F Zhang

Xinyu Geng, Xuanhua He, Sixiang Chen et al. · tencent-ai

Training tool-use agents to improve from their own experience remains challenging, as supervised fine-tuning relies on fixed teacher-distilled trajectories, while sparse-reward reinforcement learning provides weak supervision for long-horizon interactions. We present DeepSearch-Evolve, a self-distillation framework for web agents built on DeepSearch-World, a deterministic and verifiable environment with reproducible search and page-reading tools. DeepSearch-World contains 420K multi-hop QA tasks constructed from entity-level random walks and supports key agentic cognitive behaviors useful for self-evolving, including progress verification, grounded reflection, and failure recovery. DeepSearch-Evolve iteratively performs trajectory generation, filtering, data mixing, and fine-tuning to train stronger agents. Without distillation from more capable models, DeepSearch-World-9B achieves competitive performance compared with open-source agents, reaching 31.2% on BrowseComp, 61.5% on GAIA, and 93.4% on HotpotQA, showing that verifiable environments enable scalable self-evolution for long-horizon web agents. We will release the environment, 420K training pool, validation set, model, and code to facilitate future research on self-improving deep search agents.

Jinyang Wu, Shuo Yang, Zhengxi Lu et al.

Large language models are increasingly trained as interactive agents for long-horizon tasks involving multi-turn interaction, tool use, and environment feedback. Outcome-based reinforcement learning (RL) provides a practical optimization paradigm, but its sparse trajectory-level rewards offer limited guidance on intermediate decisions, leaving a supervision gap between episode-level outcomes and token-level policy learning. We propose SEED (SElf-Evolving On-Policy Distillation), a self-evolving framework that converts completed on-policy trajectories into training-time hindsight skills and distills their behavioral effect back into the policy model. SEED first fine-tunes the policy to analyze completed trajectories and generate natural-language skills that capture reusable workflows, decisive observations, or failure-avoidance rules. During RL, the current policy both collects trajectories and serves as the analyzer that extracts hindsight skills from them. Policy updates therefore improve subsequent decision making and skill analysis together, allowing hindsight supervision to evolve with the policy. SEED then re-scores the sampled actions under ordinary and skill-augmented contexts, converting the skill-induced probability shift into a dense token-level on-policy distillation signal. This signal is jointly optimized with outcome-based RL, keeping the auxiliary supervision aligned with the current trajectory distribution. Extensive experiments on text-based and vision-based agentic tasks show that SEED consistently improves performance and sample efficiency, exhibiting robust generalization to unseen scenarios. Our code is available at https://github.com/jinyangwu/SEED.

Zhuo Ye, Feng Zhang, Maxim Moraru et al.

Exa-PD is a highly parallelizable workflow designed for the construction of multi-element phase diagrams (PDs). It uses standard sampling techniques, molecular dynamics (MD) and Monte Carlo (MC) as implemented in the LAMMPS package, to simultaneously sample multiple phases over a fine temperature-composition mesh for free-energy calculations. Parsl serves as the global workflow engine, coordinating large ensembles of MD and MC tasks to achieve massive parallelization with strong scalability. The resulting free energies of liquid and solid phases are then fed to CALPHAD modeling via the PyCalphad package to construct multi-element PDs.

Aiwei Liu, Cheng Shi, Chuhan Wu et al.

Scaling Large Language Models (LLMs) has been driven mainly by enlarging the Transformer backbone, but for an already-strong model this requires another round of costly pretraining. We study whether an existing backbone can keep improving by allocating more computation to each token while leaving the Transformer backbone fixed. Depth-recurrent (looped) Transformers pursue this goal but are hard to scale, because looped computation does not fit naturally with the pipeline parallelism used to train the largest models. We add computation along the sequence-length dimension, where the extra computation is simply a longer input and stays compatible with standard large-model training. We propose Hidden Decoding, a sequence-length scaling method applied during continued pretraining (CPT). It expands each token into n streams with independent embedding tables and keeps the intermediate streams' key-value cache as context, so each token performs more internal computation without adding or widening Transformer layers. To keep this affordable at scale, we introduce Stream-Factorized Attention, in which most layers attend only within each stream and only a few layers mix across streams, reducing the attention cost from quadratic to roughly linear in n. Experiments support two scaling results. At frontier scale, we train WeLM-HD4-80B and WeLM-HD4-617B at n=4 and improve their matched non-HD baselines, making Hidden Decoding the first demonstrated sequence-length scaling method at the 100B+ MoE scale. Across expansion factors, the gains grow as n increases, showing that sequence-length expansion is a practical fixed-backbone scaling path for frontier-scale LLMs.

