arXiv Science⌕ Search

arXiv subjects

Xin Li

Publications and source records attributed to Xin Li.

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

Moment-Video: Diagnosing Temporal Fidelity of Video MLLMs on Momentary Visual Events

Video multimodal large language models (MLLMs) have made rapid progress in general and long-form video understanding, yet their ability to capture and use brief, answer-critical visual evidence remains underexplored. Momentary actions or state transitions may determine an answer while occupying only a small fraction of the video. Performance on broad video understanding tasks alone does not establish whether models can reliably identify and use this temporally localized evidence. We introduce Moment-Video, a benchmark for evaluating temporal fidelity through momentary visual event understanding. Questions are constructed around visually observable events or event sequences, requiring models to notice, count, describe, or reason about transient evidence. The benchmark contains 1,000 human-verified video-QA pairs across seven domains and 25 subcategories, covering four tasks: Temporal Occurrence, Temporal Counting, Action Description, and Temporal Reasoning. We evaluate 34 proprietary and open-source MLLMs. Under the main evaluation settings, the best-performing model, Seed-2.0-Pro, achieves 39.6% overall accuracy, while most open-source models remain below 25%. Diagnostic analyses show that denser frame sampling improves some models, but substantial errors persist even when sampled frames cover all annotated necessary event intervals in an audited subset. Performance also declines on longer videos, suggesting additional temporal-localization challenges. By exposing these limitations, Moment-Video provides a diagnostic benchmark to guide the evaluation and development of video MLLMs that more reliably understand brief, answer-critical visual events.

cs.CV↗
arXiv2026

Commit Locally, Exit Globally: Coordinating Adaptive Sampling and Early Exit in Diffusion Language Models

Diffusion language models predict tokens in parallel but still require many iterative forward passes. Adaptive sampling accelerates local token commitment, while recent global stopping methods seek to end generation early. Our trajectory analysis shows that short answers can stabilize early in decoding, whereas reasoning outputs often require more steps to settle. A fixed progress threshold therefore struggles to avoid both redundant computation on short-answer tasks and premature termination on reasoning tasks. We introduce C$^4$, a training-free framework that combines local decoding acceleration with adaptive global early exit. Confidence-Verified Early Exit (CVEE) determines when to end generation, skipping the remaining steps once its stopping criterion is met. It combines confidence and temporal agreement to adapt stopping to the observed revision history. Commit-Core-Then-Confirm (CCTC) accelerates token commitment while generation continues, reducing the decoding steps needed to complete each block. It extends confidence-threshold parallel decoding with a boundary-anchored core and one-step confirmation of deferred predictions, constraining additional commitments to mitigate premature freezing under unresolved context. With CVEE calibrated once on LLaDA, the same numerical settings transfer across all 12 zero-shot tasks and to Dream without recalibration. Under this configuration, C$^4$ reduces decoding steps by 64-95% and achieves measured decoding-loop speedups of 2.6x-18.6x, with task performance close to full decoding. Component analyses show complementary contributions: early exit supplies most savings on short-answer tasks, while local acceleration contributes more on reasoning and code generation.

cs.CL↗
arXiv2026

Structured Multi-Level-Set Inversion for 3-D Electromagnetic Inverse Scattering

Many dielectric targets encountered in 3-D electromagnetic inverse scattering exhibit piecewise-homogeneous material distributions, which provide useful structural information for inversion. To exploit this property, this paper proposes a structured multi-level-set inversion (SMLSI) method within a physics-driven neural network (PDNN) framework. Instead of directly parameterizing the contrast distribution, SMLSI represents it in terms of object support and material distribution. Multiple level-sets are mapped to soft-support functions and combined through a soft-union operation to construct the object support. The normalized contributions of these soft-support functions, together with trainable complex relative permittivities, determine the material distribution within the support. The level-set residuals are parameterized by a coordinate-based neural network, and the trainable parameters are optimized from scattered field data. Structure-aware regularization terms are introduced to promote binary object support and spatially coherent material regions while reducing material mixing and redundant activation of level-set supports. Adaptive regularization weighting is further employed to reduce the dependence on manual parameter tuning. Numerical experiments demonstrate that SMLSI improves reconstruction accuracy over direct contrast parameterization and enables the recovery of irregular 3-D geometries, closely spaced objects, and multiple material regions. Compared with CSI and $L_{2/3}$-FBE-WCIE, SMLSI achieves more accurate reconstructions on representative synthetic targets. Experimental validation using 3-D Fresnel data further demonstrates its applicability to measured scattering data.

