Searcharxiv⌕ Search

arXiv subjects

Peng Shi

Publications and source records attributed to Peng Shi.

At least 37 records · Page 2Linked to original sources
arXiv2026

FutureWeaver: Planning Test-Time Compute for Multi-Agent Systems with Modularized Collaboration

Scaling test-time computation has been shown to significantly improve large language model (LLM) performance without additional training. However, extending these techniques to multi-agent systems remains challenging: existing approaches lack principled mechanisms for allocating compute to enable effective collaboration, scaling coordination itself, or optimizing compute usage under explicit budget constraints. To address this gap, we propose FutureWeaver, a framework for planning and optimizing test-time compute allocation in multi-agent systems under fixed budgets. It introduces collaboration modules, formalized as modular, callable functions that encapsulate reusable multi-agent workflows and are automatically induced via self-play reflection from recurring interaction patterns. Building on these modules, it employs \emph{a dual-level planning architecture} that jointly performs short-horizon action selection and long-horizon abstract lookahead to optimize inference trajectories under budget constraints. Experiments on complex agent benchmarks demonstrate that FutureWeaver consistently outperforms baselines across diverse budget settings, validating its effectiveness for multi-agent collaboration in inference-time optimization.

cs.AI↗
arXiv2026

Adaptive Latent Agentic Reasoning

Large reasoning models improve performance by generating extended chain-of-thought (CoT) reasoning, but this behavior becomes inefficient when applied to LLM agents. Current LLM agents often generate verbose textual reasoning at every decision step and allocate reasoning effort nearly uniformly across turns, leading to substantial inefficiency in multi-turn agentic trajectories. We propose Adaptive Latent Agentic Reasoning (ALAR), a dual-mode framework that uses compact latent reasoning for routine turns and selectively escalates to explicit chain-of-thought when deeper deliberation is needed. ALAR learns latent reasoning by using the agent's actions as supervision anchors and is further optimized to use latent reasoning when it is sufficient for task success and reserve explicit CoT for harder decisions. Experiments on agentic search and tool-use benchmarks show that ALAR maintains comparable or better task accuracy while substantially reducing generated tokens by up to 43.6% in search and 84.6% in tool use. These results demonstrate that ALAR improves the accuracy-efficiency trade-off of LLM agents by reducing unnecessary textual reasoning while preserving explicit deliberation for harder decision steps.

cs.CL↗
arXiv2026

Nitrogen-induced ELM suppression and confinement improvement in the EAST tokamak with a full metal wall

This paper reports the achievement of an ELM-free H-mode regime with confinement improvement enabled by nitrogen (N2) seeding on the Experimental Advanced Superconducting Tokamak (EAST) with a full metal wall. Following N2 injection, large Edge-Localized Mode (ELM) bursts are completely suppressed, while global energy confinement is significantly enhanced, with the H98 factor increasing from approximately 0.9 to 1.2. A distinct edge coherent mode (ECM), localized at the pedestal foot (psi_N ~ 0.99), is identified using O-mode Poloidal Correlation Reflectometry and AXUV diagnostics. This mode operates within a frequency range of 20-50 kHz with a poloidal wavenumber of k_theta ~ 0.54 cm^-1. Linear gyrokinetic simulations performed with the CGYRO code reveal a dominant instability that quantitatively matches the experimental measurements. Detailed scans of parameters identify this mode as a Dissipative Trapped Electron Mode (DTEM). The energy and particle transport driven by this pedestal-foot DTEM effectively regulates the edge gradients, preventing the pedestal from crossing the Peeling-Ballooning stability boundary and sustaining a stationary ELM-free state. These findings provide a physical basis for an integrated scenario to maintain high confinement and protect plasma-facing components in future steady-state fusion reactors.

