Searcharxiv⌕ Search

arXiv subjects

Jiajie Xu

Publications and source records attributed to Jiajie Xu.

At least 19 recordsLinked to original sources
arXiv2026

MEDR: Query-Independent Frame Selection via Multi-Signal Event Modeling and Dynamic Rescoring

Frame selection is a fundamental component of multimodal large language models, enabling long videos to be processed under limited visual-token and computational budgets. Uniform sampling preserves temporal coverage but may miss informative content that appears only briefly. To alleviate this limitation, query-dependent methods can retrieve question-relevant frames. However, because the selected frames depend on the current question, the same visual input cannot be directly shared across different questions, and frame selection must be repeated in multi-turn video dialogue. This motivates us to seek a query-independent frame selection method that preserves the reusability of a fixed visual input while improving the coverage of informative events beyond uniform sampling. We propose Multi-Signal Event Modeling and Dynamic Rescoring (MEDR), a training-free and query-independent frame selection method. Multi-Signal Event Modeling organizes complementary visual, motion, and text signals into signal-specific temporal events. Dynamic Rescoring then iteratively reevaluates each candidate relative to the current selected set, updating its score according to frame-level signal strength, additional event coverage, and temporal proximity. The resulting fixed frame set is constructed without observing the query and can be reused across different questions. On the standard benchmark evaluations, MEDR improves model accuracy by 0.63%-0.89% on Video-MME. On the long-video subset of LongVideoBench, it improves accuracy by up to 1.23% with Qwen3-VL-8B. MEDR further improves overall accuracy by 0.53%, while reusing exactly the same frame set for every question about a video.

cs.CV↗
arXiv2026

RippleKV: Cross-Layer KV Cache Allocation via Perturbation Propagation

Long-context LLM inference is bottlenecked by KV cache memory, yet distributing a limited cache budget across layers remains challenging. Existing methods rely on proxies such as layer depth, attention statistics, or representation change. These proxies do not measure how perturbations at each layer propagate to the output and may therefore cause sensitive layers to be underallocated while tolerant layers are overallocated. To address this issue, we propose RippleKV, which allocates cache across layers by estimating how perturbations to each layer's value cache affect the final predictive distribution. RippleKV independently injects norm-adaptive perturbations into each layer's value cache and measures the induced KL divergence at the model output over a small calibration set. Averaging these responses yields a sensitivity profile specific to the model that need not vary monotonically with depth. RippleKV then converts the sensitivity profile into layer budget multipliers by normalizing the sensitivity scores and applying an exponential mapping. A ratio parameter controls the allocation disparity between sensitive and tolerant layers, while a final normalization preserves the KV cache budget. Experiments on LongBench demonstrate that RippleKV achieves the highest average performance among the evaluated KV cache compression methods under matched cache budgets.

cs.LG↗
arXiv2026

PluginEval: A Diagnostic Benchmark for Fine-Grained Error Attribution in Function Calling

Reliable evaluation of tool routing is critical as Large Language Models increasingly operate as autonomous agents. Current benchmarks face three structural limitations: data distributions that follow a power law leave rare scenarios underrepresented; the absence of adversarial hard negatives obscures performance differences across models; and annotation pipelines depend on LLM judgments that have not been validated through execution. In this paper, we introduce PluginEval, a benchmark constructed through a two-stage framework that systematically mitigates these limitations. First, we formulate tool routing as a sequence of three decisions and separate generation from verification. LLMs propose candidate calls, while deterministic validation and real API execution provide reliable quality signals. Second, we decompose each plugin by capability, intent, and boundary to identify trigger and exclusion scenarios. We then generate queries at different difficulty levels to fill coverage gaps, including adversarial negatives targeting three failure modes, and return them to the first stage for annotation. This process creates a closed loop that iterates until coverage converges. For evaluation, we move beyond aggregate accuracy. An LLM judge anchored to gold annotations classifies failures as missed calls, spurious calls, or parameter errors, producing a detailed error profile for each model. We evaluate five model families, including proprietary models and models with open weights, analyze their performance across difficulty levels and error categories, and validate the judge through agreement with human annotations.

