arXiv subjects

Shaojie Wang

Publications and source records attributed to Shaojie Wang.

At least 19 recordsLinked to original sources

arXiv2026

Adaptive Beam Hopping and Power Control for Dual-Layer Over-the-Air Online Federated Learning in LEO Satellite Networks

This paper investigates over-the-air (OTA) computation enabled online federated learning (FL) in low-Earth orbit (LEO) satellite networks. Specifically, we consider a dual-layer OTA aggregation architecture, where ground devices upload analog model updates to serving satellites via uplink OTA aggregation, and satellites forward the aggregated signals to a data processing center through the second round OTA aggregation. Then, we formulate a long-term data-utilization maximization problem in which devices continuously collect new data and untrained samples gradually lose freshness. The problem is subject to the satellite beam budget, transmit-power limit, and global mean squared error (MSE) constraint that governs end-to-end aggregation distortion. This yields a coupled mixed-integer nonlinear programming (MINLP) problem, involving tightly coupled discrete beam-hopping decisions and continuous power control. Due to the combinatorial action space and nonconvex constraints, the problem is NP-hard and computationally intractable. Furthermore, the time-varying satellite topology and dynamic data generation render it a sequential decision-making problem, necessitating adaptive online scheduling. To address these issues, we cast the problem as a Markov decision process and develop a proximal policy optimization (PPO)-based deep reinforcement learning framework that jointly optimizes adaptive beam hopping and power control, using an MSE-aware reward to balance data utilization and aggregation accuracy. Numerical simulation results verify that the proposed algorithm consistently outperforms other benchmark schemes, achieving superior long-term data utilization and faster FL convergence while satisfying the MSE requirement.

cs.IT↗

arXiv2026

Dual-Layer Over-the-Air Federated Learning in LEO Satellite Networks: Architecture, Key Technologies and Applications

Low Earth orbit (LEO) satellite networks are emerging as a pivotal infrastructure for global edge intelligence. In this context, integrating over-the-air (OTA) computation with adaptive beam hopping (BH) provides an innovative framework that seamlessly merges physical-layer analog aggregation with dynamic resource orchestration. This effectively overcomes the stringent bandwidth and power constraints of space platforms while extending federated learning (FL) to pervasive Internet-of-things (IoT) deployments. In this article, we first outline the fundamental principles of the dual-layer OTA model and introduce the adaptive BH mechanism designed for time-varying topologies. Then, we summarize the distinct advantages of this learning-centric architecture, which include decoupling aggregation latency from device density, optimizing spatio-temporal resource efficiency, and balancing data freshness with channel quality. Several application scenarios are explored to highlight the framework's potential across diverse vertical industries. Furthermore, a specific case is studied to demonstrate the practical efficacy of the proposed scheduling policy. The results reveal substantial performance gains in terms of model convergence speed and data utilization for satellite-based FL systems. Finally, we discuss the implementation challenges and outline future research directions, aiming to provide insights for the evolution of ubiquitous non-terrestrial intelligence.

eess.SP↗

arXiv2026

Knowing What to Solve Before How: Preplan Empowered LLM Mathematical Reasoning

Current plan-based reasoning methods improve large language models (LLMs) by inserting a planning stage before execution, giving rise to the question $\rightarrow$ plan $\rightarrow$ cot paradigm. While effective, a closer examination reveals an inherent paradigm-level gap: both the planning and its execution stages decide how to solve a problem, while the prior question of what to solve; recognizing the problem type, the applicable tools, and the foreseeable pitfalls; remains entirely implicit. To bridge this gap, we propose PPC (Preplan-Plan-CoT), a framework that introduces an explicit problem-understanding stage, the preplan, yielding a new question $\rightarrow$ preplan $\rightarrow$ plan $\rightarrow$ cot paradigm. Realizing this paradigm requires safeguarding the conceptual integrity of preplan at both ends. Specifically, we design a three-stage synthesis pipeline with a spoiler-score detector that filters out leakage and spoiler failures to build clean preplan supervision, and a composite GRPO reward enforces that the generated plan genuinely follows from the preplan. Experiments across four backbones and five mathematical reasoning benchmarks show that PPC achieves the best results on 39 of 40 metrics, improving maj@16 and pass@16 by +2.23 and +3.06 over the strongest baseline without introducing additional inference token overhead.

