Searcharxiv⌕ Search

arXiv subjects

Lei Li

Publications and source records attributed to Lei Li.

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

Inherited Learning in an Artificial Ecology: How Controls and Update Allocation Shape Benefits

Learning can improve an individual's behavior, yet a population risks losing that experience whenever individuals die and are replaced. Inheriting learned preferences offers a way to preserve useful behavior across generations, raising a question for artificial populations: when does inheritance improve collective performance, and how can its benefits be measured fairly? The challenge is that inheritance changes not only offspring behavior but also survival, reproduction, and opportunities for further hereditary updates. Random controls with equal update magnitudes may therefore yield misleading comparisons if they alter different states or obey different stability constraints. We investigate this problem in a resource-limited artificial ecology, combining structured random controls with interventions on newborn preferences and the allocation of hereditary updates. Preserving the state structure of random updates substantially narrows the apparent inheritance advantage, while a conditional establishment-speed benefit remains. Preference erasure and faster-learning compensation support a contribution from reduced offspring relearning. Update allocation also changes the comparison: event quotas and common time cutoffs can reverse rankings, although they also change realized update amounts. With update count and cumulative magnitudes matched, staged release improves occupancy but does not achieve the prespecified establishment criterion. These findings provide a framework for distinguishing the value of inherited preferences from the effects of control design and update allocation, clarifying how inherited learning should be evaluated in artificial populations.

cs.NE↗
arXiv2026

Video Understanding Reward Modeling: A Robust Benchmark and Performant Reward Models

Multimodal reward models have advanced substantially in text and image domains, yet progress in video understanding reward modeling remains severely limited by the lack of robust evaluation benchmarks and high-quality preference data. To address this, we propose a unified framework spanning benchmark design, data construction, and reward model training. We introduce Video Understanding Reward Bench (VURB), a benchmark featuring 2,100 preference pairs with long chain-of-thought reasoning traces (averaging 1,143 tokens) and majority voting evaluation across general, long, and reasoning-oriented video tasks. We further construct Video Understanding Preference Dataset (VUP-35K) via a fully automated pipeline, providing large-scale high-quality supervision for video reward training. Building on the data, we train VideoDRM and VideoGRM, a discriminative and a generative reward model, both achieving state-of-the-art performance on VURB and VideoRewardBench. Further analysis confirms that VUP-35K enhances both reward performance and model reasoning capability, while VideoDRM and VideoGRM yield significant gains under best-of-$N$ test-time scaling.

cs.CV↗
arXiv2026

LongEmo: Towards Emotion Understanding and Reasoning in Long Videos

While recent Multimodal Large Language Models (MLLMs) have shown promise in affective computing, their reasoning capabilities are largely confined to short video clips with limited interactions. However, real-world emotions are not merely isolated instantaneous reactions but dynamic and cumulative processes deeply shaped by past experiences and ongoing events. To bridge this gap, we introduce LongEmoBench, a benchmark dedicated to emotion understanding and reasoning in long videos. It assesses progressive capabilities scaling from continuous scene interactions to complex episodic developments. Furthermore, we propose LongEmo, a novel memory-augmented agentic framework designed to tackle the immense challenges of long-range affective reasoning. LongEmo processes continuous video streams to construct an Event Memory Graph, explicitly modeling long-range dependencies and capturing emotional dynamics across discrete events. Given a question, the agent retrieves a query-relevant event stream from the graph, iteratively integrating multimodal memories and relational dependencies to deduce the final answer. Extensive evaluations of 17 representative methods reveal that they struggle significantly with emotion understanding and reasoning in long videos. In contrast, LongEmo achieves state-of-the-art performance, demonstrating the efficacy of its event-centric memory architecture.