physics.plasm-ph↗
arXiv2026

On the Essence of Lagrange's Equations

From a new perspective, this paper rederives Lagrange's equations. By applying the chain rule of differentiation, the intrinsic relationship between the momentum theorem and the kinetic energy theorem is first established. Subsequently, expressing the differential form of energy conservation in an arbitrary coordinate system and performing suitable differential operations yields Lagrange's equations. Generalized forces and generalized displacements are shown to be component representations of forces and displacements in a chosen coordinate system. Consequently, the essence of Lagrange's equations is identified as the transformation of the kinetic energy theorem into the momentum theorem via the chain rule for composite functions, thereby revealing how energy conservation constructs momentum conservation.

physics.class-ph↗
arXiv2026

Flexible Entropy Control in RLVR with a Gradient-Preserving Perspective

Reinforcement Learning with Verifiable Rewards (RLVR) has emerged as a critical method for enhancing the reasoning capabilities of Large Language Models (LLMs). However, continuous training often leads to policy entropy collapse, characterized by a rapid decay in entropy that results in premature overconfidence, reduced output diversity, and vanishing gradient norms that inhibit learning. Gradient-Preserving Clipping is a primary factor influencing these dynamics, but existing mitigation strategies are largely static and lack a framework connecting clipping mechanisms to precise entropy control. This paper proposes reshaping entropy control in RL from the perspective of Gradient-Preserving Clipping. We first theoretically and empirically verify the contributions of specific importance sampling ratio regions to entropy growth and reduction. Leveraging these findings, we introduce a novel regulation mechanism using dynamic clipping thresholds to precisely manage entropy. Furthermore, we design and evaluate dynamic entropy control strategies, including increase-then-decrease, decrease-increase-decrease, and oscillatory decay. Experimental results demonstrate that these strategies effectively mitigate entropy collapse and achieve superior performance across multiple benchmarks.

cs.LG↗
arXiv2026

Constructing Four-Body Ballistic Lunar Transfers via Analytical Energy Conditions

The current lunar exploration, particularly the regular large-scale cargo transportation in the Earth-Moon system proposes requirements for amount of low-energy lunar transfers. The conventional grid-search method to construct low-energy lunar transfers suffers from extensive computational effort and large-scale searches. To further improve the method, this paper focuses on one type of low-energy lunar transfers termed ballistic lunar transfers, and derives prior knowledge to narrow the scale of searches. The Sun-Earth/Moon planar bicircular restricted four-body problem (PBCR4BP) is adopted as the dynamical model to construct lunar transfers. First, the analytical conditions for ballistic capture are derived and summarized in form of exact ranges of the Jacobi energy at the lunar insertion point. Both sufficient and necessary condition and necessary condition are developed. These conditions suggest an important role of the Sun-Earth/Moon PBCR4BP rather than the Earth-Moon planar restricted three-body problem in achieving lunar ballistic capture. Then, a grid-search method combined with the analytical energy conditions is proposed to construct ballistic lunar transfers. Simulations shows that a high ballistic capture ratio is achieved by the proposed method (99.87% for direct insertion and 98.72% for retrograde insertion). Examining the obtained ballistic lunar transfers, the effectiveness of the analytical energy conditions is verified. Samples of our obtained lunar transfers achieves lower or comparable impulses compared to solutions obtained in the previous works. Solutions belonging to new or less-reported transfer families are also presented, and the potential engineering applications of these trajectories are briefly discussed.

astro-ph.EP↗
arXiv2026

CAIRO: Decoupling Order from Scale in Regression

Standard regression methods typically optimize a single pointwise objective, such as mean squared error, which conflates the learning of ordering with the learning of scale. This coupling renders models vulnerable to outliers and heavy-tailed noise. We propose CAIRO (Calibrate After Initial Rank Ordering), a framework that decouples regression into two distinct stages. In the first stage, we learn a scoring function by minimizing a scale-invariant ranking loss; in the second, we recover the target scale via isotonic regression. We theoretically characterize a class of "Optimal-in-Rank-Order" objectives -- including variants of RankNet and Gini covariance -- and prove that they recover the ordering of the true conditional mean under mild assumptions. We further show that subsequent monotone calibration recovers the true regression function at the population level and mathematically guarantees that finite-sample predictions are strictly auto-calibrated. Empirically, CAIRO combines the representation learning of neural networks with the robustness of rank-based statistics. It matches the performance of state-of-the-art tree ensembles on tabular benchmarks and significantly outperforms standard regression objectives in regimes with heavy-tailed or heteroskedastic noise.

