arXiv subjects

Pan Li

Publications and source records attributed to Pan Li.

At least 37 records · Page 2Linked to original sources

arXiv2026

Towards Direct Latent-Space Synthesis for Parallel Branches in LLM-Agent Workflows

Large language models increasingly serve as execution engines for agentic systems, yet they still consume context through a sequential text interface. This creates a mismatch with modern structured agent workflows, in which independent branches explore subtasks, retrieve evidence, or generate candidate solutions before a final synthesis step. Existing systems typically merge these branches by concatenating their textual outputs, which discards the parallel structure and incurs redundant prefill computation. In this work, we introduce Parallel-Synthesis, a plug-and-play framework that enables a synthesizer to directly consume the KV caches produced by parallel worker agents. Parallel-Synthesis combines a cache mapper that calibrates independently generated branch caches with a fine-tuned synthesizer adapter that enables generation from this non-sequential cache interface. We train Parallel-Synthesis using data that exposes the synthesizer to parallel cache contexts, teaches aggregation across cached branches, and distills reasoning behavior from standard text-concatenation-based synthesis. Across nine downstream datasets spanning math, science QA, code generation, GAIA, and multi-agent database diagnosis, Parallel-Synthesis matches or outperforms text-based synthesis on seven datasets and remains close on the other two. It also reduces time-to-first-token by 2.5x-11x, suggesting that direct cache-based synthesis is a promising interface for more native and efficient synthesis over parallel agent branches.

cs.AI↗

arXiv2026

Towards Stable, Globally Expressive Graph Representations with Laplacian Eigenvectors

A popular way to improve the expressive power of graph neural networks (GNNs) is to use Laplacian eigenvectors as additional node features, since they can serve both as structural identifiers and global coordinates of nodes. Properly handling the orthogonal group symmetry among eigenvectors is crucial for the stability and generalizability of Laplacian eigenvector augmented GNNs. Previous studies have shown that using a naive $O(p)$-group invariant encoder for each $p$-dimensional eigenspace often leads to expressivity loss and numerical instability. In this paper, we propose a novel method exploiting Laplacian eigenvectors to generate \emph{stable} and globally \emph{expressive} graph representations. The main difference from previous works is that (i) our method utilizes \textbf{learnable} $O(p)$-invariant representations for each Laplacian eigenspace of dimension $p$, which are built upon powerful orthogonal group equivariant neural network layers already well studied in the literature, and that (ii) our method deals with numerically close eigenvalues in a \textbf{smooth} fashion, ensuring its better robustness against perturbations. Experiments on various graph learning benchmarks witness the competitive performance of our method, especially its great potential to learn global properties of graphs.

cs.LG↗

arXiv2026

On Information Self-Locking in Reinforcement Learning for Active Reasoning of LLM agents

Reinforcement learning (RL) has become a de facto paradigm for building LLM-based agents that act, interact, and reason over extended task horizons. However, in active reasoning where agents must elicit new observations through interaction with the environment to solve the task, we find that outcome-based RL can induce a systematic failure mode which we call information self-locking (SeL): agents fail both to elicit informative feedback and to internalize obtained evidence. To understand the issue, we trace agentic behaviors into two coupled capabilities: Action Selection (AS), which determines observation streams, and Belief Tracking (BT), which updates the agent's internal task understanding. Theoretical and empirical analyses reveal a bidirectional bottleneck that leads to SeL: weak BT obscures the credit of informative actions, while weak AS deprives BT of useful evidence. This coupling weakens the learning signal for both capabilities and leads to SeL. To mitigate this issue, we propose AREW, a simple yet effective Advantage Reweighting method that uses easy-to-obtain directional critiques to reallocate credit within trajectories. Extensive experiments across 9 agentic tasks of varying complexity show that AREW significantly mitigates SeL, yielding up to 60-point gains in final performance. Code is available at https://github.com/unimpor/T3.

cs.AI↗

arXiv2026

Beyond Test-Time Memory: State-Space Optimal Control for LLM Reasoning

Associative memory has long underpinned the design of sequential models. Beyond recall, humans reason by projecting future states and selecting goal-directed actions, a capability that modern language models increasingly require but do not natively encode. While prior work uses reinforcement learning or test-time training, planning remains external to the model architecture. We formulate reasoning as optimal control and introduce the Test-Time Control (TTC) layer, which performs finite-horizon LQR planning over latent states at inference time, represents a value function within neural architectures, and leverages it as the nested objective to enable planning before prediction. To ensure scalability, we derive a hardware-efficient LQR solver based on a symplectic formulation and implement it as a fused CUDA kernel, enabling parallel execution with minimal overhead. Integrated as an adapter into pretrained LLMs, TTC layers improve mathematical reasoning performance by up to +27.8% on MATH-500 and 2-3x Pass@8 improvements on AMC and AIME, demonstrating that embedding optimal control as an architectural component provides an effective and scalable mechanism for reasoning beyond test-time training.