cs.CV↗
arXiv2026

SecureVibe: Making Vibe Coding More Secure

As vibe coding becomes increasingly capable and widespread, security vulnerabilities in even functionally correct solutions are a growing concern. When investigating functionally correct but insecure solutions, we find that the insecure agent is less than half as likely to conduct effective planning and testing for the hidden security risks behind the functional requirements. Motivated by this, we develop SECUREVIBE, a training recipe that explicitly targets planning and testing for code security. SECUREVIBE constructs training signals around these security behaviors. It includes supervised fine-tuning on the security suite with 4 security tasks, and post-training methods, SECUREVIBE_rl and SECUREVIBE_hg, to enhance security capabilities from verifiable execution feedback and hint-based self-supervision. Our SECUREVIBE outperforms the baseline on two types of security coding tasks across 4 benchmarks. Specifically, SECUREVIBE improves the security pass@1 by 6.9 points on BaxBench. The gains extend to unseen CWE categories, with improvements of 11.5 points on SusVibes. Meanwhile, it also improves functionality pass@1 by 13.6 points on the security coding task SusVibes and 4.1 points on the generic coding task SWE-bench Verified. Further analysis offers two practical insights: (i) diversifying supervision across security planning, coding, and testing strengthens security behaviors more effectively than adding coding trajectories alone, and (ii) hint-guided supervision is particularly valuable when the agent's existing security capabilities are insufficient to learn effectively from outcome feedback.

cs.CR↗
arXiv2026

Groupwise Agentic Grading and Advantage Redistribution for Code Agent RL

Reinforcement learning (RL) for code agents often uses executable tests to provide binary rewards. With these rewards, Group Relative Policy Optimization (GRPO) assigns identical advantages to test-passing trajectories within each rollout group, overlooking differences in implementation quality and adherence to task requirements. This leaves the policy without a learning signal that favors clean, targeted implementations over those containing unnecessary or out-of-scope changes. We introduce GAGAR, a framework for quality-aware credit redistribution in code agent RL. Built on dynamic sampling that retains groups containing both passing and failing trajectories, GAGAR places all trajectories from each group in a shared workspace, where an SFT-trained agentic grader jointly inspects them and ranks the test-passing candidates. Based on this ranking, we downweight lower-ranked trajectories and proportionally rescale the advantages of all test-passing trajectories to restore their original sum. This sum-preserving redistribution retains the relative weights established by quality-based downweighting while shifting credit toward higher-quality implementations. We evaluate GAGAR at industrial scale using pre-RL SFT checkpoints of MiMo-V2.6-Flash (310B total parameters) and MiMo-V2.6-Pro (1.02T total parameters). Controlled code-only Flash experiments show improved code agent performance, reduced trajectory-length growth, and more stable training. We further apply GAGAR in large-scale mixed-task RL with both Flash and Pro. Our results support combining test-based verification with groupwise agentic grading to improve the quality and stability of code agent RL.

cs.CL↗
arXiv2026

SkillRefine: Cross-Source Skill Induction and Execution Validation for LLM Agents in Refinery Planning Software

Operating industrial planning software such as AspenTech PIMS (Process Industry Modeling System) requires an LLM agent to combine structural knowledge, procedural knowledge from expert records, and constraints revealed only during execution. These evidence sources are heterogeneous and individually incomplete: documentation describes tables and interfaces but omits task-level coordination, expert CASE records expose multi-table modification patterns without explicit schema grounding, and execution feedback reveals latent constraints only when a plan is executed. We present \textsc{SkillRefine}, a framework that exploits the Documentation--Practice Gap between documented structure and expert practice to extract candidate coordination patterns, ground them against table definitions and COM specifications, and compile them into progressively disclosed skill packages with provenance. The library is then refined through label-free compliance screening, oracle-based match decomposition over table, row, column, and value dimensions, and signal-conditioned trajectory attribution for localized repair. We evaluate \textsc{SkillRefine} on PIMS-Bench, a benchmark built from two AspenTech PIMS demonstration models, using disjoint construction and held-out test tasks. On the held-out test set, \textsc{SkillRefine} achieves absolute component match F1 gains of 14\%--30\% across four LLM backbones, with the largest gains on complex multi-table coordination tasks.

cs.CE↗
arXiv2026

Anatomy-Aligned Surface Field Learning for Myocardial Reconstruction from Sparse Short-Axis Cine MRI