stat.ME↗
arXiv2026

Data Mining-Based Cislunar Escape-Family Analysis in The Multi-Body Models

Escape trajectories from the Earth-Moon system play an important role in interplanetary transfer. This paper focuses on the escape trajectories from a 167 km circular Earth orbit in the Earth-Moon planar circular three-body problem and the Sun-Earth/Moon planar bicircular four-body problem and is denoted to providing a comprehensive analysis on these escape trajectories. To achieve these purposes, the global maps of escape trajectories are constructed, and escape trajectories with one lunar gravity assist are pre-filtered. Then, an effective method to identification escape families is proposed based on dynamical analysis and data mining techniques. Once the escape families are identified, the corresponding characteristics are analyzed to provide insights into the construction of escape trajectories. Based on these escape families, the effects of the solar gravity perturbation on the number of escape trajectories, the emergence and disappearance of escape families, variation in generalized energy, and transfer characteristics are further summarized, providing insights into the model selection in the escape trajectory construction. This paper establishes an analysis methodology of escape trajectories from a perspective of escape families, deepening the understanding of escape dynamics.

astro-ph.EP↗
arXiv2026

TreeCUA: Efficiently Scaling GUI Automation with Tree-Structured Verifiable Evolution

Effectively scaling GUI automation is essential for computer-use agents (CUAs); however, existing work primarily focuses on scaling GUI grounding rather than the more crucial GUI planning, which requires more sophisticated data collection. In reality, the exploration process of a CUA across apps/desktops/web pages typically follows a tree structure, with earlier functional entry points often being explored more frequently. Thus, organizing large-scale trajectories into tree structures can reduce data cost and streamline the data scaling of GUI planning. In this work, we propose TreeCUA to efficiently scale GUI automation with tree-structured verifiable evolution. We propose a multi-agent collaborative framework to explore the environment, verify actions, summarize trajectories, and evaluate quality to generate high-quality and scalable GUI trajectories. To improve efficiency, we devise a novel tree-based topology to store and replay duplicate exploration nodes, and design an adaptive exploration algorithm to balance the depth (\emph{i.e.}, trajectory difficulty) and breadth (\emph{i.e.}, trajectory diversity). Moreover, we develop world knowledge guidance and global memory backtracking to avoid low-quality generation. Finally, we naturally extend and propose the TreeCUA-DPO method from abundant tree node information, improving GUI planning capability by referring to the branch information of adjacent trajectories. Experimental results show that TreeCUA and TreeCUA-DPO offer significant improvements, and out-of-domain (OOD) studies further demonstrate strong generalization. All trajectory node information and code will be available at https://github.com/UITron-hub/TreeCUA.

cs.CV↗
arXiv2026

Length-Unbiased Sequence Policy Optimization: Revealing and Controlling Response Length Variation in RLVR

Recent applications of Reinforcement Learning with Verifiable Rewards (RLVR) to Large Language Models (LLMs) and Vision-Language Models (VLMs) have demonstrated significant success in enhancing reasoning capabilities for complex tasks. During RLVR training, an increase in response length is often regarded as a key factor contributing to the growth of reasoning ability. However, the patterns of change in response length vary significantly across different RLVR algorithms during the training process. To provide a fundamental explanation for these variations, this paper conducts an in-depth analysis of the components of mainstream RLVR algorithms. We present a theoretical analysis of the factors influencing response length and validate our theory through extensive experimentation. Building upon these theoretical findings, we propose the Length-Unbiased Sequence Policy Optimization (LUSPO) algorithm. Specifically, we rectify the length bias inherent in Group Sequence Policy Optimization (GSPO), rendering its loss function unbiased with respect to response length and thereby resolving the issue of response length collapse. We conduct extensive experiments across mathematical reasoning benchmarks and multimodal reasoning scenarios, where LUSPO consistently achieves superior performance. Empirical results demonstrate that LUSPO represents a novel, state-of-the-art optimization strategy compared to existing methods such as GRPO and GSPO.