cs.LG↗
arXiv2026

MemCode: Discrete Semantic Representations for Long-Term Agent Memory

Long-term memory remains a fundamental bottleneck for large language model (LLM) agents across sessions. Existing systems typically adopt a write-then-retrieve paradigm, storing context-dependent dialogue fragments and retrieving them via nearest-neighbor search in a continuous embedding space. This design yields unstable access units and systematically misses logically related but lexically distant evidence. We propose MemCode, a long-term agent memory framework that couples retrieval-oriented writing with discrete associative retrieval. MemCode rewrites dialogue into self-contained atomic memory units (AMUs), maps AMUs into a learnable codebook with isomorphic semantic quantization (ISQ), and uses codebook co-occurrence topology as a retrieval path complementary to vector search. For long-term memory stores, MemCode separates online atomic writes from asynchronous offline consolidation. On LongMemEval-S and LoCoMo with GPT-4o-mini and Qwen3-30B-A3B-Instruct-2507 backbones, MemCode achieves the best overall accuracy while keeping per-dialogue memory construction within roughly 40~130 seconds, making it 3~21 times faster than LightMem and over an order of magnitude faster than A-MEM and MemoryOS.

cs.AI↗
arXiv2026

Spectral World Renderer: Unified 3D Hyperspectral Unmixing

Cameras and the foundation models built on them recognize what a 3D scene contains, but not what it is made of: surfaces of the same colour can be different materials. Hyperspectral unmixing explains each spectrum as a mixture of material signatures, yet in 3D it is applied to rendered images after the views are composited, or tied to slow implicit fields, and no multi-view data with known composition exist to check the result. We present the Spectral World Renderer (SWR), which unifies world generation and unmixing. A coding agent builds executable 3D worlds that are rendered with measured spectra, so every hyperspectral image comes with exact labels and sub-pixel proportions; the resulting SWR-Bench provides eight indoor and outdoor worlds in which decompositions can be verified. SWR then unmixes inside rendering: each Gaussian of a compact 3D representation is decoded into material abundances before compositing, so one render yields spectra and material maps, and any surface can be queried by material name, a clicked pixel or a reference spectrum. Without material labels, SWR identifies materials best in all eight worlds, has the best mean PSNR on real captures, trains 3.7-12.9x faster than unmixing baselines, renders 34.9 labelled views/s and separates same-coloured materials.

cs.CV↗
arXiv2026

Cracking Mobile Manipulation: From Failure Diagnosis to Agentic Robot System

Recent vision-language-action (VLA) models have shown strong performance on local robot manipulation, yet their success does not readily translate to long-horizon mobile manipulation. In our experiments, a pretrained subtask VLA achieves 80% success rate on atomic manipulation tasks but only 19% when these skills are composed into mobile manipulation. We argue that this gap stems from failures across the execution process rather than from local manipulation capability alone. To uncover these bottlenecks, we systematically diagnose failures in long-horizon mobile manipulation and identify three dominant regimes: task planning, navigation and base positioning, and local manipulation. Based on this diagnosis, we develop failure-guided agentic intervention that augments a fixed subtask VLA with demonstration-based planning guidance, adaptive visual memory, accessibility-aware base adjustment, and local end-effector adjustment. On RoboCasa365, these interventions progressively improve task success from 19% to 47%, with the finalized system reaching 49% on held-out episode seeds, compared with 18% for the baseline. The same design improves performance from 14% to 37% on BEHAVIOR-1K and from 5/20 to 13/20 on real-robot tasks. These results suggest that reliable mobile manipulation can be advanced by diagnosing and intervening on the current bottleneck, rather than uniformly scaling the underlying policy. Our code will be released at https://github.com/PKU-VaLuE-Lab/Cracking-Mobile-Manipulation.git.

