arXiv subjects

Fei Yu

Publications and source records attributed to Fei Yu.

At least 19 recordsLinked to original sources

arXiv2026

Invert4TVG: A Temporal Video Grounding Framework with Inversion Tasks Preserving Action Understanding Ability

Temporal Video Grounding (TVG) aims to localize video segments corresponding to a given textual query, which often describes human actions. However, we observe that current methods, usually optimizing for high temporal Intersection-over-Union (IoU), frequently struggle to accurately recognize or understand the underlying actions in both the video and query, thus reducing the effectiveness of these methods. To address this, we propose a novel TVG framework that integrates inversion-based TVG as auxiliary objectives to maintain the model's action understanding ability. We introduce three kinds of inversion TVG tasks derived from the original TVG annotations: (1) Verb Completion, predicting masked verbs (actions) in queries given video segments; (2) Action Recognition, identifying query-described actions; and (3) Video Description, generating descriptions containing query-relevant actions given video segments. These inversion tasks are entirely derived from the original TVG tasks and are probabilistically integrated with them within a reinforcement learning framework. By leveraging carefully designed reward functions, the model preserves its ability to understand actions, thereby improving the accuracy of temporal grounding. Experiments show our method outperforms state-of-the-art approaches, achieving a 7.1\% improvement in R1@0.7 on Charades-STA for a 3B model.

cs.AI↗

arXiv2026

Test-Time Policy Adaptation for Enhanced Multi-Turn Interactions with LLMs

Large Language Models (LLMs) employ multi-turn interaction as a fundamental paradigm for completing complex tasks. However, their performance often degrades in extended interactions, as they are typically trained on static, single-turn data, which hinders their ability to adapt to real-time user feedback. To address this limitation, we first propose a new paradigm: Test-Time Policy Adaptation for Multi-Turn Interactions (T2PAM), which utilizes user feedback from the ongoing interaction as a reward signal to estimate a latent optimal policy aligned with user preferences, then updates a small subset of parameters to steer the model toward this policy, ultimately enabling efficient in-conversation self-correction. We then introduce Optimum-Referenced One-Step Adaptation (ROSA), a lightweight algorithm that operationalizes T2PAM. ROSA guides the model parameters toward a theoretical optimal policy in a single, efficient update step, avoiding costly iterative gradient-based optimization and minimizing computational overhead. We provide a rigorous theoretical analysis guaranteeing that the policy of ROSA converges to the preference of user as the number of interactions increases. Extensive experiments on challenging benchmark demonstrate that ROSA achieves significant improvements in both task effectiveness and efficiency.

cs.CL↗

arXiv2026

OptiCom : A Unified Framework for State-Conditioned Composition in LLM-Driven Optimization

Large language models (LLMs) are increasingly deployed to solve complex scientific and practical problems via iterative optimization. However, dynamically coordinating diverse search mechanisms as candidate quality, failure modes, and resource budgets evolve remains a critical open challenge. Targeted empirical diagnostics reveal that mechanism effectiveness is highly state-dependent. Motivated by this, we analyze how individual decisions drive final outcomes, decomposing the expected terminal improvement under a shared budget into cumulative decision opportunities minus cumulative selection losses. Guided by this opportunity-loss theoretical foundation, we propose OptiCom, a unified framework that represents LLM-driven optimizers within a shared configuration space: C=(A,Q,O,E,M,S), corresponding to artifact, query, operator, evaluation, memory, and strategy. Operating within this space, a fast LLM-based Optimization Controller dynamically composes immediate mechanisms through structured Action Packages, while a slower Strategy Adapter refines long-term selection preferences, operator weights, and templates based on accumulated trajectory feedback. Comprehensive evaluations across 32 benchmark groups demonstrate the superiority of framework: OptiCom achieves an average Max-score rank of 1.72 among 14 evaluated configurations, securing the top score in 23 groups. Ultimately, these results highlight the broad applicability and high extensibility of OptiCom as a general-purpose paradigm for robust LLM test-time scaling.

cs.AI↗

arXiv2026

Freezing Mechanism and Topology Dependence of Information Flow Structure in Small-Scale Particle Swarm Optimization