Patient-specific 4D myocardial reconstruction from cine MRI supports quantitative functional assessment, regional motion analysis, and simulation-based modeling. However, routinely acquired short-axis (SAX) cine MRI is sparsely sampled along the through-plane direction, making dense and anatomically consistent surface reconstruction challenging. In this study, we propose an anatomy-aligned surface learning framework that parameterizes the epicardial and endocardial surfaces on a shared circumferential-longitudinal UV domain. This formulation converts irregular 3D reconstruction into structured coordinate-field completion with explicit correspondence across subjects and cardiac phases. Sparse SAX contours are encoded as UV observation fields, coverage-aware sampling improves robustness to incomplete slice coverage, and topology- and distortion-aware learning preserves circumferential continuity and local surface quality. Experiments on three public cine MRI datasets showed that the proposed method consistently outperformed representative mesh-based and implicit reconstruction approaches, achieving overall Chamfer distances of $2.887$~mm on ACDC, $2.641$~mm on M\&Ms, and $2.810$~mm on M\&Ms-2. The reconstructed sequences also preserved ventricular function, with end-diastolic volume and ejection fraction errors of $3.3$~mL and $1.1 \%$, respectively. These results demonstrate that anatomy-aligned UV learning provides an accurate, efficient, and correspondence-aware representation for sparse cine MRI reconstruction and myocardial modeling. The source code will be available at https://github.com/yuan-xiaohan/SAX2MyoSurf.

cs.CV↗
arXiv2026

Deep-learning-aided dismantling of interdependent networks

Identifying the minimal set of nodes whose removal breaks a complex network apart, also referred as the network dismantling problem, is a highly non-trivial task with applications in multiple domains. Whereas network dismantling has been extensively studied over the past decade, research has primarily focused on the formulations of the optimization problem for single-layer networks, neglecting that many, if not all, real networks display multiple layers of interdependent interactions. In such networks, the optimization problem is fundamentally different as the effect of removing nodes propagates within and across layers in a way that can not be predicted using a single-layer perspective. Here, we propose a dismantling algorithm named MultiDismantler, which leverages multiplex network representation and deep reinforcement learning to optimally dismantle multi-layer interdependent networks. MultiDismantler is trained on small synthetic multiplex graphs; when applied to large, real and synthetic networks, it displays exceptional dismantling performance, clearly outperforming all existing methods that rely on a single-layer approach to network dismantling. We show that MultiDismantler is effective in guiding strategies for the containment of diseases in social networks characterized by multiple layers of social interactions. Also, we show that MultiDismantler is useful in the design of protocols aimed at delaying the onset of cascading failures in interdependent critical infrastructures.

cs.SI↗
arXiv2026

Is Vibe Coding Safe? Benchmarking Vulnerability of Agent-Generated Code in Real-World Tasks

Vibe coding is a new software development paradigm in which human engineers prompt a large language model (LLM) agent to complete complex coding tasks with little supervision. Although vibe coding is increasingly adopted, is the generated code really safe to deploy in production? To investigate this question, we propose SUSVIBES, a benchmark consisting of 186 feature-request software engineering tasks from real-world open-source projects, for which, human programmers committed vulnerable implementations. We evaluate 12 widely used coding agentic settings with frontier models on the benchmark. Disturbingly, all agents perform poorly in terms of software security. Although 57% of the solutions from SWE-Agent with Claude 4 Sonnet are functionally correct, only 11.8% are secure. Further experiments demonstrate that preliminary security strategies, such as augmenting the feature request with vulnerability hints, cannot mitigate these security issues. Our findings raise serious concerns about the widespread adoption of vibe coding, particularly in security-sensitive applications. The code and dataset are available at https://github.com/LeiLiLab/susvibes. The leaderboard is at https://leililab.github.io/susvibes-leaderboard.

cs.SE↗
arXiv2026

Logics of Filter Bubbles

Filter bubbles can be understood as configurations characterized by within-group proximity and separation from outsiders. Personalized feeds may form or preserve such configurations. We develop hybrid-style logics for reasoning about filter bubbles. The static logic characterizes bubble-shaped configurations, while its dynamic extension captures their formation and persistence under personalized feeds. We provide sound and strongly complete axiomatizations for the static and dynamic logics and prove their decidability.