cs.AI↗
arXiv2026

From Enumeration to Covering: Near-Optimal Densest P-Partite Subgraph Search over Large Heterogeneous Information Networks

Given a heterogeneous information network (HIN) and a query meta-path P of length i, the densest P-partite subgraph problem finds the subgraph, spanning the i typed layers of P, that maximizes a parameter-free density: the number of meta-path instances over the geometric mean of the layer sizes. It has applications across bibliographic, e-commerce, and biomedical networks. The state-of-the-art approximation linearizes the geometric-mean objective by fixing per-layer weights, but solves one subproblem for every feasible weight set, of which there are $O((n/i)^i)$, and on each achieves only a $1/i$ approximation. We show that neither the exhaustive enumeration nor the loose guarantee is necessary. First, we replace enumeration by covering: polylogarithmically many representative weight sets, localized further by a data-dependent bound, cover all feasible ones while losing only a tunable factor $1+η$ in density. Second, we cast each fixed-weight subproblem as a weighted supermodular densest-subgraph instance and solve it near-optimally, lifting the overall guarantee to $(1-δ)/(1+η)$. To our knowledge, this is the first near-optimal density approximation beyond the bipartite ($i=2$) case, and it yields a PTAS for every fixed i. Algorithmically, our solver is an adaptive peeling scheme that never materializes the meta-path instances, whose number can exceed the graph size by orders of magnitude. An incumbent-driven reduction further discards representative weight sets before their subproblems are solved. Experiments on five real HINs show that our algorithms achieve substantial speedups over enumeration-based baselines and can further certify the near-optimality of the returned subgraph.

cs.DS↗
arXiv2026

Integrated Discovery and State-Aware Servicing for Mobile AUVs With UOWC: Modeling and Performance Analysis

Underwater wireless optical communication (UWOC) is an enabling technology for high-throughput subsea networks, yet its long-term deployment is constrained by the finite energy budget of underwater nodes. To address this challenge, we investigate a mobile system wherein an autonomous underwater vehicle (AUV) performs joint wireless information transfer (WIT) and wireless power transfer (WPT) for a network of randomly distributed sensor nodes. This paper develops \textcolor{blue}{an integrated mission-level framework} that combines stochastic node discovery with state-aware servicing. First, we present an analytical model for node discovery based on a signal-to-noise ratio (SNR) analysis, deriving performance metrics that include the probability distribution of the discovery distance. Second, we introduce \textcolor{blue}{a threshold-based scheduling framework}, termed State-Aware Optimal Point Servicing (SA-OPS), which \textcolor{blue}{selects one of three actions according to the node's real-time energy state: preemptive charging, communication followed by charging, or communication only.} Simulations and multi-criteria decision analysis show that, \textcolor{blue}{under the considered assumptions and parameter ranges}, SA-OPS can improve the tradeoff between AUV energy expenditure and network-wide energy health relative to the adopted baseline strategies. The results also indicate that the selected charging threshold can be approximated by \textcolor{blue}{a simple state-dependent heuristic}, providing a practical guideline for autonomous energy replenishment in underwater networks.

eess.SY↗
arXiv2026

Task Decomposition-Guided Reranking for Adaptive Agent Skill Retrieval

Skill usage can significantly enhance the ability of modern agent systems to complete complex tasks. However, the growing scale of skill libraries makes accurate skill selection increasingly challenging. In real-world scenarios, ambiguous semantic matching often arises between a specific task requirement and multiple generic yet semantically similar candidate skills. Moreover, existing methods tend to overlook the dynamic influence of task difficulty and skill applicability when selecting the optimal target skill set. To address these issues, we propose SkillReranker, an inference-time reranking framework for adaptive skill selection. Specifically, we first perform semantic decomposition on both the task and skill sides, yielding informative subtask and execution-state descriptions as well as transition-state descriptions that characterize each skill's functionality. These descriptions are then used to construct a directed acyclic execution graph, where intermediate task states are modeled as nodes and candidate skills as edges, thereby establishing a structured task-skill correspondence. On this basis, SkillReranker determines whether each state node satisfies the split condition to identify subtask intervals. For each task interval, we employ a cross-encoder to perform comprehensive scoring over candidate skills and select the most suitable ones to form the final target skill set. Experiments on ALFWorld and ScienceWorld with three backbone LLMs show that SkillReranker effectively improves task performance, reduces environment interaction steps, and lowers token consumption compared with existing skill selection baselines.