cs.LG↗

arXiv2026

When Do Graph Foundation Models Transfer? A Data-Centric Theory

Graph foundation models (GFMs) aim to reuse a single backbone across diverse graph domains, yet their transfer is often uneven and can exhibit negative transfer. While most prior work improves transfer through architectural or adaptation choices, we ask a data-centric question: which properties of two graph domains determine how much a fixed representation model changes its outputs? Using a graphon-based continuous limit for dense graphs, we show that for both set-based and message-passing tokenizations, any Lipschitz backbone admits an explicit decomposition of cross-domain output shift into (i) graph-specific finite-sample approximation terms and (ii) an intrinsic, relabeling-invariant domain discrepancy capturing structural mismatch. A key ingredient is positional-encoding (PE) stability: we establish stability guarantees for spectral PEs and highlight contrasting behaviors of eigenvector- versus subspace-based PEs. Experiments on synthetic and real graphs validate the theory and translate the decomposition into guidance for data curation in GFM transfer.

cs.LG↗

arXiv2026

LongDPM: Overlap-Aware 4D Reconstruction from Long Monocular Videos

Recovering a dynamic 3D scene from a long monocular video is crucial for dense geometry, camera motion, and temporal correspondence to remain consistent in a shared coordinate system. Existing methods face two key challenges: (1) feed-forward reconstruction models provide accurate local predictions but are limited to short clips, and (2) long-range trackers preserve correspondences without producing dense sequence-level reconstruction. This paper presents LongDPM, a novel overlap-aware framework for scalable long-range monocular dynamic reconstruction. First, LongDPM processes long videos in overlapping chunks, keeping inference memory bounded by the chunk length. Second, it connects chunk-local coordinate systems through confidence-weighted registration with static-aware overlap abstraction. Third, it associates dynamic identities across chunk boundaries and fuses matched trajectories to recover coherent long-range 3D motion. Experimental results demonstrate that LongDPM achieves superior long-range reconstruction and tracking performance, reducing dense tracking EPE over V-DPM on PointOdyssey, Kubric-F, and Kubric-G, while obtaining the best TUM-dynamics ATE for camera pose estimation.

cs.CV↗

arXiv2026

Privacy Amplification in Differentially Private Zeroth-Order Optimization with Hidden States

Zeroth-order optimization has emerged as a promising approach for fine-tuning large language models under differential privacy (DP) and memory constraints. While privacy amplification by iteration (PABI) provides convergent DP bounds for first-order methods, establishing similar guarantees for zeroth-order methods remains an open problem. First-order PABI analysis relies on the fact that gradients are perturbed with isotropic noise, allowing privacy bounds to be iteratively tracked via shifted Rényi divergence. In contrast, DP zeroth-order methods inject scalar noise along random update directions to maintain utility. This anisotropic update fails standard shifted divergence frameworks, as the global Lipschitz property no longer holds almost surely. We provide the first convergent hidden-state DP bound for zeroth-order optimization by proposing a hybrid noise mechanism and a novel coupling analysis. We bypass the purely shifted-divergence approach by constructing a coupled auxiliary process, which circumvents the global Lipschitz barrier and yields a convergent privacy bound. Furthermore, our results induce better DP zeroth-order algorithmic designs that are previously unknown to the literature.

cs.LG↗

arXiv2026

Accelerated Sequential Flow Matching: A Bayesian Filtering Perspective

Sequential probabilistic inference from streaming observations requires modeling distributions over future trajectories as new observations arrive. Although diffusion and flow-matching models are effective at capturing high-dimensional, multimodal distributions, their deployment in real-time streaming settings typically relies on repeatedly sampling from a non-informative initial distribution. This results in substantial inference latency, particularly when multiple samples are needed to characterize the predictive distribution. In this work, we introduce Sequential Bayesian Flow Matching, a framework inspired by Bayesian filtering. By learning a probability flow that transports the posterior distribution from one time step to the next time step conditioned on new observations, it mirrors the recursive structure of Bayesian belief updates. Crucially, by using the previous belief as an informative source distribution, it enables substantially faster sampling than naive resampling from scratch. Across scientific forecasting tasks spanning accelerator beam spill dynamics, fluid dynamics, and weather forecasting, as well as decision-making benchmarks, our method achieves performance competitive with full-step diffusion on distributional metrics while using far fewer sampling steps, substantially reducing inference latency. Our code is available at https://github.com/Graph-COM/Sequential_Flow_Matching.