cs.CL↗
arXiv2026

Metis-SPECS: Decoupling Multimodal Learning via Self-distilled Preference-based Cold Start

Reinforcement learning (RL) with verifiable rewards has recently catalyzed a wave of "MLLM-r1" approaches that bring RL to vision language models. Most representative paradigms begin with a cold start, typically employing supervised fine-tuning (SFT), to initialize the policy before RL. However, SFT-based cold start adopts the reasoning paradigm intertwined with task solution and output format, which may induce instruction-style overfitting, weakens out-of-distribution generalization, and ultimately affects downstream RL. We revisit the cold start along two views, its training method and data construction, and introduce the Generalization Factor (GF) coefficient to quantify the generalization capability under different methods. Our empirical study finds that preference-based training methods (e.g. DPO) generalizes better than SFT-based methods in cold start. Motivated by this, we propose SPECS-a Self-distilled, Preference-based Cold Start framework that decouples multimodal learning: (1) generates introspective preference data pairs via self-distillation, avoiding reliance on larger teachers or manual annotation; (2) performs preference-based training to learn, focusing on shallow, transferable surface-form criteria (format, structure, style) rather than memorizing content; and (3) hands off to RL with verifiable rewards for deep reasoning results. Experimental results across multiple multimodal benchmarks show that our decoupling learning framework yields consistent performance gains over strong baselines, improving MEGA-Bench by 4.1% and MathVista by 12.2%. Additional experiments indicate that SPECS contributes to reducing in-distribution "stuckness," improving exploration, stabilizing training, and raising the performance ceiling. Project Page: https://kwen-chen.github.io/SPECS-VL/

cs.LG↗
arXiv2026

Quality or Quantity? Error-Informed Selective Online Learning with Gaussian Processes in Multi-Agent Systems: Extended Version

Effective cooperation is pivotal in distributed learning for multi-agent systems, where the interplay between the quantity and quality of the machine learning models is crucial. This paper reveals the irrationality of indiscriminate inclusion of all models on agents for joint prediction, highlighting the imperative to prioritize quality over quantity in cooperative learning. Specifically, we present the first selective online learning framework for distributed Gaussian process (GP) regression, namely distributed error-informed GP (EIGP), that enables each agent to assess its neighboring collaborators, using the proposed selection function to choose the higher quality GP models with less prediction errors. Moreover, algorithmic enhancements are embedded within the EIGP, including a greedy algorithm (gEIGP) for accelerating prediction and an adaptive algorithm (aEIGP) for improving prediction accuracy. In addition, approaches for fast prediction and model update are introduced in conjunction with the error-informed quantification term iteration and a data deletion strategy to achieve real-time learning operations. Numerical simulations are performed to demonstrate the effectiveness of the developed methodology, showcasing its superiority over the state-of-the-art distributed GP methods with different benchmarks.

cs.LG↗
arXiv2025

Learning When to Look: A Disentangled Curriculum for Strategic Perception in Multimodal Reasoning

