arXiv subjects

Zhiyuan Lu

Publications and source records attributed to Zhiyuan Lu.

16 recordsLinked to original sources

arXiv2026

Route-MHT: Multimodal Transformer Guardrails for Thermal Visual Place Recognition

Strong mapped-region thermal visual place recognition (VPR) does not ensure safe rejection of unmapped queries. We identify and quantify this gap in AnyThermal: high Map-In retrieval accuracy coexists with confident false loop closures in Map-Out. We address it with ROUTE-MHT, a multimodal transformer guardrail. Causal motion forms a route-local candidate pool beyond the frontend's Top-$K$. Closed-form $\text{SE}(2)$ SVD verifies candidates, while frozen visual features, nine-dimensional SVD residuals, and motion proxies enter a masked multi-head transformer (MHT). Their interactions yield a contextual confidence correction to reject unsupported matches without altering geometric pose alignment. We collect an indoor thermal dataset with a physical mobile robot (Dataset-A), forming five same-day/cross-day map-query pairs; the protected interface reaches macro R@1@5m of .610/.861. Dataset-B comprises 20 map-hole scenarios derived from public STheReO-KAIST recordings. Across five scenario-held-out folds and three seeds, ROUTE-MHT reduces FPR from .116 for the SVD baseline to .061 (paired 95% CI [-.100, -.014]), while improving AUC from .945 to .967 and recall from .884 to .914. Public benchmark transfer checks on STheReO-KAIST, MS2, and IRSLAM-KRI extend the evaluation under their released metric or route-progress protocols.

cs.RO↗

arXiv2026

Multi2AV-Safety: Benchmarking Safety in Multimodal-to-Audio-Video Generation

Audio-video generation is rapidly moving from prompt-driven synthesis toward multimodal conditioning, where text, images, audio, and video can jointly shape the generated output. This shift changes the nature of safety evaluation: harmful intent may no longer reside in any single input, but instead emerge from how otherwise benign or weakly harmful conditions interact across modalities and time. Existing safety benchmarks, however, remain largely prompt-centric or tied to fixed conditioning interfaces, leaving such compositional risks difficult to study systematically. To bridge this gap, we introduce Multi2AV-Safety, the first safety benchmark, to the best of our knowledge, to cover all 11 non-singleton T/I/A/V conditioning configurations for audio-video generation, comprising 11,024 attack instances. Evaluation on Multi2AV-Safety reveals systematic weaknesses in representative multimodal safety guards across attack mechanisms and harm-evidence structures. Our evaluation reveals two complementary failure modes: harmful semantics can emerge from the combination of individually benign inputs, while explicit harmful cues can become harder to detect when mixed with benign multimodal context. Together, these results identify \emph{compositional risk perception} as a central capability gap in safeguarding multimodal-conditioned audio-video generation: current safety guards fail to reliably integrate safety evidence across modalities and time, even when all conditioning inputs are observable. The dataset will be publicly released in October 2026.

cs.AI↗

arXiv2026

Data-Driven Generation of Compact Quasi-Isodynamic Stellarators

Stellarator design explores a vast space of three-dimensional plasma boundaries, only a small fraction of which yields usable equilibria. Data-driven models can narrow this search by learning from existing optimized configurations. Building on the ConStellaration database, we extend conditional boundary generation to four-field-period QI configurations, focusing on the sparsely sampled low-aspect-ratio regime. The approach learns the common geometric structure of known QI equilibria and then adapts it using a small high-fidelity compact dataset, allowing target magnetic properties to guide generation beyond the original data distribution. This adaptation reduces the compact-domain test loss by approximately 87% and yields converged ultra-compact candidates consistent with the prescribed conditions. Several candidates show favorable confinement indicators, and one provides a useful seed for further QI optimization and finite-beta assessment. The method therefore serves as a data-informed front end to high-fidelity physics and optimization.

physics.plasm-ph↗

arXiv2026

Direct Optimization of Stellarator Omnigenity from the Second Adiabatic Invariant

Stellarators offer a steady-state, disruption-free path to fusion energy but suffer from enhanced particle losses due to their three-dimensional geometry. Existing optimization methods rely on geometric proxies rather than the actual trapped-particle orbit condition. We present a differentiable framework that directly optimizes the fundamental orbit action governing particle confinement. The approach yields compact stellarator designs with excellent fast-ion confinement, finite-pressure stability, and coil compatibility, demonstrating that first-principles orbit optimization can be integrated with engineering constraints in a single computational workflow for fusion reactor design.