cs.LO↗
arXiv2026

CodeMidas: Scaling Agentic Coding RL Environments from Code Itself

Training capable coding agents via reinforcement learning (RL) requires diverse tasks with reliable verifiers. Open-source codebases offer a rich source of such tasks, while existing methods typically rely on development artifacts such as issues and commits, limiting the range of tasks that can be extracted. To better scale RL environments, we present CodeMidas, an agentic pipeline that turns implemented functionality in existing codebases into executable RL environments using source code as its only task-specific input. CodeMidas allocates agentic compute to every stage of environment construction: agents explore implemented functionality to formulate behavioral specifications, construct tests grounded in execution of the original code, and validate and filter candidate tasks through execution checks and repeated solution rollouts. The resulting dataset has 5,545 training tasks from 3,185 open-source codebases spanning 23 programming languages and 15 technical domains. Training MiMo-V2.5 on these tasks with GRPO improves performance on all five diverse benchmarks, covering issue repair (DeepSWE + 11.7%), whole-program construction (ProgramBench +17%), and terminal work (Terminal-Bench v2.1 +8.5%). Ablations show that increasing the number of high-quality training tasks improves performance. Trajectory analysis shows the RL-trained agent demonstrates better behaviors like increasing codebase exploration and more diverse self-verification. These results establish source code as a scalable foundation for constructing RL environments that improve coding agents across diverse software tasks.

cs.AI↗
arXiv2026

Reverse Neighbor Sliding and Order Selection for Efficient Multi-Proximity Graph Merging

Approximate k Nearest Neighbor (AKNN) search in high-dimensional space is a foundational problem in vector databases with widespread applications. Among the numerous AKNN indexes, Proximity Graph-based indexes achieve state-of-the-art search efficiency across various benchmarks. In many real-world scenarios, datasets are maintained as multiple segment-level graph indexes to support continuous writes and segment management. However, these fragmented indexes complicate maintenance and degrade search efficiency, making fast graph index merging essential. In this paper, we focus on the efficient merging of multiple existing graph indexes into a single one. To achieve this, we propose a Reverse Neighbor Sliding Merge (RNSM) that exploits structural information to boost merging efficiency. We further propose Merge Order Selection (MOS) to minimize total merge cost across multiple indexes by eliminating redundant operations. Experiments show that our approach yields up to a 3.86x speedup over existing index merge methods and a 9.92x speedup over index reconstruction, while maintaining comparable search performance. Moreover, our method scales to merging up to 50 sub-indexes on datasets of 100 million vectors, maintaining consistent speedups.

cs.DB↗
arXiv2026

Parametric charge-conservative mixed finite element method for 3D incompressible inductionless MHD equations on curved domains

This paper develops a charge-conservative mixed finite element method with optimal convergence rates for the stationary incompressible inductionless MHD equations on three-dimensional curved domains. The discretization employs the isoparametric Taylor-Hood elements with grad-div stabilization for the velocity-pressure pair, and parametric Brezzi-Douglas-Marini elements for the current density. For sufficiently small meshsize, the discrete inf-sup conditions for both the velocity-pressure and current density-electric potential finite element pairs are established on curved meshes. Utilizing the Piola's transformation, the discrete current density is exactly divergence-free. By employing suitable extensions and projections, optimal a priori error estimates are derived in both the energy norm and the $L^2$-norm. Numerical experiments are presented to confirm the theoretical results.

math.NA↗
arXiv2026

Hilbert Expansion of the coupled radiation-Euler in the equilibrium regime

In this paper, we establish the validity of the Hilbert expansion for the coupled radiation-Euler model with non-relativistic source term for general initial data on the torus $\mathbb{T}^{3}$, which leads to the simplified equilibrium-diffusion limit and the initial layer corrections for the radiation intensity coupled with the temperature.

math.AP↗
arXiv2026

Deep operator learning for efficient sampling from invariant measures of stochastic differential equations