cs.AI↗
arXiv2026

SSR3D-LLM: Structured Spatial Reasoning via Latent Steps for Fine-Grained Grounding in Unified 3D-LLMs

3D object grounding localizes referred objects in a 3D scene from natural language. Unified instance-centric 3D-LLMs aim to solve grounding together with dialog, QA, and captioning, yet many rely on a single pointer-style grounding decision that compresses a relational instruction into one selection. This is brittle for fine-grained queries where multiple same-class candidates must be ruled out by context objects and spatial relations. We propose Structured Spatial Reasoning 3D-LLM (SSR3D-LLM), a structured grounding interface for unified 3D-LLMs. Given fixed Mask3D object proposals, the LLM writes a sequence of latent spatial reasoning steps and memory tokens from the query, and a geometry-aware scorer reads these latent steps in order to refine candidate rankings step by step with step-length masking. The latent steps are learned from standard benchmark target supervision with auxiliary referential-cue supervision during training, while inference uses only the input query and Mask3D proposals. Across ReferIt3D, ScanRefer, and Multi3DRef, SSR3D-LLM achieves the strongest results among unified 3D-LLM baselines, with substantial gains over the single-pointer QPG baseline on fine-grained grounding and consistent improvements over prior unified 3D-LLMs, while preserving the default language-task route.

cs.CV↗
arXiv2026

R$^3$L: Reflect-then-Retry Reinforcement Learning with Language-Guided Exploration, Pivotal Credit, and Positive Amplification

Reinforcement learning drives recent advances in LLM reasoning and agentic capabilities, yet current approaches struggle with both exploration and exploitation. Exploration suffers from low success rates on difficult tasks and high costs of repeated rollouts from scratch. Exploitation suffers from coarse credit assignment and training instability: Trajectory-level rewards penalize valid prefixes for later errors, and failure-dominated groups overwhelm the few positive signals, leaving optimization without constructive direction. To this end, we propose R$^3$L, Reflect-then-Retry Reinforcement Learning with Language-Guided Exploration, Pivotal Credit, and Positive Amplification. To synthesize high-quality trajectories, R$^3$L shifts from stochastic sampling to active synthesis via reflect-then-retry, leveraging language feedback to diagnose errors, transform failed attempts into successful ones, and reduce rollout costs by restarting from identified failure points. With errors diagnosed and localized, Pivotal Credit Assignment updates only the diverging suffix where contrastive signals exist, excluding the shared prefix from gradient update. Since failures dominate on difficult tasks and reflect-then-retry produces off-policy data, risking training instability, Positive Amplification upweights successful trajectories to ensure positive signals guide the optimization process. Experiments on agentic and reasoning tasks demonstrate 5\% to 52\% relative improvements over baselines while maintaining training stability. Our code is released at https://github.com/shiweijiezero/R3L.