The neighborhood topology of particle swarm optimization (PSO) determines information propagation and regulates the exploration-exploitation balance. However, existing evaluation methods based on fitness and spatial diversity fail to characterize the dynamic evolution internal interaction structure of the population. To address this gap, this paper focuses on small-scale PSO (20-50 particles) with regular topologies and models the iterative process as a dynamic weighted influence graph. We construct a multi-scale information-flow evaluation framework covering micro-, meso-, and macro-level metrics, and propose a freeze-point detection method based on effective nodes, a macro-level distributional metric. Experiments on CEC2017 benchmark functions show that the information flow structure freezes early in the search (3 percent - 22 percent for representative configurations and 15 percent-49 percent for global medians) and remains unchanged thereafter. The freeze time follows the order Gbest$<$ Von Neumann $<$Ring, consistent across standard, LDIW, and TVAC variants. Parameter-adaptive variants delay freezing without altering the topological order, and the freeze point is significantly negatively correlated with final solution quality. Intervention experiments under Gbest show that weakening hub nodes delays freezing and improves performance on multimodal functions, while strengthening hubs produces no additional deterioration due to saturation, revealing a decoupling between structural and performance effects. Among the three metric levels, only macro-level distributional metrics effectively capture freezing under regular topologies, because micro- and meso-level structures are predetermined by the topology. These findings reveal the freezing mechanism of information flow in small-scale PSO and provide structural prior signals for adaptive topology control.

cs.NE↗

arXiv2026

SecJev: Bringing Security Expertise to System One Decision Models

Security workflows need models that turn complex observations and explicit policies into decisions. System One models introduced by Jev return typed predictions and probabilities; security specialization supplies the domain expertise behind those predictions. We introduce SecJev, to our knowledge the first family of Jev-style decision models specialized for security, spanning 0.8B to 9B parameters. Built on Kev's single-pass candidate scorer, SecJev learns Boolean, choice, and ordered decisions from text, telemetry, and observation histories. We develop SecJev-Corpus to unify source-label prediction and explicit-policy evaluation across 14 tasks and eight sources. It covers tool outputs, traffic, federated updates, consensus, authentication, and vehicle messages. Scene-weighted training adapts the models across these domains while preserving a shared typed decision interface. Security specialization improves every model in the family; SecJev-0.8B outperforms general Kev-9B by 20.51 percentage points in task-macro accuracy. At 0.8B and 4B, SecJev achieves accuracy and latency close to answer-only generative fine-tuning with lower peak allocated GPU memory. SecJev-4B improves on Kev-4B and security-trained generation in prompt-injection and traffic transfer tests. We release adapters, decision heads, SecJev-Corpus, and training and inference code.

cs.CR↗

arXiv2026

Accurate Molecular Property Regression Using Simple Fusion of Complementary Classical and Graph Predictors

Accurate prediction of continuous molecular properties supports chemical discovery and design. Many recent approaches combine multiple molecular representations within increasingly complex neural architectures, aiming to achieve higher accuracy. However, classical machine-learning models remain competitive when trained on chemically meaningful molecular representations. We therefore constructed a chemistry-guided classical branch from Morgan fingerprints, RDKit molecular descriptors, and conformer-derived geometric features selected to provide complementary views of molecular structure. We combined four tree learners trained on these representations with a CoMPT graph predictor and integrated their out-of-fold predictions using Ridge regression. Across ESOL, FreeSolv, and Lipophilicity datasets, the fused model achieved lower mean root-mean-square error (RMSE) than Chemprop and competitive graph and multimodal methods, including CoMPT, MoleSG, KFLM2, and MCMPP. After retraining on the separately curated AqSolDB dataset, the same approach, without redesign or retuning, reduced mean absolute error (MAE) relative to the classical ensemble and achieved lower mean MAE than Chemprop. These results show that accurate and transferable molecular property regression does not necessarily require a complex joint multimodal architecture: classical learning remains competitive when chemical insight guides the choice of molecular representations, and a graph model can add complementary information through simple prediction-level fusion.

physics.chem-ph↗

arXiv2026

Simpson filtrations and closure relations of Hodge moduli spaces

In this paper, we study closure relations of Simpson strata in Hodge moduli spaces over a smooth complex projective curve of genus $g\geq2$ and prove in every rank that the Heinloth weight strictly increases whenever a specialization changes the fixed component. For stable full chains, we refine this numerical condition by comparing the degrees of the corresponding filtration steps and construct counterexamples to Simpson's nestedness conjecture for the Dolbeault and de Rham moduli spaces in every rank $n\geq4$. Using the maximum property of this weight, we also discuss Simpson filtrations from the viewpoint of $Θ$ stratifications.