cs.CL↗

arXiv2026

2DESR: a two-dimensional Fourier-space gyrokinetic eigenvalue code for the ion-temperature-gradient modes in tokamaks

A two-dimensional (2D) gyrokinetic eigenvalue solver, 2DESR, has been developed to solve the 2D gyrokinetic eigenvalue problem in the poloidal Fourier space for the ion-temperature-gradient (ITG) modes in tokamaks. With full kinetic effects of ions retained, the 2D gyrokinetic eigenvalue equations in the poloidal Fourier space have been derived and numerically solved in the 2DESR code. In the linear ITG Cyclone test with adiabatic electrons, the 2DESR code benchmarks well against the gyrokinetic initial-value codes GENE and NLT. It is found that two branches of ITG modes coexist in the system.

physics.plasm-ph↗

arXiv2026

From Meta-Thought to Execution: Cognitively Aligned Post-Training for Generalizable and Reliable LLM Reasoning

Current LLM post-training methods optimize complete reasoning trajectories through Supervised Fine-Tuning (SFT) followed by outcome-based Reinforcement Learning (RL). While effective, a closer examination reveals a fundamental gap: this approach does not align with how humans actually solve problems. Human cognition naturally decomposes problem-solving into two distinct stages: first acquiring abstract strategies (i.e., meta-knowledge) that generalize across problems, then adapting them to specific instances. In contrast, by treating complete trajectories as basic units, current methods are inherently problem-centric, entangling abstract strategies with problem-specific execution. To address this misalignment, we propose a cognitively-inspired framework that explicitly mirrors the two-stage human cognitive process. Specifically, Chain-of-Meta-Thought CoMT focuses supervised learning on abstract reasoning patterns without specific executions, enabling acquisition of generalizable strategies. Confidence-Calibrated Reinforcement Learning (CCRL) then optimizes task adaptation via confidence-aware rewards on intermediate steps, preventing overconfident errors from cascading and improving execution reliability. Experiments across four models and ten benchmarks show 2.10% and 3.86% improvements in-distribution and out-of-distribution respectively over standard methods, while remaining highly robust to variations in teacher model selection, optimization methods, and symbolic perturbations.

cs.AI↗

arXiv2026

Electric field induced by radial redistribution of the energetic ion pressure in a fusion plasma

It is found by using the gyrokinetic theory that significant radial electric fields, or zonal flows, can be generated by the radial redistribution of energetic ion pressure in a tokamak fusion device. Trapped energetic ions are more effective to generate the radial electric field than the isotropic energetic ions. This suggests that the energetic $\alpha$ particles produced by DT fusion may induce significant radial electric field and thus help to improve the core plasma confinement in a fusion reactor.

physics.plasm-ph↗

arXiv2026

From Implicit to Explicit: Token-Efficient Logical Supervision for Mathematical Reasoning in LLMs

Recent studies reveal that large language models (LLMs) exhibit limited logical reasoning abilities in mathematical problem-solving, instead often relying on pattern-matching and memorization. We systematically analyze this limitation, focusing on logical relationship understanding, which is a core capability underlying genuine logical reasoning, and reveal that errors related to this capability account for over 90\% of incorrect predictions, with Chain-of-Thought Supervised Fine-Tuning (CoT-SFT) failing to substantially reduce these errors. To address this bottleneck, we propose First-Step Logical Reasoning (FSLR), a lightweight training framework targeting logical relationship understanding. Our key insight is that the first planning step-identifying which variables to use and which operation to apply-encourages the model to derive logical relationships directly from the problem statement. By training models on this isolated step, FSLR provides explicit supervision for logical relationship understanding, unlike CoT-SFT which implicitly embeds such relationships within complete solution trajectories. Extensive experiments across multiple models and datasets demonstrate that FSLR consistently outperforms CoT-SFT under both in-distribution and out-of-distribution settings, with average improvements of 3.2\% and 4.6\%, respectively. Moreover, FSLR achieves 4-6x faster training and reduces training token consumption by over 80\%.