cs.LG↗

arXiv2026

Structural Alignment Improves Graph Test-Time Adaptation

Graph-based learning excels at capturing interaction patterns in diverse domains like recommendation, fraud detection, and particle physics. However, its performance often degrades under distribution shifts, especially those altering network connectivity. Current methods to address these shifts typically require retraining with the source dataset, which is often infeasible due to computational or privacy limitations. We introduce Test-Time Structural Alignment (TSA), a novel algorithm for Graph Test-Time Adaptation (GTTA) that adapts a pretrained model to align graph structures during inference without the cost of retraining. Grounded in a theoretical understanding of graph data distribution shifts, TSA employs three synergistic strategies: uncertainty-aware neighborhood weighting to accommodate neighbor label distribution shifts, adaptive balancing of self-node and aggregated neighborhood representations based on their signal-to-noise ratio, and decision boundary refinement to correct residual label and feature shifts. Extensive experiments on synthetic and real-world datasets demonstrate TSA's consistent outperformance of both non-graph TTA methods and state-of-the-art GTTA baselines.

cs.LG↗

arXiv2026

How Far Are VLMs from Privacy Awareness in the Physical World? An Empirical Study

As Vision-Language Models (VLMs) are increasingly deployed as autonomous cognitive cores for embodied assistants, evaluating their privacy awareness in physical environments becomes critical. Unlike digital chatbots, these agents operate in intimate spaces, such as homes and hospitals, where they possess the physical agency to observe and manipulate privacy-sensitive information and artifacts. However, current benchmarks remain limited to unimodal, text-based representations that cannot capture the demands of real-world settings. To bridge this gap, we present ImmersedPrivacy, an interactive audio-visual evaluation framework that simulates realistic physical environments using a Unity-based simulator. ImmersedPrivacy evaluates physically grounded privacy awareness across three progressive tiers that test a model's ability to identify sensitive items in cluttered scenes, adapt to shifting social contexts, and resolve conflicts between explicit commands and inferred privacy constraints. Our evaluation of 12 state-of-the-art models reveals consistent deficits. In cluttered scenes, all models exhibit monotonic performance decay as scene complexity grows due to perceptual deficit. When social context shifts, no model exceed 65% selection accuracy. Under conflicting commands, the best model gemini-3.1-pro perfectly balances task completion and privacy preservation in only 51% of cases. These findings reveal that current VLMs in the physical world suffer from perceptual fragility and fail to let their knowledge of privacy cues govern their situated behavior. Our code and data is available at https://github.com/immersed-privacy/immersed-privacy .

cs.CR↗

arXiv2026

Can Explanations Improve Recommendations? Evidence from Prediction-Informed Explanations

Recommender systems are central to digital platforms, yet they face a fundamental trade-off between accuracy and explainability. Black-box models achieve strong performance but lack interpretability needed for trust and adoption. Existing explainable AI approaches either treat explanations as post-hoc or at the cost of accuracy. We challenge this view, proposing that explanations, when designed as an integral component of a system and aligned with prediction outcomes, can improve both interpretability and performance. We introduce RecPIE (Recommendation with Prediction-Informed Explanations), a framework that jointly optimizes recommendation predictions and natural-language explanations generated by LLMs. RecPIE embeds explanation generation into the learning loop: predictions guide explanation generation (prediction-informed explanations), which are fed back to refine subsequent predictions (explanation-informed predictions) via alternating training. The LLM is fine-tuned using LoRA and reinforcement learning with a customized reward derived from recommendation accuracy. Drawing on multi-environment statistical learning theory, we formally ground why explanation generation and prediction can be mutually reinforcing. We evaluate RecPIE on large-scale point-of-interest recommendation data from Google Maps, where user preferences span diverse place categories. RecPIE improves predictive accuracy by 3-4% over state-of-the-art baselines and matches the best performing model using only 12% of the training data. In human evaluations with 566 participants, RecPIE explanations are preferred 61.5% of the time (versus 16.6% for the best baseline) and rated closer to human-generated explanations. These results reframe explainability not as a constraint on performance but as a design lever for improving AI systems, with implications for trust, data efficiency, and marketplace deployment.