math.AG↗

arXiv2026

Pistis Technical Report

We introduce the Pistis model family, comprising 27B- and 9B-parameter multimodal large language models built on Qwen3.6 and Qwen3.5, respectively, and developed through a general and scalable post-training framework. The framework first establishes a strong foundation through large-scale multimodal supervised fine-tuning (SFT). Building on this SFT foundation, we propose Interleaved Distillation and Reinforcement Learning (IDRL), a novel post-training paradigm that tightly integrates on-policy distillation and reinforcement learning within a single training loop. By alternating between the two objectives, rather than optimizing either in isolation or combining them in a static joint loss, IDRL enables more effective knowledge transfer, greater optimization stability, and more precise credit assignment for long-horizon agentic trajectories, leading to stronger performance while mitigating common capability trade-offs. At both model scales, the framework produces two specialized variants: Pistis-Thinking, designed to strengthen deep multimodal reasoning, and Pistis-Agentic, which additionally incorporates agentic trajectory data to support long-horizon planning, iterative reasoning, and tool use. Pistis-Agentic is particularly strong in multimodal search. Both scales outperform their corresponding base models. Beyond model-parameter optimization, we further introduce Pistis-Auto-Harnessing (PAH), a system-level method that automatically improves the agent's inference harness through iterative optimization. Experiments demonstrate that PAH enhances the model performance without updating the model parameters or increasing the interaction budget.

cs.AI↗

arXiv2026

Divergence Timing and Cumulative Disagreement under KV-Cache Eviction

KV-cache eviction perturbs the conditional token distributions governing autoregressive generation. We investigate how first-divergence timing and subsequent token mismatch determine cumulative disagreement. We derive an exact decomposition under a specified stepwise maximal coupling: the expected mismatch fraction equals a first-mismatch contribution plus post-divergence exposure multiplied by its mismatch rate. An explicit construction over unrestricted autoregressive kernel pairs realizes the sharp interval of risks compatible with a finite divergence-aligned observation window. Residual-branch conditional Monte Carlo provides unbiased joint estimates of occurrence, occupation, and window/tail contributions, with per-replicate variance dominance for total token loss. Complete trajectories from Meta-Llama-3.1-8B-Instruct and Qwen2.5-7B-Instruct show that SnapKV at 50% retention enters divergence later and less often than SnapKV-512 or recent-token retention with the same 50% prompt-cache budget, while post-divergence total variation (TV) remains high. In an exploratory analysis of 288 documents, post-divergence exposure accounts for 85-90% of four aggregate mismatch gaps. On 288 independent documents at 90% retention, prespecified comparisons show higher branch-aligned TV in the late than in the early window in both models.

cs.LG↗

arXiv2026

Prediction Dynamics in Depth-Recurrent Language Models

Depth-recurrent language models refine predictions through repeated latent updates. Why can intermediate answers agree with the endpoint while their scores continue to change? We derive a sharp margin characterization that decomposes the conservatism of a magnitude bound into common translation, direction relative to the winner, and the pairing of each competitor's update with its score gap. Across Huginn-3.5B and Ouro-1.4B, accounting for update direction and competitor pairing reduces the mean earliest qualifying depth by a further 22.5-34.4% of the total depth beyond translation removal under full answer-text scoring. This retrospective comparison uses completed trajectories. Substantial contributions also occur under label scoring. For shared predictive distributions, we separate common and contrast motion orthogonally and express the common component through candidate-set mass and within-set concentration. Common and contrast energies can attenuate at different rates, allowing a growing preference-change share to coexist with shrinking absolute updates. These findings explain finite-depth answer preservation through the geometry and composition of observed score changes.

cs.CL↗

arXiv2026

The MiniMax-M2 Series: Mini Activations Unleashing Max Real-World Intelligence

We introduce the MiniMax-M2 series, a family of Mixture-of-Experts language models built around the principle that mini activations can unleash maximum real-world intelligence. The flagship M2 contains 229.9B total parameters with only 9.8B activated per token. Designed end-to-end for agentic deployment, the M2 series rests on three components: (i) agent-driven data pipelines producing large-scale, verifiable trajectories across agentic coding and agentic cowork, each grounded in an executable workspace and an artifact-aligned reward; (ii) Forge, a scalable agent-native RL system that adapts to long-horizon agent trajectories, paired with windowed-FIFO scheduling, prefix-tree merging, inference optimization, and a clean training-inference-agent decoupling that supports both white-box and black-box agents; (iii) the latest M2.7 checkpoint takes an early step toward self-evolution -- autonomously debugging training runs and modifying its own scaffold. Across M2 through M2.7, this combination translates a mini-activation footprint into frontier-tier performance on agentic coding, deep search, office-task, and reasoning benchmarks.