cs.RO↗
arXiv2026

Share the Symbol, Not the Vector: Training-Free Fact Communication Across Language Models

Multi-agent systems allow language models to solve tasks collaboratively by sharing complementary information. These systems typically communicate through text, while recent approaches also explore exchanging internal model states. However, sharing these states across architectures often requires learned translators or assumptions about model compatibility. We introduce SYNC (Symbol-Indexed Neural Communication), a word-indexed interface that connects heterogeneous models without fine-tuning or a learned translator. From descriptions only it can see, the sender selects words from a shared list and sends their indices for each object, without seeing the questions. The receiver writes the described facts into its hidden states using its own word directions, vectors fitted once from its activations. We test selective use by swapping one attribute between two objects and requiring correct answers about both changed and unchanged attributes before and after the swap, four answers in total. We choose an early write layer on four development receivers and then fix it. With three different senders and nine newer receivers from five families, four-answer success on new scenes reaches 72.0-98.3%, up to 96 percentage points above middle-layer writing, and the two largest receivers match plain text. A task-free probe predicted in advance which depths each new receiver can use (45 of 45 cases). Without retuning, messages listing every fact keep 89.6-100% success under new layouts, and sender messages reach 23.9-64.0% on a landmark task with multi-token symbols, below text but from 0% without a message. These results show that shared word indices and receiver-local directions can support selective fact use across model families without additional training.

cs.CL↗
arXiv2026

BluePixel: Toward Native 4K Agentic Image Editing

Instruction-driven image editing has reached native 4K, yet how such editors are trained remains undisclosed, and pretraining one from scratch is beyond most teams. The task itself compounds the difficulty: a single end-to-end editor cannot make precise parametric adjustments, match specialized experts, or interpret the vague instructions users actually write. We argue that 4K editing is a system problem rather than a single-model problem, and present BluePixel, which couples a native-4K editing model with an agent harness; the complete training recipe is disclosed in this paper. The model makes 4K training practical: it turns a 1K-pretrained editor into a native-4K one through progressive VAE adaptation that keeps 4K training as affordable as 1K, absorbs a long tail of vertical editing skills through continual learning, and refines the aesthetic-fidelity trade-off with reinforcement learning. Distillation via source-anchored target projection (SATP) then compresses 50 denoising steps to 4, completing a 4K edit in 8.7 s on a single Kunlunxin P800 XPU while preserving task success rate and background consistency. The harness, BluePixel Agent, makes editing controllable: it decomposes user instructions into specialist and general operations, verifies every result with the same reward model that guided training, and retries within bounded attempts. On VINS-4KEval++ (3,305 native-4K cases), BluePixel ranks second among seven models, behind only GPT Image 2 in non-edited-region consistency.

cs.CV↗
arXiv2026

Reflect After Failure: Brain-Inspired Self-Evolving Agents for Embodied Spatial Intelligence