Multimodal Large Language Models (MLLMs) demonstrate significant potential but remain brittle in complex, long-chain visual reasoning tasks. A critical failure mode is "visual forgetting", where models progressively lose visual grounding as reasoning extends, a phenomenon aptly described as "think longer, see less". We posit this failure stems from current training paradigms prematurely entangling two distinct cognitive skills: (1) abstract logical reasoning "how-to-think") and (2) strategic visual perception ("when-to-look"). This creates a foundational cold-start deficiency -- weakening abstract reasoning -- and a strategic perception deficit, as models lack a policy for when to perceive. In this paper, we propose a novel curriculum-based framework to disentangle these skills. First, we introduce a disentangled Supervised Fine-Tuning (SFT) curriculum that builds a robust abstract reasoning backbone on text-only data before anchoring it to vision with a novel Perception-Grounded Chain-of-Thought (PG-CoT) paradigm. Second, we resolve the strategic perception deficit by formulating timing as a reinforcement learning problem. We design a Pivotal Perception Reward that teaches the model when to look by coupling perceptual actions to linguistic markers of cognitive uncertainty (e.g., "wait", "verify"), thereby learning an autonomous grounding policy. Our contributions include the formalization of these two deficiencies and the development of a principled, two-stage framework to address them, transforming the model from a heuristic-driven observer to a strategic, grounded reasoner. \textbf{Code}: \url{https://github.com/gaozilve-max/learning-when-to-look}.

cs.CV↗
arXiv2025

Reading or Reasoning? Format Decoupled Reinforcement Learning for Document OCR

Reading text from images or scanned documents via OCR models has been a longstanding focus of researchers. Intuitively, text reading is perceived as a straightforward perceptual task, and existing work primarily focuses on constructing enriched data engineering to enhance SFT capabilities. In this work, we observe that even advanced OCR models exhibit significantly higher entropy in formatted text (\emph{e.g.}, formula, table, etc.) compared to plain text, often by an order of magnitude. These statistical patterns reveal that advanced OCR models struggle with high output uncertainty when dealing with format sensitive document, suggesting that reasoning over diverse reading pathways may improve OCR performance. To address this, we propose format decoupled reinforcement learning (FD-RL), which leverages high-entropy patterns for targeted optimization. Our approach employs entropy-based data filtration strategy to identify format-intensive instances, and adopt format decoupled rewards tailored to different format types, enabling format-level validation rather than token-level memorization. FD-RL achieves an average score of 90.41 on OmniDocBench, setting a new record for end-to-end models on this highly popular benchmark. More importantly, we conduct comprehensive ablation studies over data, training, filtering, and rewarding strategies, thoroughly validating their effectiveness.

cs.CV↗
arXiv2025

LongCat-Flash-Omni Technical Report

We introduce LongCat-Flash-Omni, a state-of-the-art open-source omni-modal model with 560 billion parameters, excelling at real-time audio-visual interaction. By adopting a curriculum-inspired progressive training strategy that transitions from simpler to increasingly complex modality sequence modeling tasks, LongCat-Flash-Omni attains comprehensive multimodal capabilities while maintaining strong unimodal capability. Building upon LongCat-Flash, which adopts a high-performance Shortcut-connected Mixture-of-Experts (MoE) architecture with zero-computation experts, LongCat-Flash-Omni integrates efficient multimodal perception and speech reconstruction modules. Despite its immense size of 560B parameters (with 27B activated), LongCat-Flash-Omni achieves low-latency real-time audio-visual interaction. For training infrastructure, we developed a modality-decoupled parallelism scheme specifically designed to manage the data and model heterogeneity inherent in large-scale multimodal training. This innovative approach demonstrates exceptional efficiency by sustaining over 90% of the throughput achieved by text-only training. Extensive evaluations show that LongCat-Flash-Omni achieves state-of-the-art performance on omni-modal benchmarks among open-source models. Furthermore, it delivers highly competitive results across a wide range of modality-specific tasks, including text, image, and video understanding, as well as audio understanding and generation. We provide a comprehensive overview of the model architecture design, training procedures, and data strategies, and open-source the model to foster future research and development in the community.