cs.AI↗

arXiv2026

ARD-REFSM: Enhancing Reflection Symmetry Detection with Asymmetric Denoising and Rotation Equivariance

Reflection symmetry detection remains challenging due to interference from asymmetric regions and arbitrary orientations of symmetric patterns. Asymmetric regions introduce background clutter that disrupts symmetric pattern matching, whereas conventional convolutional neural networks lack rotation equivariance, leading to inconsistent feature representations under rotational transformations. To address these issues, we propose an Asymmetric Region Denoising (ARD) module and a Rotation Equivariant Feature Similarity Matching (REFSM) module. The ARD module suppresses asymmetric interference to refine symmetric patterns, while the REFSM module enhances rotation equivariance through feature similarity matching between original and rotated images. Specifically, our dual-input REFSM framework leverages rotation loss to maximize consistency between the score maps of original and rotated images, thereby enabling precise prediction of rotation-equivariant symmetry axes. Furthermore, we introduce GMSYM, a new benchmark dataset that categorizes images into diverse scenarios and incorporates various interferences to address the limitations of existing reflection symmetry detection benchmarks. Extensive experiments on four standard datasets (DENDI, NYU, LDRS, SDRW) and our proposed GMSYM dataset demonstrate that our method achieves state-of-the-art performance in both accuracy and robustness.

cs.CV↗

arXiv2026

ABot-World-0: Infinite Interactive World Rollout on a Single Desktop GPU

We present ABot-World-0, an action-conditioned video world model for real-time, long-horizon closed-loop interaction, supported by a multi-source data infrastructure spanning AAA games, simulation engines, and internet videos to learn controllable world dynamics. WorldExplorer performs agent-driven collection guided by training feedback, while a unified pipeline applies 14 deterministic quality checks, VLM-based assessment, and synchronized action and text annotation. We progressively distill a bidirectional action-conditioned teacher into a causal student through teacher forcing and ODE distillation, and introduce LongForcing to align long student self-rollouts with an extended-horizon teacher, mitigating accumulated distribution shift and autoregressive drift. Raw keyboard actions provide a unified control interface for scene roaming and third-person character interaction, while reference-character memory provides persistent appearance cues for identity consistency during third-person rollouts. For deployment, we co-design a streaming inference stack with a lightweight VAE decoder, efficient attention, memory-aware scheduling, and low-bit DiT inference. Across optimized low-bit configurations, ABot-World-0 streams 720P video at up to 16 FPS on a single NVIDIA RTX 5090 desktop GPU, with 1.2s action-to-first-frame latency and approximately 19GiB peak VRAM. Experiments on WorldRoamBench and extended interactive rollouts demonstrate competitive controllability and coherent long-horizon world evolution.

cs.CV↗

arXiv2026

ABot-3DWorld 0: A Universal World Model to Explore Any 3D Space

We present ABot-3DWorld 0, a universal multimodal 3D world model that turns text, image, and video inputs into high-fidelity, explorable 3D worlds. At the heart of our framework is a unified Spatial Generative Primitive (SGP), a compact tuple of a high-quality panorama and a spatial point cloud that delivers an efficient description of any 3D space. Multimodal inputs are first lifted into this primitive; a 3D-consistent panoramic video generator then explores the primitive along a planned trajectory; finally, our panoramic video reconstruction engine converts the generated video into a clean, photorealistic 3D Gaussian Splatting (3DGS) world. This pipeline covers two regimes: rich inputs (multi-view sets, casual video) are lifted into the SGP through a geometry-rigorous recovery that mirrors the observed scene, while a single image or sentence is completed generatively into a creative world. The result is one low-barrier engine for general 3D content creation that further anchors generated worlds to geographic points of interest, enabling map-native spatial exploration at consumer scale. Experiments show that ABot-3DWorld 0 sets the state of the art among open-source methods and demonstrates stronger scene fidelity than Marble under rich multimodal inputs.