Vision-Language Models (VLMs) have made substantial progress in spatial reasoning, and self-evolving agents further extend these capabilities to embodied environments by learning from interactions. However, current self-evolution remains particularly challenging in embodied spatial tasks, complex and multi-step decisions from partial visual evidence making it difficult to determine why an execution fails. The failure may stem from misperception, insufficient exploration, or incorrect planning, yet execution feedback alone rarely reveals which component is responsible. As a result, simply accumulating interaction experience can reinforce ineffective strategies. To address this, we propose HiMeco, a brain-inspired self-evolving agent for embodied spatial intelligence, inspired by metacognitive monitoring and hippocampal memory consolidation. Through Hippocampal-Metacognitive Reflection, HiMeco diagnoses failures in perception-exploration-planning (PEP), then compares execution attempts to derive reusable strategies and guidance for avoiding recurring errors. It also abstracts interaction programs into parameterized skills, using evidence contracts and execution-based validation to determine their retention and reuse. These strategies and skills guide subsequent interactions, with memory updated only between rounds and model parameters kept frozen. HiMeco achieves state-of-the-art performance on ESI-Bench, and it also improves accuracy on EmbSpatial-Bench by 3.94 percentage points without evolution.

cs.AI↗
arXiv2026

Counterfactual Neighborhood Risk Fields: Evidence-Grounded Selective Safety for Vision-Language Models

Representation-based safety defenses for vision-language models (VLMs) often learn to separate harmful from benign inputs, but separation alone does not establish what a risk score means or when it should be trusted. We introduce Counterfactual Neighborhood Risk Fields (CNRF), an evidence-grounded framework for selective VLM safety. CNRF constructs context-matched benign-harmful counterfactual pairs, uses their representation differences as capability-related risk directions, and estimates neighborhood support separately from risk. This separation prevents unfamiliarity from being conflated with harmfulness and enables selective intervention rather than blanket rejection. A victim-adaptive interface accommodates different safety observations across VLM architectures and routes selected requests for same-victim response review, which either releases acceptable candidates or produces constrained alternatives with reduced harmful capability. Across three VLMs and five multimodal jailbreak families, CNRF achieves 86.06-89.30% macro defense success rate while retaining substantial benign-task utility. Controlled experiments further show that query-matched references and preserved context-direction correspondence materially improve risk ranking in the low-false-positive regime. CNRF thus reframes representation-based VLM safety from learning harmfulness boundaries to constructing applicable evidence for selective risk decisions.

cs.AI↗
arXiv2026

DMAD: Distribution Matching as Adversarial Distillation for Fast Visual Generation

Distribution Matching Distillation (DMD) trains a few-step student from the difference between separately estimated target and student scores, so it must keep an auxiliary diffusion model fitted to the student's evolving distribution at extra memory and computation cost. We introduce DMAD, Distribution Matching as Adversarial Distillation, which recasts distribution matching as classification and learns the required log-density ratios directly. Two discriminator heads on a shared backbone distinguish real data and teacher samples from the student's, and linear losses on their logits train the student without auxiliary score fitting. We prove that at the discriminator optimum these losses recover the distribution-matching gradient underlying DMD, through the classical identity linking discriminator logits to log-density ratios. We further introduce gap-based reweighting, which adapts teacher supervision across noise levels from the real-data head's empirical logit gap between real and teacher samples. DMAD reaches a Fréchet Inception Distance (FID) of 1.04 with one-step generation on ImageNet-64x64, 14.47 with four-step SDXL on COCO-10K, and a VBench total score of 85.15 with four-step Wan2.1-T2V-14B, the best values among the compared few-step methods and the multi-step teachers. On MiniMax-H3-33B, our four-step student achieves overall human preference rates of 79.1% over DMD2 and 84.6% over rCM for joint audio-video generation, excluding ties. Our code, models and demos are available at https://yzmblog.github.io/projects/DMAD.

cs.CV↗
arXiv2026

Many Right Answers, One Training Example: Removing Responder Noise from GRPO for LLM Recommenders

A recommendation request on Reddit draws several replies, each naming different titles. When an LLM recommender is trained on such threads with reinforcement learning (RL), which replies should a generated list be rewarded against? The state-of-the-art recipe, Rank-GRPO, picks one reply per training example, so the advantage mostly records which reply was drawn (85% of its variance) rather than how good the list is. We argue that RL should train on the request, not on each answer, and propose ReMaRk: it scores every list once against the union of all replies' titles and samples each request equally. This makes RL cheaper. On seven backbones and three datasets, ReMaRk reaches the validation recall of 2,000 per-reply RL steps in 25 to 53% of the steps and leads at that budget; trained until neither improves, it is never significantly behind Rank-GRPO and in four of seven pairs reaches Rank-GRPO's best with fewer GPU-hours. The same reward helps on ProtoQA, where about 100 people answer each question, which points to a simple rule for RL with several valid human answers: reward them jointly.