physics.plasm-ph↗

arXiv2026

Self-supervised Automatic Matting

High-quality alpha mattes are notoriously expensive to annotate, creating a fundamental data bottleneck for deep image matting. While prior work attempts to reduce annotation cost using coarser labels like trimaps or masks, they remain reliant on costly per-pixel supervision, limiting scalability and generalization. In this work, we push the boundary further and ask: can we train an automatic matting model using only RGB images, with no manual annotation at all? We answer this by presenting SSMatte, a self-supervised framework that for the first time achieves performance on par with fully-supervised automatic matting. Our key insight is to decompose the problem into semantic anchoring and detail matting. SSMatte first generates a semantic matting prompt from frozen self-supervised ViT features by propagating class-token seeds via a novel, training-efficient semantic anchoring loss based on a generalized Rayleigh quotient. This prompt then anchors a detail matting network, which is optimized via a fixed-point-based loss that enforces alpha-RGB consistency. Extensive experiments show SSMatte outperforms prior weakly-supervised methods, matches the performance of fully-supervised models on portrait benchmarks, and demonstrates favorable scaling and generalization behaviors with additional data. Our work pushes automatic matting to an fresh, fully annotation-free paradigm. Code will be available.

cs.CV↗

arXiv2026

Log-linear Model for Dual System Estimation and Computational Considerations

The use of dual system estimation (DSE) is heavily used in Census Bureau operations. With DSE methods, it is important to implement methods to infer the population size among those with missing data from one or both data sources. The use of log-linear models, calculated through EM algorithms, promises a way for estimation of counts among all groups with incomplete recorded data, as displayed by Van der Heijden et al. 2022. Unfortunately, the numerical computations involved scale very poorly the more the population is divided, to the point where simultaneous analysis of several demographic and geographic factors, such as state of residence and ethnicity, becomes computationally infeasible. Here, an alternative method to calculate the log-linear estimates will be provided, which can calculate the maximum likelihood estimator in orders of computation lower than through the EM algorithm.

stat.CO↗

arXiv2026

CorpusQA: A 10 Million Token Benchmark for Corpus-Level Analysis and Reasoning

While large language models now handle million-token contexts, their capacity for reasoning across entire document repositories remains largely untested. Existing benchmarks are inadequate, as they are mostly limited to single long texts or rely on a "sparse retrieval" assumption-that answers can be derived from a few relevant chunks. This assumption fails for true corpus-level analysis, where evidence is highly dispersed across hundreds of documents and answers require global integration, comparison, and statistical aggregation. To address this critical gap, we introduce CorpusQA, a new benchmark scaling up to 10 million tokens, generated via a novel data synthesis framework. By decoupling reasoning from textual representation, this framework creates complex, computation-intensive queries with programmatically guaranteed ground-truth answers, challenging systems to perform holistic reasoning over vast, unstructured text without relying on fallible human annotation. We further demonstrate the utility of our framework beyond evaluation, showing that fine-tuning on our synthesized data effectively enhances an LLM's general long-context reasoning capabilities. Extensive experiments reveal that even state-of-the-art long-context LLMs struggle as input length increases, and standard retrieval-augmented generation systems collapse entirely. Our findings indicate that memory-augmented agentic architectures offer a more robust alternative, suggesting a critical shift is needed from simply extending context windows to developing advanced architectures for global information synthesis.

cs.CL↗

arXiv2025

QwenLong-L1.5: Post-Training Recipe for Long-Context Reasoning and Memory Management

We introduce QwenLong-L1.5, a model that achieves superior long-context reasoning capabilities through systematic post-training innovations. The key technical breakthroughs of QwenLong-L1.5 are as follows: (1) Long-Context Data Synthesis Pipeline: We develop a systematic synthesis framework that generates challenging reasoning tasks requiring multi-hop grounding over globally distributed evidence. By deconstructing documents into atomic facts and their underlying relationships, and then programmatically composing verifiable reasoning questions, our approach creates high-quality training data at scale, moving substantially beyond simple retrieval tasks to enable genuine long-range reasoning capabilities. (2) Stabilized Reinforcement Learning for Long-Context Training: To overcome the critical instability in long-context RL, we introduce task-balanced sampling with task-specific advantage estimation to mitigate reward bias, and propose Adaptive Entropy-Controlled Policy Optimization (AEPO) that dynamically regulates exploration-exploitation trade-offs. (3) Memory-Augmented Architecture for Ultra-Long Contexts: Recognizing that even extended context windows cannot accommodate arbitrarily long sequences, we develop a memory management framework with multi-stage fusion RL training that seamlessly integrates single-pass reasoning with iterative memory-based processing for tasks exceeding 4M tokens. Based on Qwen3-30B-A3B-Thinking, QwenLong-L1.5 achieves performance comparable to GPT-5 and Gemini-2.5-Pro on long-context reasoning benchmarks, surpassing its baseline by 9.90 points on average. On ultra-long tasks (1M~4M tokens), QwenLong-L1.5's memory-agent framework yields a 9.48-point gain over the agent baseline. Additionally, the acquired long-context reasoning ability translates to enhanced performance in general domains like scientific reasoning, memory tool using, and extended dialogue.

