellite network traffic forecasting provides key information for routing and resource allocation, which is important the efficient operation of satellite network. However, due to self-similarity long-range dependence (LRD) traffic, traditional linear or non-linear models cannot achieve sufficient accuracy. A combined model based on BPNN ARIMA proposed. adopted forecast residuals. EMD-ARIMA results are forecasting. Experimental data generated by ON/OFF show that accuracy proposed method...