cs.CL↗
arXiv2026

AesCode: Aesthetic Code Generation with Decoupled Cross-Modal Rewards

Generating information-rich visual artifacts such as slides, posters, and dashboards as code is appealing because the outputs are structured, editable, and verifiable. Yet code models cannot see how their choices of layout and color come together on the canvas, often limiting spatial planning and aesthetic quality. Image generators excel at visual design but can misrender requested content, making their outputs unreliable targets for direct reproduction. We therefore introduce AesCode, which generates code grounded in the content prompt while using an image generated from the same prompt as an aesthetic guidance. To do so, AesCode represents the target as a design graph that describes the whole canvas and makes each property attributable. From this graph, we derive complementary cross-modal rewards: deterministic verifiers score properties parsable from code and rendering, while a VLM judge assesses the non-parsable ones with a sample-specific Visual Graph Rubric (VGR) tied to graph elements and relations. We optimize these decoupled rewards through reinforcement learning with per-channel normalization to jointly improve spatial planning, aesthetic quality, and verifiable correctness. To evaluate this setting, we build a benchmark that scores verifiable correctness and visual quality separately, with visual scores highly consistent with human judgments. We find that reference images substantially improve the visual quality of code generated by existing models, and ablations show that AesCode benefits from references both as model input and as a source of visual rewards. Relative to same-scale Qwen3-VL baselines given the same references, AesCode-8B and AesCode-32B improve visual quality by 31.1 and 22.4 percentage points, reaching scores comparable to those of closed-source frontier models such as GPT-5.5 and Claude Opus 4.8.

cs.AI↗
arXiv2026

ErrOPD: When Does On-Policy Distillation Transfer Teacher Errors?

A distilled student can copy its teacher's mistakes along with its skills. On-policy distillation (OPD), used in recent post-training recipes, has the teacher grade every token of the student's own outputs. Which teacher errors does it pass on? We build ErrOPD, a testbed whose teachers are fine-tuned to close a JSON list one item early, an error their students almost never make. The answer depends on the KL direction: across four model families, forward-KL OPD transmits the error within the training budget (+13-23 points), while reverse-KL OPD at our reference learning rate adds under 0.05 points. A duplicated ID and, where the decision arises, two SQL errors show the same split. The reverse-KL update scales with the student's own probability of a token, and an intervention tests this: giving reverse KL forward KL's gradient magnitude at the error decision alone makes it transmit on two model pairs. Reverse-KL OPD can delay such errors rather than block them: some runs adopt the error with longer training, and Qwen3 students that already make it, even rarely, adopt it within the budget. Because greedy evaluation on Qwen3 misses the inherited error, distilled students should also be audited on sampled outputs.

cs.LG↗
arXiv2026

Why Agentic Workflows Fail: A Taxonomy Derived from Literature and Empirical Data

Agentic workflows combine large language models with planning, memory, tools, execution environments, and iterative feedback. Their failures are difficult to diagnose because the same user-visible symptom may originate from model behavior, workflow orchestration, external tools, infrastructure, or their interaction. This paper presents a source-aware taxonomy of agentic workflow failures and applies it to evidence from academic literature and GitHub. The taxonomy combines source attribution with three empirical dimensions: a concrete failure-type dictionary, workflow-stage localization, and agent-capability localization. Its central structure is a two-axis localization matrix: workflow stages form the horizontal process axis and agent capabilities form the vertical functional axis. This design moves beyond taxonomies that mainly classify final failure manifestations by localizing intermediate process states inside agent operation. Our empirical comparison shows that literature-derived failures emphasize planning, reasoning, learning, perception, and knowledge-related mechanisms, whereas GitHub-derived failures concentrate on execution-stage, actuation-related, tool, API, and infrastructure failures. The dominant hotspot is iterative execution and actuation. These findings indicate that agentic workflow reliability should be treated as a systems problem rather than only a model-output problem. The taxonomy provides a basis for structured evaluation, root cause analysis, failure injection, and engineering triage.