cs.CL↗

arXiv2025

Tree Training: Accelerating Agentic LLMs Training via Shared Prefix Reuse

Agentic large language model (LLM) training often involves multi-turn interaction trajectories that branch into multiple execution paths due to concurrent tool use, think-mode, sub-agent, context management and other runtime designs. As a result, the tokens produced by a single task naturally form a tree-structured token trajectory with shared prefixes, rather than a linear sequence. Existing training pipelines linearize such trajectories and treat each branch independently, leading to substantial redundant computation in both forward and backward passes. We derive that averaging the loss over all branches independently is algebraically identical to a per-token weighted loss, where each token's weight equals the fraction of branches passing through it. The problem therefore reduces to computing the log-probability of every token in the prefix tree exactly once, with no repeated computation across shared prefixes: we propose DFS serialization of the tree, which visits every token exactly once, and adapt full-attention and SSM layers to ensure the resulting log-probabilities match independent per-branch calculation exactly. In practice, a single trajectory tree can be too large to fit in GPU memory; we therefore propose Redundancy-Free Tree Partitioning, which handles memory-constrained settings with zero redundant computation and peak memory bounded by a single root-to-leaf path. Together, these contributions form Tree Training, an efficient framework for training LLMs on tree-structured trajectories, achieving up to 6.2x end-to-end training speedup on dense and MoE models for both supervised fine-tuning and reinforcement learning.

cs.LG↗

arXiv2025

KAT-Coder Technical Report

Recent advances in large language models (LLMs) have enabled progress in agentic coding, where models autonomously reason, plan, and act within interactive software development workflows. However, bridging the gap between static text-based training and dynamic real-world agentic execution remains a core challenge. In this technical report, we present KAT-Coder, a large-scale agentic code model trained through a multi-stage curriculum encompassing Mid-Term Training, Supervised Fine-Tuning (SFT), Reinforcement Fine-Tuning (RFT), and Reinforcement-to-Deployment Adaptation. The Mid-Term stage enhances reasoning, planning, and reflection capabilities through a corpus of real software engineering data and synthetic agentic interactions. The SFT stage constructs a million-sample dataset balancing twenty programming languages, ten development contexts, and ten task archetypes. The RFT stage introduces a novel multi-ground-truth reward formulation for stable and sample-efficient policy optimization. Finally, the Reinforcement-to-Deployment phase adapts the model to production-grade IDE environments using Error-Masked SFT and Tree-Structured Trajectory Training. In summary, these stages enable KAT-Coder to achieve robust tool-use reliability, instruction alignment, and long-context reasoning, forming a deployable foundation for real-world intelligent coding agents. Our KAT series 32B model, KAT-Dev, has been open-sourced on https://huggingface.co/Kwaipilot/KAT-Dev.

cs.CL↗

arXiv2025

Superconductivity at 22.3 K in Compressed Sodium-intercalated Graphite

Graphite intercalation compounds (GICs) have long been recognized as promising candidates for high-temperature superconductivity by intercalation or charge doping, yet experimental progress has stalled with transition temperatures (Tc) limited to 11.5 K at ambient pressure and 15.1 K at 7.5 GPa in calcium-intercalated graphite over decades. Here, we report robust superconductivity in sodium-intercalated graphite with Tc of 22.3 K, as demonstrated by clear zero-resistance behavior. Our approach involves simply room-temperature grinding of graphite with sodium, followed by slight compression up to 7.1 GPa, circumventing complex synthesis procedures. Through synchrotron X-ray diffraction combined with first-principles calculations, we identify the major superconducting phase as an orthorhombic stage-2 GIC structure with slightly over-stoichiometric composition (Na1+xC8). Electron-phonon coupling calculations reveal that superconductivity primarily emerges from the interactions between out-of-plane carbon electrons and low-frequency Na/C vibrations.The enhancement in Tc establishes sodium as superior for achieving higher-Tc in GICs and illustrates promising pathway for further optimization through compositional and structural tuning.

cond-mat.supr-con↗

arXiv2025

SeamlessFlow: A Trainer Agent Isolation RL Framework Achieving Bubble-Free Pipelines via Tag Scheduling