We introduce an amortized neural sampler that combines operator learning with flow methods for sampling. It maps SDE coefficient functions to pushforwards from a reference measure to the invariant measures, enabling efficient sampling across families of stochastic differential equations. Our framework shifts traditional sampling cost to an initial training phase, after which new SDE instances require only one encoder pass and a few ODE solver steps, independent of mixing time. To handle problems in high dimensions, we use Lagrangian trajectory sensors for the coefficient functions and cross attention in the architecture. We also theoretically establish the expressivity and resolution invariance of our framework. Experiments on 1D and 2D SDE families show competitive accuracy with substantial speedups over MCMC in regimes with slow mixing, transfer across sensor counts, and demonstration results on a 64D interacting particle SDE where traditional grid approaches are infeasible.

math.NA↗
arXiv2026

HiRAD: A Flexible Large-Scale AGV Routing System

Automatic Guided Vehicles (AGVs) substantially boost warehouse throughput, but routing large-scale AGV fleets remains challenging. Classical Multi-Agent Pathfinding solvers suffer from exploding combinatorial complexity and super-quadratic runtime, while relying on idealized grid or piecewise-linear motion models that mismatch real-world kinematics. Recent Reinforcement Learning (RL) solutions improve flexibility via decentralized agent policies but depend on discretized spatiotemporal representations, require millions of episodes to converge, and incur full-map observation at every step, which leads to large models, slow convergence, and high inference latency that violates real-time industrial control constraints. To address these bottlenecks, we propose HiRAD, a hierarchical RL framework for continuous-space AGV routing with real-time guarantees: (1) a step-level spatiotemporal representation that translates continuous motion into a differentiable RL problem, (2) a hierarchical strategy that splits heading choice from velocity control to reduce the action space, and (3) an asynchronous event-driven decision pipeline that lowers inference complexity from O(n^2) to O(n) and cuts per-step latency by as much as 71 percent. Across random graphs and two warehouse maps, HiRAD reduces makespan by 45 percent to 63 percent and shortens end-to-end runtime.

cs.RO↗
arXiv2026

Numerical index one characterises commutative JB*-triples

We establish that a JB$^*$-triple $E$ is commutative if and only if its numerical index $n(E)$ is one. As a consequence, $n(E) = 1$ implies $n(E^*) = n(E^{**}) = 1$. This settles a question left open in the literature since 2008.

math.OA↗
arXiv2026

Towards Multi-modal Multi-turn Safety: From Agentic Interaction to Strategic Alignment

Despite remarkable capability in multi-modal understanding, deploying Multi-modal Large Language Models (MLLMs) in open-ended conversational scenarios introduces safety risks that remain poorly addressed by existing alignment methods. Unlike simple malicious visual question and answer (VQA) pairs , multi-turn interactions enable adversaries to incrementally reconstruct harmful intent across dialogues, progressively bypassing safety constraints in ways that are difficult to detect at any individual turn. Meanwhile, conventional reinforcement learning from human feedback (RLHF) approaches are unsuitable for this situation: designed primarily for VQA tasks, they neither capture cross-turn risk dynamics nor scale efficiently without costly manual preference annotation. To close this gap, we introduce \textbf{MINT-Safe}, an open-source visual multi-turn training dataset comprising 11,270 multi-image dialogues and 500 refusal VQA pairs, constructed via multi-agent interaction with text-to-image (T2I) tool-call augmentation. Building on MINT-Safe, we propose \textbf{TAD-Align}, a dialogue safety alignment framework centered on a turn-aware dual-objective reward function. Rather than treating all dialogue turns uniformly, TAD-Align leverages rollout-based safety score variance to dynamically identify turns where the model exhibits inconsistent safety behavior, and adaptively up-weights these turns during optimization. Experiments on Qwen2.5-VL-7B-Instruct and LLaVA-NeXT-7B demonstrate reductions of over 10\% in Attack Success Rate (ASR), alongside improvements of at least 8\% in harmlessness and 13\% in helpfulness on multi-modal multi-turn safety benchmarks, while preserving general model capabilities.

cs.CL↗