cs.LG↗
arXiv2026

SpecBlock: Block-Iterative Speculative Decoding with Dynamic Tree Drafting

Speculative decoding accelerates LLM inference by drafting a tree of candidate continuations and verifying it in one target forward. Existing drafters fall into two camps with opposite weaknesses. Autoregressive drafters such as EAGLE-3 preserve dependence along each draft path but call the drafter once per tree depth, making drafting a non-trivial share of per-iteration latency. Parallel drafters cut drafter calls by predicting multiple future positions in one forward, but each position is predicted without seeing the others, producing paths the verifier rejects. In this paper, we propose SpecBlock, a block-iterative drafter that combines path dependence with cheap drafting. Each drafter forward produces K dependent positions and we call this a block. The draft tree grows through repeated block expansions. Two mechanisms explicitly carry path dependence to keep later draft positions accurate. Within each block, a layer-wise shift carries the previous position's hidden state into every decoder layer. Across blocks, each new block can start from any position of the previous block, inheriting its hidden state to extend the path. To spend verifier budget where acceptance is likely, a co-trained rank head replaces the fixed top-k tree by allocating per-position branching during drafting. To avoid training the drafter on prefixes it never produces at inference, a valid-prefix mask drops the loss at later positions once an earlier one is wrong. Beyond static drafting, a cost-aware bandit at deployment uses free verifier feedback to update the drafter selectively, only when the expected throughput gain exceeds the update cost. Experiments show that SpecBlock improves mean speedup by 8-13% over EAGLE-3 at 44-52% of its drafting cost, and cost-aware adaptation extends this lead to 11-19%.

cs.CL↗
arXiv2026

Wi-Fi HaLow (IEEE 802.11ah) for Long-Range Monitoring Links: Point-to-Point NLoS/LoS and LoS Mesh Field Characterization

Monitoring deployments often require reliable long-range wireless links to intermittently upload sensor logs and short video snapshots. Wi-Fi HaLow (IEEE~802.11ah) is a promising candidate due to sub-1 GHz propagation and bandwidth-flexible PHY modes. This summary paper reports a field characterization organized around three deployment-driven regimes: (i) point-to-point Non-Line-of-Sight (NLoS) links; (ii) point-to-point Line-of-Sight (LoS) links over several-hundred-meter distances; and (iii) LoS mesh networking with fixed relay nodes for range extension. Using commodity HaLow dongle-class nodes in all regimes, we report application-layer goodput and monitoring-centric update latency based on transferring a representative ``heavy'' object (a $\sim$30 s video file). The measurements reveal (a) a clear bandwidth--range tradeoff and an NLoS coverage boundary around $\sim$120 m, (b) gradual throughput decay under LoS up to 814 m in single-hop with 0.15 Mbps at the farthest point, and (c) kilometer-class extension under LoS when fixed relays are introduced, reaching 901 m (two fixed relays) and 1110 m (three fixed relays

cs.NI↗
arXiv2026

Cognitive-Uncertainty Guided Knowledge Distillation for Accurate Classification of Student Misconceptions

Accurately identifying student misconceptions is crucial for personalized education but faces three challenges: (1) data scarcity with long-tail distribution, where authentic student reasoning is difficult to synthesize; (2) fuzzy boundaries between error categories with high annotation noise; (3) deployment parado-large models overlook unconventional approaches due to pretraining bias and cannot be deployed on edge, while small models overfit to noise. Unlike traditional methods that increase diversity through large-scale data synthesis, we propose a two-stage knowledge distillation framework that mines high-value samples from existing data. The first stage performs standard distillation to transfer task capabilities. The second stage introduces a dual-layer marginal selection mechanism based on cognitive uncertainty, identifying four types of critical samples based on teacher model uncertainty and confidence differences. For different data subsets, we design difficulty-adaptive mechanism to balance hard/soft label contributions, enabling student models to inherit inter-class relationships from teacher soft labels while distinguishing ambiguous error types. Experiments show that with augmented training on only 10.30% of filtered samples, we achieve MAP@3 of 0.9585 (+17.8%) on the MAP-Charting dataset, and using only a 4B parameter model, we attain 84.38% accuracy on cross-topic tests of middle school algebra misconception benchmarks, significantly outperforming sota LLM (67.73%) and standard fine-tuned 72B models (81.25%). Our code is available at https://github.com/RoschildRui/acl2026_map.

cs.LG↗
arXiv2026

LoopGuard: Breaking Self-Reinforcing Attention Loops via Dynamic KV Cache Intervention