cs.IR↗

arXiv2026

Simultaneous Secrecy and Covert Communications (SSACC) in Mobility-Aware RIS-Aided Networks

In this paper, we propose a simultaneous secrecy and covert communications (SSACC) scheme in a reconfigurable intelligent surface (RIS)-aided network with a cooperative jammer. The scheme enhances communication security by maximizing the secrecy capacity and the detection error probability (DEP). Under a worst-case scenario for covert communications, we consider that the eavesdropper can optimally adjust the detection threshold to minimize the DEP. Accordingly, we derive closedform expressions for both average minimum DEP (AMDEP) and average secrecy capacity (ASC). To balance AMDEP and ASC, we propose a new performance metric and design an algorithm based on generative diffusion models (GDM) and deep reinforcement learning (DRL). The algorithm maximizes data rates under user mobility while ensuring high AMDEP and ASC by optimizing power allocation. Simulation results demonstrate that the proposed algorithm achieves faster convergence and superior performance compared to conventional deep deterministic policy gradient (DDPG) methods, thereby validating its effectiveness in balancing security and capacity performance.

eess.SP↗

arXiv2026

Joint Secrecy and Covert Communication (JSACC): An Enhanced Physical Layer Security Approach

In this paper, we propose an enhanced physical layer security approach, named joint secrecy and covert communication (JSACC), which aims to improve the performance of physical layer security (PLS). The JSACC system can dynamically switch between secrecy mode and covert mode according to the channel difference between legitimate and illegitimate receivers. We further leverage reconfigurable intelligent surface (RIS) to extend the communication range. For each scenario, we derive the closed-form expressions for the outage probability (OP) and ergodic rate (ER). To further understand system performance, we derive asymptotic approximations in the high signal-to-noise ratio (SNR) regime to obtain the diversity order and high-SNR slope. We demonstrate that the diversity order of the JSACC depends on Nakagami fading parameters and the RIS reflecting element number. Simulation results are consistent with our theoretical analysis and reveal the superiority of the JSACC system over the conventional secrecy communication (SC) system.

eess.SP↗

arXiv2026

CREWS: Collaborative Robust Edge WiFi Sensing with Asynchronous and Incomplete Observations

Existing collaborative WiFi sensing systems rely on perfect node synchronization and complete data availability. However, real-world edge deployments suffer from heterogeneous computing and network dropouts, leading to asynchronous and incomplete features. We propose CREWS, a robust collaborative sensing framework that inherently resists these network volatility. First, CREWS employs a topology-agnostic aggregator invariant to the arrival order and subset size of incoming features. Second, rather than discarding delayed observations, it utilizes a staleness-aware adaptive replay mechanism. By treating stale features from lagging nodes as system-induced hard samples, CREWS transforms synchronization delays into beneficial training regularization. We theoretically prove the joint convergence of this architecture and demonstrate how replay bounds the bias-variance trade-off. Extensive evaluations and an 8-node heterogeneous hardware testbed demonstrate its superior resilience. Under severe conditions i.e., 50\% transient dropout rate or out-of-distribution jitter, CREWS restricts accuracy degradation to merely 2.2 percentage points, substantially outperforming state-of-the-art baselines.

cs.NI↗

arXiv2026

Time-reversed Flow Matching with Worst Transport in High-dimensional Latent Space for Image Anomaly Detection

Likelihood-based deep generative models have been widely investigated for Image Anomaly Detection (IAD), particularly Normalizing Flows, yet their strict architectural invertibility needs often constrain scalability, particularly in large-scale data regimes. Although time-parameterized Flow Matching (FM) serves as a scalable alternative, it remains computationally challenging in IAD due to the prohibitive costs of Jacobian-trace estimation. This paper proposes time-reversed Flow Matching (rFM), which shifts the objective from exact likelihood computation to evaluating target-domain regularity through density proxy estimation. We uncover two fundamental theoretical bottlenecks in this paradigm: first, the reversed vector field exhibits a non-Lipschitz singularity at the initial temporal boundary, precipitating explosive estimation errors. Second, the concentration of measure in high-dimensional Gaussian manifolds induces structured irregularities, giving rise to a Centripetal Potential Field (CPF) that steers trajectories away from Optimal Transport (OT) paths. We identify these observations as the inherent dualities between FM and rFM. To address these issues, we introduce local Worst Transport Flow matching (WT-Flow), which amplifies the observed CPF of rFM to mitigate the initial singularity while circumventing the need for exact distribution transformations via density proxy. Experiments on five datasets demonstrate that WT-Flow achieves state-of-the-art performance among single-scale flow-based methods, and competitive performance against leading multi-scale approaches. Furthermore, the proposed framework enables superior one-step inference, achieving a per-image flow latency of only 6.7 ms. Our code is available on https://github.com/lil-wayne-0319/fmad.