Fengzhuo Zhang, Zhuoran Yang, Dirk Bergemann

Large Language Models (LLMs) have revolutionized AI services, but a critical tension emerges: while personalization improves model performance, it consumes scarce computational resources that users must share. When should a user invest in expensive Supervised Fine-Tuning (SFT) versus lightweight In-Context Learning (ICL)? How does congestion from other users' personalization choices reshape these incentives? And what strategies should platforms adopt when offering multiple personalization algorithms? We develop a tractable framework for LLM serving that captures the statistical-economic trade-offs users face. Our analysis yields several surprising insights. First, we show that ICL and SFT dominate in different regimes, determined by an interplay between pretraining coverage and data signal-to-noise ratios, but congestion can flip these rankings. Second, equilibrium resource consumption exhibits pronounced non-monotonicity: improving pretraining precision reduces the congestion, while broader pretraining coverage and harder tasks sometimes increase it. Third, we prove that offering both personalization methods never hurts the platform's maximal profits, despite potentially increasing computational load. Experiments with GPT-2 on linear regression tasks validate our theoretical predictions about algorithm performance. Complementing these results, our review of documentation from 21 major AI platforms shows that the share offering both SFT and ICL increased from 9.5% in 2021 to 71.4% in 2025, consistent with our platform-design implications.

Jing-Xiao Liao, Tianwei Zhang, Yu-Hao Jiang et al.

The pursuit of autonomously self-improving models has attracted growing interest in the era of large-scale foundation models. Drawing inspiration from the concept of "enlightenment" or "aha moment" in human brain, we hypothesize that large models exhibit an analogous enlightenment phenomenon-a latent capacity for sudden capability boost. Then, we propose Enlightenment, a novel training-free post-tuning paradigm for large-scale models. Our approach modifies shortcuts for key modules/layers without weight updates, while existing training-free ones predominantly manipulate attention weights. We introduce two architecture-specific instantiations: i) For large language models, we propose attention head-mixing shortcuts that recalibrate attention weights by linking the initial attention head's output to all other target heads, modulated by an adaptive scaling factor initialization strategy. ii) For vision-language models, we apply a lightweight scalar-modulated factor to residual connections in the decoder layers, regulating information flow. Extensive experiments show that Enlightenment efficiently unlocks the latent potential of pre-trained networks, yielding remarkable performance improvements across diverse benchmarks and models.

Qi Lou, Yijun Xu, Yang Cao et al.

The computational speed of electromagnetic transient programs (EMTP) is severely limited by both the curse of dimensionality and the ill-conditioned system matrix, which collectively degrade solver performance. However, existing research on EMTP acceleration has largely overlooked the issue of ill-conditioning. This letter presents a first systematic, EMT-oriented investigation of the ill-conditioning of the EMTP admittance matrix by establishing a link between its physical origins and mathematical pathologies, thereby revealing the underlying mechanism by which network topology induces ill-conditioning. Building upon these structural insights, a preconditioner-based strategy is developed that significantly accelerates computation while preserving numerical accuracy. Simulation results demonstrate the outstanding efficiency and robustness of the proposed approach.

Yanbing Shen, Fan Zhang, Haitao Xu

Firmware rehosting executes firmware images in emulated environments such as QEMU to enable scalable dynamic analysis of Internet of Things (IoT) devices. In practice, rehosting pipelines remain fragile across diverse real-world firmware images, as reaching an externally observable execution state depends on tightly coupled artifacts spanning boot scripts, persistent configuration (e.g., NVRAM-like key-value state), and network setup. Template-driven frameworks often fail to accommodate long-tail vendor conventions, while unconstrained use of large language models (LLMs) risks unsupported modifications and irreproducible executions. We introduce FirmPilot, an evidence-guided multi-agent framework for environment recovery in firmware rehosting. FirmPilot reformulates rehosting as iterative environment reconstruction in which a search agent grounds decisions through similarity-based retrieval, a planner coordinates execution-accepted transitions, and specialized agents recover filesystem/init artifacts, persistent state, and network exposure. Through repeated execution and evidence-grounded artifact deltas, the system resolves cross-layer dependencies across boot, state, and networking that otherwise prevent firmware executions from reaching a stable, externally reachable state in emulation. Evaluated on the large-scale, real-world LFwC firmware corpus, FirmPilot improves web-service reachability over FirmAE from 25.49% to 52.39% and network reachability from 39.30% to 71.93%. The resulting rehosts raise the average number of detected services per firmware from 0.86 to 1.62 and support downstream analysis workflows, including RouterSploit interaction and protocol-aware fuzzing over recovered service surfaces. The evaluation shows that evidence- and feedback-grounded agent coordination improves rehosting success, service recovery, and downstream utility in automated firmware rehosting.

Yongjian Zhu, Yusen Tao, Feitian Zhang

Miniature hybrid underwater gliders have attracted increasing attention for long-endurance ocean observation and confined-space inspection. Large-range wing reconfiguration offers a promising yet largely unexplored approach for simultaneously enhancing maneuverability and shape adaptability in constrained underwater environments. However, such morphing introduces substantial challenges in mechanical integration, dynamic modeling, and hydrodynamic characterization. This paper presents FoDeGlider, a miniature hybrid underwater glider equipped with two independently actuated wings capable of large-range folding and deflection. To capture configuration-dependent variations in mass distribution, center-of-geometry location, and hydrodynamic loading, a multibody dynamics model is developed by treating wing configuration as a structural variable. A composite rigid body algorithm (CRBA)-based projection formulates the composite inertia, wrench transformations, and component-level hydrodynamics into a unified Fossen-form dynamic model applicable to arbitrary wing configurations. A sequential parameter-identification framework is further proposed to estimate fuselage and wing hydrodynamic coefficients, resulting in an open benchmark dataset for model identification and validation. Extensive experiments are conducted, the results of which demonstrate accurate dynamic modeling and parameter identification across diverse morphing configurations. Gate traversal experiments further validate FoDeGlider's ability to actively reconfigure its morphology during locomotion, enabling enhanced navigation in confined underwater environments.