cs.SE↗
arXiv2026

FirstFit coloring process of random bipartite graphs and an online coloring lower bound in the random order model

FirstFit colors each arriving vertex with the smallest positive integer absent from its previously colored neighborhood. We prove that this rule can require $n^{Ω(1)}$ many colors on certain bipartite graph of $n$ vertices even when the vertices arrive in a uniformly random order. This refutes the conjecture of Bosek, Gutowski, Lasoń, and Przybyło [BGLP24] that FirstFit uses a polylogarithmic expected number of colors on every bipartite graph in the random order model. We obtain the lower bound by analyzing the FirstFit coloring process on random bipartite graphs, where vertices arrive to either side with equal probability, and with an independent edge density $n^{-x}$ for some $0 < x < 1$. Our result can also be viewed as a qualitative strengthening of Kučera [Kuč91], who showed that, for every $ε> 0$, there are graphs with chromatic number at most $n^ε$ on which FirstFit uses $Ω(n/\log n)$ colors in expectation under a uniformly random order. Our result shows that such poor behavior of FirstFit already occurs for $χ(G) = 2$.

cs.DS↗
arXiv2026

Stable Long-Range Charging and Boundary-Mediated Thresholds in a Waveguide Quantum Battery

The waveguide-integrated architecture holds promise for scalable on-chip integration of a charger, a transmission channel, and a quantum battery in a modular, independently optimizable design. We investigate a minimal charging configuration consisting of two emitters coupled to a semi-infinite waveguide and identify the optimal charging mechanism, wherein non-Markovian dynamics driven by out-of-band bound states give rise to single-frequency oscillations of the stored energy. We show that near-unity charging efficiency is achievable at weak to moderate coupling strengths via detuning control, provided that the battery-to-boundary separation $n$ is less than three times the charger-to-boundary separation $m$ ($n<3m$)---a constraint imposed by the mirror-induced feedback---while the average charging power decays exponentially with the spatial separation $n-m$. These findings offer useful insights for monolithic solid-state quantum energy storage and transport.

quant-ph↗
arXiv2026

Data-Adaptive Mahalanobis Metric Learning for Cross-Head Attention in Transformers

In this paper, we propose Data-Adaptive Mahalanobis Cross-Head Attention (DAMCHA), a simple and effective attention mechanism that generalizes Multi-Head Attention (MHA) through the lens of metric learning. Standard MHA implicitly learns a static, head-wise Euclidean similarity corresponding to the diagonal blocks of a unified cross-head metric matrix, discarding all off-diagonal interactions across heads. DAMCHA addresses this limitation via two key innovations. First, it learns the full cross-head metric matrix, inducing a Mahalanobis-like similarity that explicitly captures inter-head interactions beyond the restricted block-diagonal structure of MHA. Second, it parameterizes the metric matrix as an input-dependent function, enabling the attention geometry to dynamically adapt to the intrinsic manifold structure of the data rather than remaining fixed after training. We further introduce principled regularization strategies, most notably stack-wise parameter sharing, to ensure computational efficiency and stable optimization. DAMCHA serves as a drop-in replacement for MHA and its variants in various Transformer architectures, delivering superior expressive power and learning performance with comparable or reduced computational overhead. Extensive experiments demonstrate that DAMCHA-based Transformers consistently outperform competing baselines across a range of benchmark tasks. Code is available at https://github.com/erwinmsmith/DAMCHA.

cs.LG↗