cs.CV↗

arXiv2026

Electron-beam Writing of Spectrally Uniform Green Single-photon Emitters in Hexagonal Boron Nitride

Scalable quantum photonic technologies require single-photon emitters whose positions and emission energies can be engineered simultaneously. Hexagonal boron nitride (hBN) is an attractive room-temperature host, but deterministic creation of spectrally reproducible emitters remains challenging. Here, we use a standard scanning electron microscope as a direct-writing tool to activate bright green single-photon emitters in hBN at predefined sites, without ion implantation or post-fabrication thermal annealing. The written emitters exhibit reproducible zero-phonon-line emission centered near 536 nm, room-temperature antibunching with g(2)(0) as low as 0.08, high brightness, strong linear polarization, and stable emission. Thickness-dependent activation, stacking experiments, cathodoluminescence spectroscopy, and first-principles calculations support a carbon-related defect complex as the most plausible origin of the emission. As a proof of nanophotonic compatibility, we further activate emitters in a nanoparticle-on-mirror plasmonic nanocavity and observe photoluminescence enhancement accompanied by shortened emission lifetimes. These results establish electron-beam direct writing as a practical route to site-selective, spectrally uniform green quantum emitters in hBN, offering a promising basis for integrated room-temperature quantum photonic architectures.

physics.optics↗

arXiv2026

Selective Expert Guidance for Effective and Diverse Exploration in Reinforcement Learning of LLMs

Reinforcement Learning with Verifiable Rewards (RLVR) has become a widely adopted technique for enhancing the reasoning ability of Large Language Models (LLMs). However, the effectiveness of RLVR strongly depends on the capability of base models. This issue arises because it requires the model to have sufficient capability to perform high-quality exploration, which involves both effectiveness and diversity. Unfortunately, existing methods address this issue by imitating expert trajectories, which improve effectiveness but neglect diversity. To address this, we argue that the expert only needs to provide guidance only at critical decision points rather than the entire reasoning path. Based on this insight, we propose MENTOR: Mixed-policy Expert Navigation for Token-level Optimization of Reasoning, a framework that provides expert guidance only at critical decision points to perform effective and diverse exploration in RLVR. Extensive experiments show that MENTOR enables models capture the essence of expert strategies rather than surface imitation, thereby performing high-quality exploration and achieving superior overall performance. Our code is available online.

cs.AI↗

arXiv2026

Socratic agents for autonomous scientific discovery in high-dimensional physical systems

The automation of scientific discovery has reached an inflection point. While AI systems now operate instruments, optimize parameters and generate hypotheses, most remain procedural: they execute workflows fixed by human designers. True autonomous science demands epistemic autonomy--the capacity to construct, challenge and revise physical explanations in response to evidence. Here we introduce AHOIS, a multi-agent AI scientist that embeds Socratic midwifery into closed-loop experimentation. A physics-critic agent interrogates hypotheses through causal questioning, constraint checking, counterexample generation and falsification-criteria formulation. We evaluate AHOIS on a real multimode-fibre optical platform, a high-dimensional system with complex wave transformations, indirect detection, environmental drift and multi-modal acquisition. Without prior encoding schemes, classifiers or speckle models, the system autonomously proposed and validated a random-interference encoding hypothesis, discovered task-adaptive sparse-measurement strategies, diagnosed distinct failure modes (encoding instability, fluorescence contamination and detector noise) and translated a published imaging protocol into an executable workflow on a non-original configuration. The discovered encoding yielded 16x16 measurements with effective rank 56.9 and classification accuracies of 76.97% on MNIST and 83.17% on Fashion-MNIST. Ablations show that Socratic interrogation improves physical consistency, hypothesis completeness, uncertainty calibration and experimental-plan validity. These results establish a route from workflow automation towards evidence-grounded, self-correcting autonomous discovery in complex physical environments.

cs.AI↗

arXiv2026

Non-asphericity of strata of genus-one differentials and stability spaces

We show that when the number of zeros or poles is at least four, every connected component of the strata of differentials in genus one with prescribed zero and pole orders is not an orbifold $K(π,1)$. For quadratic differentials, this provides infinitely many counterexamples to a conjecture attributed to Kontsevich, as well as to a folklore conjecture concerning the contractibility of spaces of Bridgeland stability conditions.

math.GT↗