We introduce SeamlessFlow, a server based reinforcement learning (RL) framework that addresses two core challenges in industrial scale RL: (1) decoupling RL training from the complex execution flow of agents; (2) maximizing GPU utilization with minimal idle time while preserving the stability and scalability required for large-scale deployments. First, SeamlessFlow introduces a data plane that decouples the RL trainer from diverse, complex agent implementations while sustaining high throughput. A central trajectory manager maintains complete interaction histories and supports partial rollout, allowing rollout to pause for weight updates and resume seamlessly, keeping agents unaware of service interruptions. Second, we propose a tag driven scheduling paradigm that abstracts hardware into capability tagged resources, unifying colocated and disaggregated architectures. Based on this, SeamlessFlow introduces a spatiotemporal multiplexing pipeline that dynamically reassigns idle training nodes to rollout in a train rollout separated setup, eliminating pipeline bubbles and fully exploiting heterogeneous cluster resources. By combining these innovations, SeamlessFlow delivers both stability and high performance, making it well suited for multi agent, long horizon, and other complex RL tasks.

cs.LG↗

arXiv2025

KAT-V1: Kwai-AutoThink Technical Report

We present Kwaipilot-AutoThink (KAT), an open-source 40B large language model developed to address the overthinking problem in reasoning-intensive tasks, where an automatic thinking training paradigm is proposed to dynamically switch between reasoning and non-reasoning modes based on task complexity. Specifically, first, we construct the dual-regime dataset based on a novel tagging pipeline and a multi-agent synthesis strategy, and then we apply Multi-Token Prediction (MTP)-enhanced knowledge distillation, enabling efficient and fine-grained reasoning transfer with minimal pretraining cost. Besides, we implement a cold-start initialization strategy that introduces mode-selection priors using majority-vote signals and intent-aware prompting. Finally, we propose Step-SRPO, a reinforcement learning algorithm that incorporates intermediate supervision into the GRPO framework, offering structured guidance over both reasoning-mode selection and response accuracy. Extensive experiments across multiple benchmarks demonstrate that KAT consistently matches or even outperforms current state-of-the-art models, including DeepSeek-R1-0528 and Qwen3-235B-A22B, across a wide range of reasoning-intensive tasks while reducing token usage. Notably, KAT outperforms all open-source models and even surpasses o3-mini on the leakage-controlled LiveCodeBench Pro. Beyond academic evaluation, KAT has been successfully deployed in Kwaipilot (i.e., Kuaishou's internal coding assistant), where it improves real-world development workflows with high accuracy, efficiency, and controllable reasoning behaviors. Moreover, we are actively training a 200B Mixture-of-Experts (MoE) model with 40B active parameters, and early results already show significant gains, further demonstrating the scalability of the AutoThink paradigm.

cs.CL↗

arXiv2025

Rethinking the Understanding Ability across LLMs through Mutual Information

Recent advances in large language models (LLMs) have revolutionized natural language processing, yet evaluating their intrinsic linguistic understanding remains challenging. Moving beyond specialized evaluation tasks, we propose an information-theoretic framework grounded in mutual information (MI) to achieve this. We formalize the understanding as MI between an input sentence and its latent representation (sentence-level MI), measuring how effectively input information is preserved in latent representation. Given that LLMs learn embeddings for individual tokens, we decompose sentence-level MI into token-level MI between tokens and sentence embeddings, establishing theoretical bounds connecting these measures. Based on this foundation, we theoretically derive a computable lower bound for token-level MI using Fano's inequality, which directly relates to token-level recoverability-the ability to predict original tokens from sentence embedding. We implement this recoverability task to comparatively measure MI across different LLMs, revealing that encoder-only models consistently maintain higher information fidelity than their decoder-only counterparts, with the latter exhibiting a distinctive late-layer "forgetting" pattern where mutual information is first enhanced and then discarded. Moreover, fine-tuning to maximize token-level recoverability consistently improves understanding ability of LLMs on tasks without task-specific supervision, demonstrating that mutual information can serve as a foundation for understanding and improving language model capabilities.

cs.CL↗

arXiv2025

SRPO: A Cross-Domain Implementation of Large-Scale Reinforcement Learning on LLM

Recent advances of reasoning models, exemplified by OpenAI's o1 and DeepSeek's R1, highlight the significant potential of Reinforcement Learning (RL) to enhance the reasoning capabilities of Large Language Models (LLMs). However, replicating these advancements across diverse domains remains challenging due to limited methodological transparency. In this work, we present two-Staged history-Resampling Policy Optimization (SRPO), which surpasses the performance of DeepSeek-R1-Zero-32B on the AIME24 and LiveCodeBench benchmarks. SRPO achieves this using the same base model as DeepSeek (i.e. Qwen2.5-32B), using only about 1/10 of the training steps required by DeepSeek-R1-Zero-32B, demonstrating superior efficiency. Building upon Group Relative Policy Optimization (GRPO), we introduce two key methodological innovations: (1) a two-stage cross-domain training paradigm designed to balance the development of mathematical reasoning and coding proficiency, and (2) History Resampling (HR), a technique to address ineffective samples. Our comprehensive experiments validate the effectiveness of our approach, offering valuable insights into scaling LLM reasoning capabilities across diverse tasks.