Through systematic experiments on long-context generation, we observe a damaging failure mode in which decoding can collapse into persistent repetition loops. We find that this degeneration is driven by collapsed attention patterns, where a subset of heads locks onto a narrow suffix of the history, and is further stabilized by inference-time KV cache reuse. Crucially, since many existing KV cache policies rely on attention-based importance, this collapse can produce spuriously high scores for repetitive tokens, causing cache management to inadvertently amplify repetition. To study this phenomenon in a controlled and reproducible manner, we introduce LoopBench, a benchmark with explicit loop-inducing conditions and loop-oriented metrics that quantify repetition severity and generation instability beyond downstream task scores. Building on these insights, we propose LoopGuard, a lightweight, plug-in KV cache guard that detects loop onset online and disrupts the feedback cycle by pruning repetitive tail spans under a fixed cache budget. Experiments on LoopBench show that LoopGuard reduces loop incidence by over 90 percentage points, while restoring output diversity and reducing token waste.

cs.AI↗
arXiv2026

Head-wise Modality Specialization within MLLMs for Robust Fake News Detection under Missing Modality

Multimodal fake news detection (MFND) aims to verify news credibility by jointly exploiting textual and visual evidence. However, real-world news dissemination frequently suffers from missing modality due to deleted images, corrupted screenshots, and similar issues. Thus, robust detection in this scenario requires preserving strong verification ability for each modality, which is challenging in MFND due to insufficient learning of the low-contribution modality and scarce unimodal annotations. To address this issue, we propose Head-wise Modality Specialization within Multimodal Large Language Models (MLLMs) for robust MFND under missing modality. Specifically, we first systematically study attention heads in MLLMs and their relationship with performance under missing modality, showing that modality-critical heads serve as key carriers of unimodal verification ability through their modality specialization. Based on this observation, to better preserve verification ability for the low-contribution modality, we introduce a head-wise specialization mechanism that explicitly allocates these heads to different modalities and preserves their specialization through lower-bound attention constraints. Furthermore, to better exploit scarce unimodal annotations, we propose a Unimodal Knowledge Retention strategy that prevents these heads from drifting away from the unimodal knowledge learned from limited supervision. Experiments show that our method improves robustness under missing modality while preserving performance with full multimodal input.

cs.CV↗
arXiv2026

Offset Pointing for Energy-efficient Reception in Underwater Optical Wireless Communication: Modeling and Performance Analysis

Underwater Wireless Optical Communication is a key enabling technology for future space-air-ground-sea integrated networks. However, UOWC faces critical hurdles from spatial randomness and stringent energy constraints. These challenges fundamentally limit network lifetime and sustainability. This paper develops a comprehensive stochastic geometry framework to perform a differential energy analysis of UOWC links.Instead of relying on simplified models, we employ a three-dimensional truncated Poisson point process to accurately capture the anisotropic nature of the underwater environment, specifically the disparity between horizontal spread and vertical depth. It incorporates a Lambertian emission pattern, random receiver positions and orientations, and a realistic channel model with extinction effects. Under this model, we derive a full suite of closed-form expressions for key performance indicators. These include the nearest-neighbor distance distribution, expected received power, SNR, and BER. A principal and counter-intuitive finding of our analysis is an offset-pointing strategy. This strategy involves intentionally misaligning the receiver by a deterministically optimal angle. This approach maximizes the integrated received power across the aperture, contrary to the conventional pursuit of perfect alignment. We formulate and solve an energy-efficiency optimization problem. Our results demonstrate that this strategy enhances system robustness and yields substantial performance gains. Simulation results validate our analytical models. They show that the optimal offset strategy can reduce the required transmit power by nearly 20\% to achieve a target BER. This reduction directly translates into extended network lifetime and higher total data throughput. These findings offer a new design paradigm for deploying robust, cost-effective, and sustainable UOWC networks.