cs.MM↗
arXiv2025

Metis-HOME: Hybrid Optimized Mixture-of-Experts for Multimodal Reasoning

Inspired by recent advancements in LLM reasoning, the field of multimodal reasoning has seen remarkable progress, achieving significant performance gains on intricate tasks such as mathematical problem-solving. Despite this progress, current multimodal large reasoning models exhibit two key limitations. They tend to employ computationally expensive reasoning even for simple queries, leading to inefficiency. Furthermore, this focus on specialized reasoning often impairs their broader, more general understanding capabilities. In this paper, we propose Metis-HOME: a Hybrid Optimized Mixture-of-Experts framework designed to address this trade-off. Metis-HOME enables a ''Hybrid Thinking'' paradigm by structuring the original dense model into two distinct expert branches: a thinking branch tailored for complex, multi-step reasoning, and a non-thinking branch optimized for rapid, direct inference on tasks like general VQA and OCR. A lightweight, trainable router dynamically allocates queries to the most suitable expert. We instantiate Metis-HOME by adapting the Qwen2.5-VL-7B into an MoE architecture. Comprehensive evaluations reveal that our approach not only substantially enhances complex reasoning abilities but also improves the model's general capabilities, reversing the degradation trend observed in other reasoning-specialized models. Our work establishes a new paradigm for building powerful and versatile MLLMs, effectively resolving the prevalent reasoning-vs-generalization dilemma. Code and weights are available at https://github.com/MM-Thinking/Metis-HOME.

cs.CV↗
arXiv2025

Perceptual-Evidence Anchored Reinforced Learning for Multimodal Reasoning

Reinforcement Learning with Verifiable Rewards (RLVR) has significantly advanced the reasoning capabilities of Large Language Models (LLMs) and is now being applied to Vision-Language Models (VLMs). However, vanilla RLVR for VLMs verifies only the final textual output, critically neglecting the foundational step of visual perception. This oversight leads to visual hallucinations and reward hacking, as reasoning built upon flawed perception is inherently unreliable. To address this, we propose PEARL (Perceptual-Evidence Anchored Reinforced Learning), a dual-branch, perception-reasoning synergistic that strengthens multimodal reasoning by explicitly anchoring it to verified visual evidence. For each reasoning-oriented QA instance, PEARL first derive a perception checklist -- a set of perception-oriented sub-questions with verifiable answers that probe the model's understanding of key visual evidence. During training, auxiliary rollouts on this checklist yield a perceptual reward that both directly reinforces the model's perception ability and acts as a fidelity gate for reasoning. If the model passes the perception check, its policy update is biased towards evidence-anchored reasoning. Otherwise, the process is halted to prevent reasoning from flawed premises. PEARL can be seamlessly integrated with popular RL methods like GRPO and DAPO. Comprehensive experiments show PEARL achieves substantial gains on multimodal reasoning benchmarks, e.g., a +9.7% improvement over the baseline and +6.6% over GRPO on MathVerse.

cs.CV↗
arXiv2025

Lossless Compression: A New Benchmark for Time Series Model Evaluation

The evaluation of time series models has traditionally focused on four canonical tasks: forecasting, imputation, anomaly detection, and classification. While these tasks have driven significant progress, they primarily assess task-specific performance and do not rigorously measure whether a model captures the full generative distribution of the data. We introduce lossless compression as a new paradigm for evaluating time series models, grounded in Shannon's source coding theorem. This perspective establishes a direct equivalence between optimal compression length and the negative log-likelihood, providing a strict and unified information-theoretic criterion for modeling capacity. Then We define a standardized evaluation protocol and metrics. We further propose and open-source a comprehensive evaluation framework TSCom-Bench, which enables the rapid adaptation of time series models as backbones for lossless compression. Experiments across diverse datasets on state-of-the-art models, including TimeXer, iTransformer, and PatchTST, demonstrate that compression reveals distributional weaknesses overlooked by classic benchmarks. These findings position lossless compression as a principled task that complements and extends existing evaluation for time series modeling.

cs.LG↗