cs.LG↗

arXiv2025

Optimizing stellarators with hidden symmetry

Stellarators confine fusion plasmas using three-dimensional magnetic fields composed of nested toroidal magnetic surfaces. In generic stellarators, trapped particles can drift across these surfaces and degrade plasma confinement. Certain topological properties of the magnetic field strength can suppress these drifts. However, conventional stellarator design approaches typically enforce restrictive constraints to realize such properties, thereby segmenting and limiting the accessible configuration space. In this work, we reformulate the conditions for efficient confinement as constraints on a homeomorphic straightening transformation of the field contours. Within this framework, the various families of stellarator magnetic fields optimized for plasma confinement arise naturally as specific realizations of a unified mapping. This new perspective provides a significantly more comprehensive description of viable stellarator configurations, enabling systematic exploration of trade-offs among confinement quality, geometric complexity, and engineering requirements. We illustrate this approach by presenting a highly compact stellarator design that nevertheless achieves plasma performance comparable to that of leading reactor-scale designs with much larger aspect ratios.

physics.plasm-ph↗

arXiv2025

Geometric curvature effect on suppressing the Ion-Temperature-Gradient mode near the magnetic axis

Global gyrokinetic simulation of the ion temperature gradient mode shows that the radial electric field ($E_r$) well upshifts the critical temperature gradient near the magnetic axis, in the weak but not in the strong magnetic shear configuration. The geometric curvature effect significantly influences the $E \times B$ shear and the wave number near the axis, so that the $E_r$ well suppresses the high-n modes but has little effect on the low-n modes, which are suppressed by the weak magnetic shear effect. This new finding unravels the formation mechanism of the internal transport barrier in the weak central magnetic shear discharges.

physics.plasm-ph↗

arXiv2024

Unified mesoscale picture of nonlinear generation of zonal flows in toroidal geometry

Recent experimental findings on limit-cycle-oscillations indicate that the nonlinear driving of the turbulent poloidal Reynolds stress to zonal flows is not a significant factor in the toroidal geometry, sparking fundamental controversial issues within the fusion community. By using the global nonlinear gyrokinetic simulations, we propose a unified mesoscale picture of nonlinear driving of zonal flows in the ion-temperature-gradient turbulence. Zonal flows are nonlinearly driven by the turbulent energy flux and the turbulent poloidal Reynolds stress. The turbulent energy flux is not shielded by the toroidal geometry effect in nonlinearly driving zonal flows. The turbulent poloidal Reynolds stress is not shielded on the time scale shorter than the ion bounce period; however, on the time scale longer than the ion bounce period, the turbulent poloidal Reynolds stress is indeed shielded by the toroidal geometry effect.

physics.plasm-ph↗

arXiv2024

SES: Bridging the Gap Between Explainability and Prediction of Graph Neural Networks

Despite the Graph Neural Networks' (GNNs) proficiency in analyzing graph data, achieving high-accuracy and interpretable predictions remains challenging. Existing GNN interpreters typically provide post-hoc explanations disjointed from GNNs' predictions, resulting in misrepresentations. Self-explainable GNNs offer built-in explanations during the training process. However, they cannot exploit the explanatory outcomes to augment prediction performance, and they fail to provide high-quality explanations of node features and require additional processes to generate explainable subgraphs, which is costly. To address the aforementioned limitations, we propose a self-explained and self-supervised graph neural network (SES) to bridge the gap between explainability and prediction. SES comprises two processes: explainable training and enhanced predictive learning. During explainable training, SES employs a global mask generator co-trained with a graph encoder and directly produces crucial structure and feature masks, reducing time consumption and providing node feature and subgraph explanations. In the enhanced predictive learning phase, mask-based positive-negative pairs are constructed utilizing the explanations to compute a triplet loss and enhance the node representations by contrastive learning.

cs.LG↗