cs.IT↗
arXiv2026

NLoS Localization with Single Base Station Based on Radio Map

Accurate outdoor localization in Non-Line-of-Sight (NLoS) environments remains a critical challenge for wireless communication and sensing systems. Existing methods, including positioning based on the Global Navigation Satellite System (GNSS) and triple Base Stations (BSs) techniques, cannot provide reliable performance under NLoS conditions, particularly in dense urban areas with strong multipath effects. To address this limitation, we propose a single BS localization framework that integrates sequential signal measurements with prior radio information embedded in the Radio Map (RM). Using temporal measurement features and matching them with radio maps, the proposed method effectively mitigates the adverse impact of multipath propagation and reduces the dependence on LoS paths. Simulation experiments further evaluate the impact of different radio map construction strategies and the varying lengths of the measurement sequence on localization accuracy. Results demonstrate that the proposed scheme achieves sub-meter positioning accuracy in typical NLoS environments, highlighting its potential as a practical and robust solution for single-base-station deployment.

eess.SY↗
arXiv2026

Low-Complexity RSS-based Underwater Localization with Unknown Transmit Power

Underwater wireless sensor networks (UWSNs) have received significant attention due to their various applications, with underwater target localization playing a vital role in enhancing network performance. Given the challenges and high costs associated with UWSN deployments, Received Signal Strength (RSS)-based localization offers a viable solution due to its minimal hardware requirements and cost-effectiveness. In this paper, we assign distance-based weights to RSS measurements, providing higher reliability to closer anchor nodes. Using the weighted RSS measurements and generalized trust region subproblem (GTRS), we propose the GTRS-based localization technique with Unknown Transmit Power (GUTP), which can be solved by a simple bisection method. Unlike conventional localization methods that require prior knowledge of the target node's transmit power, GUTP jointly estimates both the location and transmit power of the target node, broadening its practical use. Additionally, we derive the Cramer-Rao lower bounds (CRLBs) for RSS-based underwater localization with known and unknown transmit power, respectively. Extensive simulations demonstrate that GUTP achieves enhanced accuracy and significantly lower computational complexity in estimating the target node's location and transmit power compared to existing semidefinite programming (SDP)-based techniques.

eess.SP↗
arXiv2026

ACR: Adaptive Context Refactoring via Context Refactoring Operators for Multi-Turn Dialogue

Large Language Models (LLMs) have shown remarkable performance in multi-turn dialogue. However, in multi-turn dialogue, models still struggle to stay aligned with what has been established earlier, follow dependencies across many turns, and avoid drifting into incorrect facts as the interaction grows longer. Existing approaches primarily focus on extending the context window, introducing external memory, or applying context compression, yet these methods still face limitations such as \textbf{contextual inertia} and \textbf{state drift}. To address these challenges, we propose the \textbf{A}daptive \textbf{C}ontext \textbf{R}efactoring \textbf{(ACR)} Framework, which dynamically monitors and reshapes the interaction history to mitigate contextual inertia and state drift actively. ACR is built on a library of context refactoring operators and a teacher-guided self-evolving training paradigm that learns when to intervene and how to refactor, thereby decoupling context management from the reasoning process. Extensive experiments on multi-turn dialogue demonstrate that our method significantly outperforms existing baselines while reducing token consumption.

cs.CL↗
arXiv2025

Using LLMs for Late Multimodal Sensor Fusion for Activity Recognition

Sensor data streams provide valuable information around activities and context for downstream applications, though integrating complementary information can be challenging. We show that large language models (LLMs) can be used for late fusion for activity classification from audio and motion time series data. We curated a subset of data for diverse activity recognition across contexts (e.g., household activities, sports) from the Ego4D dataset. Evaluated LLMs achieved 12-class zero- and one-shot classification F1-scores significantly above chance, with no task-specific training. Zero-shot classification via LLM-based fusion from modality-specific models can enable multimodal temporal applications where there is limited aligned training data for learning a shared embedding space. Additionally, LLM-based fusion can enable model deploying without requiring additional memory and computation for targeted application-specific multimodal models.

cs.LG↗