cs.CV↗

arXiv2026

Long-Horizon Plan Execution in Large Tool Spaces through Entropy-Guided Branching

Large Language Models (LLMs) have significantly advanced tool-augmented agents, enabling autonomous reasoning via API interactions. However, executing multi-step tasks within massive tool libraries remains challenging due to two critical bottlenecks: (1) the absence of rigorous, plan-level evaluation frameworks and (2) the computational demand of exploring vast decision spaces stemming from large toolsets and long-horizon planning. To bridge these gaps, we first introduce SLATE (Synthetic Large-scale API Toolkit for E-commerce), a large-scale context-aware benchmark designed for the automated assessment of tool-integrated agents. Unlike static metrics, SLATE accommodates diverse yet functionally valid execution trajectories, revealing that current agents struggle with self-correction and search efficiency. Motivated by these findings, we next propose Entropy-Guided Branching (EGB), an uncertainty-aware search algorithm that dynamically expands decision branches where predictive entropy is high. EGB optimizes the exploration-exploitation trade-off, significantly enhancing both task success rates and computational efficiency. Extensive experiments on SLATE demonstrate that our dual contribution provides a robust foundation for developing reliable and scalable LLM agents in tool-rich environments.

cs.AI↗

arXiv2026

Open-Set Supervised 3D Anomaly Detection: An Industrial Dataset and a Generalisable Framework for Unknown Defects

Although self-supervised 3D anomaly detection assumes that acquiring high-precision point clouds is computationally expensive, in real manufacturing scenarios it is often feasible to collect a limited number of anomalous samples. Therefore, we study open-set supervised 3D anomaly detection, where the model is trained with only normal samples and a small number of known anomalous samples, aiming to identify unknown anomalies at test time. We present Open-Industry, a high-quality industrial dataset containing 15 categories, each with five real anomaly types collected from production lines. We first adapt general open-set anomaly detection methods to accommodate 3D point cloud inputs better. Building upon this, we propose Open3D-AD, a point-cloud-oriented approach that leverages normal samples, simulated anomalies, and partially observed real anomalies to model the probability density distributions of normal and anomalous data. Then, we introduce a simple Correspondence Distributions Subsampling to reduce the overlap between normal and non-normal distributions, enabling stronger dual distributions modeling. Based on these contributions, we establish a comprehensive benchmark and evaluate the proposed method extensively on Open-Industry as well as established datasets including Real3D-AD and Anomaly-ShapeNet. Benchmark results and ablation studies demonstrate the effectiveness of Open3D-AD and further reveal the potential of open-set supervised 3D anomaly detection.

cs.CV↗

arXiv2026

RoboMatch: A Unified Mobile-Manipulation Teleoperation Platform with Auto-Matching Network Architecture for Long-Horizon Tasks

This paper presents RoboMatch, a novel unified teleoperation platform for mobile manipulation with an auto-matching network architecture, designed to tackle long-horizon tasks in dynamic environments. Our system enhances teleoperation performance, data collection efficiency, task accuracy, and operational stability. The core of RoboMatch is a cockpit-style control interface that enables synchronous operation of the mobile base and dual arms, significantly improving control precision and data collection. Moreover, we introduce the Proprioceptive-Visual Enhanced Diffusion Policy (PVE-DP), which leverages Discrete Wavelet Transform (DWT) for multi-scale visual feature extraction and integrates high-precision IMUs at the end-effector to enrich proprioceptive feedback, substantially boosting fine manipulation performance. Furthermore, we propose an Auto-Matching Network (AMN) architecture that decomposes long-horizon tasks into logical sequences and dynamically assigns lightweight pre-trained models for distributed inference. Experimental results demonstrate that our approach improves data collection efficiency by over 20%, increases task success rates by 20-30% with PVE-DP, and enhances long-horizon inference performance by approximately 40% with AMN, offering a robust solution for complex manipulation tasks. Project website: https://robomatch.github.io

cs.RO↗