Jiyue Tao, Shunheng Xin, Tongsheng Shen et al.

Decentralized multi-robot coordination under partial observability remains challenging, especially in communication-free settings where agents must act solely from local sensor observations. Privileged policy distillation provides a promising approach by transferring knowledge from a globally informed oracle to sensor-constrained students. However, in multi-agent systems, the same local observation may correspond to multiple global configurations requiring qualitatively different cooperative actions, making the conditional action distribution inherently multi-modal. Standard deterministic distillation collapses these modes to their mean, often yielding invalid or hesitant actions. To address this issue, we propose CoDiMAD, a three-stage framework that trains a privileged oracle with MAPPO, constructs an offline dataset of local-observation-oracle-action pairs, and distills the oracle into decentralized students parameterized as conditional denoising diffusion probabilistic models. By approximating the conditional oracle-action distribution through the diffusion reverse process, CoDiMAD samples decisive actions from coherent coordination modes rather than averaging across them. Theoretical analysis characterizes the mode-averaging failure of deterministic distillation and the distributional recovery property of diffusion-based distillation. Experiments on three cooperative tasks show that CoDiMAD consistently outperforms direct local MARL and deterministic distillation baselines. The source code will be made publicly available upon acceptance.

Ziqi Chen, Yingli Zhou, Fangyuan Zhang et al.

Leveraging large language models (LLMs) to analyze complex documents -- such as academic papers, technical manuals, and financial reports -- has emerged as a mainstream and critical task in both research and industry. In practice, users must first filter relevant documents from large collections and then conduct in-depth analysis (e.g. question answering) over the selected subset, yet existing systems flatten documents into plain-text chunks, discarding the rich hierarchical structures (sections, tables, figures, equations) and degrading downstream performance. We present DocMaster, a hierarchical structure-aware document analysis system. DocMaster parses documents into hierarchical document trees preserving original layouts and constructs a structure-aware semantic index that enables accurate document filtering and in-depth analysis. We demonstrate DocMaster through an interactive web interface that enables users to upload document collections, construct tree-based and multi-view semantic indices, filter relevant documents via natural-language conditions, and perform follow-up question answering over the filtered results. The source code, data, and demo are available at https://doc-master.github.io/.

Xin Li, Jiaju Han, Ma Yaqi et al.

Infrared vision-language models are increasingly used for perception under low-light and adverse visual conditions, yet their robustness to localized structured perturbations remains underexplored. Existing infrared adversarial studies mainly focus on object detectors, leaving the security of infrared vision-language models less systematically examined. We present InfraQR, a QR-inspired structured patch attack for infrared vision-language models. Unlike localized attacks that attach perturbations to the target object, InfraQR places a compact structured patch along image boundaries and optimizes learnable grid cells through surrogate CLIP-style encoders. The resulting patch has a near-binary structured appearance, but is not required to be a valid or machine-readable QR code. We evaluate InfraQR on infrared classification, caption transfer, and question-answer-aware visual question answering (VQA) tasks. On a 300-image infrared benchmark, InfraQR sharply reduces the accuracy of multiple CLIP-style classifiers, including reducing OpenAI CLIP accuracy from 98.67% to 0.70%. The generated adversarial images also transfer to black-box captioning and VQA models, causing semantic degradation in captions and more error-prone answers under GPT-5.4-based evaluation. These results show that infrared vision-language models remain vulnerable to structured edge-placed perturbations, motivating further study of cross-task robustness beyond direct object occlusion.

Fan Zhang, Richie Suganda, Jinfeng Chen et al.

A learning-enabled disturbance-rejection framework based on a Neural Extended State Observer (Neural-ESO) is presented in this letter. Unlike existing learning-based control methods that largely rely on the learned model once deployed, Neural-ESO adopts a dual-pathway architecture: a predictive pathway uses a neural network to provide a feedforward disturbance estimate that accelerates convergence, while a corrective pathway employs a conventional ESO to compensate prediction errors and prevent over-reliance on the neural component. Using Lyapunov theory and a small-gain analysis, we show that enforcing a Lipschitz bound on the learning component guarantees uniform ultimate boundedness of the closed-loop error dynamics. The proposed framework is validated on a quadrotor landing task subject to strong ground-effect disturbances across normal and out-of-distribution scenarios, demonstrating accuracy-robustness trade-off and greater operational reliability during training, deployment, and transfer compared with state-of-the-art baselines.