cs.CL↗

arXiv2025

Addressing Information Loss and Interaction Collapse: A Dual Enhanced Attention Framework for Feature Interaction

The Transformer has proven to be a significant approach in feature interaction for CTR prediction, achieving considerable success in previous works. However, it also presents potential challenges in handling feature interactions. Firstly, Transformers may encounter information loss when capturing feature interactions. By relying on inner products to represent pairwise relationships, they compress raw interaction information, which can result in a degradation of fidelity. Secondly, due to the long-tail features distribution, feature fields with low information-abundance embeddings constrain the information abundance of other fields, leading to collapsed embedding matrices. To tackle these issues, we propose a Dual Attention Framework for Enhanced Feature Interaction, known as Dual Enhanced Attention. This framework integrates two attention mechanisms: the Combo-ID attention mechanism and the collapse-avoiding attention mechanism. The Combo-ID attention mechanism directly retains feature interaction pairs to mitigate information loss, while the collapse-avoiding attention mechanism adaptively filters out low information-abundance interaction pairs to prevent interaction collapse. Extensive experiments conducted on industrial datasets have shown the effectiveness of Dual Enhanced Attention.

cs.IR↗

arXiv2025

Graph Foundation Models: Concepts, Opportunities and Challenges

Foundation models have emerged as critical components in a variety of artificial intelligence applications, and showcase significant success in natural language processing and several other domains. Meanwhile, the field of graph machine learning is witnessing a paradigm transition from shallow methods to more sophisticated deep learning approaches. The capabilities of foundation models in generalization and adaptation motivate graph machine learning researchers to discuss the potential of developing a new graph learning paradigm. This paradigm envisions models that are pre-trained on extensive graph data and can be adapted for various graph tasks. Despite this burgeoning interest, there is a noticeable lack of clear definitions and systematic analyses pertaining to this new domain. To this end, this article introduces the concept of Graph Foundation Models (GFMs), and offers an exhaustive explanation of their key characteristics and underlying technologies. We proceed to classify the existing work related to GFMs into three distinct categories, based on their dependence on graph neural networks and large language models. In addition to providing a thorough review of the current state of GFMs, this article also outlooks potential avenues for future research in this rapidly evolving domain.

cs.LG↗

arXiv2025

Efficient Portrait Matte Creation With Layer Diffusion and Connectivity Priors

Learning effective deep portrait matting models requires training data of both high quality and large quantity. Neither quality nor quantity can be easily met for portrait matting, however. Since the most accurate ground-truth portrait mattes are acquired in front of the green screen, it is almost impossible to harvest a large-scale portrait matting dataset in reality. This work shows that one can leverage text prompts and the recent Layer Diffusion model to generate high-quality portrait foregrounds and extract latent portrait mattes. However, the portrait mattes cannot be readily in use due to significant generation artifacts. Inspired by the connectivity priors observed in portrait images, that is, the border of portrait foregrounds always appears connected, a connectivity-aware approach is introduced to refine portrait mattes. Building on this, a large-scale portrait matting dataset is created, termed LD-Portrait-20K, with $20,051$ portrait foregrounds and high-quality alpha mattes. Extensive experiments demonstrated the value of the LD-Portrait-20K dataset, with models trained on it significantly outperforming those trained on other datasets. In addition, comparisons with the chroma keying algorithm and an ablation study on dataset capacity further confirmed the effectiveness of the proposed matte creation approach. Further, the dataset also contributes to state-of-the-art video portrait matting, implemented by simple video segmentation and a trimap-based image matting model trained on this dataset.

cs.CV↗

arXiv2024

Data-centric Graph Learning: A Survey

The history of artificial intelligence (AI) has witnessed the significant impact of high-quality data on various deep learning models, such as ImageNet for AlexNet and ResNet. Recently, instead of designing more complex neural architectures as model-centric approaches, the attention of AI community has shifted to data-centric ones, which focuses on better processing data to strengthen the ability of neural models. Graph learning, which operates on ubiquitous topological data, also plays an important role in the era of deep learning. In this survey, we comprehensively review graph learning approaches from the data-centric perspective, and aim to answer three crucial questions: (1) when to modify graph data, (2) what part of the graph data needs modification to unlock the potential of various graph models, and (3) how to safeguard graph models from problematic data influence. Accordingly, we propose a novel taxonomy based on the stages in the graph learning pipeline, and highlight the processing methods for different data structures in the graph data, i.e., topology, feature and label. Furthermore, we analyze some potential problems embedded in graph data and discuss how to solve them in a data-centric manner. Finally, we provide some promising future directions for data-centric graph learning.

cs.LG↗

arXiv2024

DropEdge not Foolproof: Effective Augmentation Method for Signed Graph Neural Networks

The paper discusses signed graphs, which model friendly or antagonistic relationships using edges marked with positive or negative signs, focusing on the task of link sign prediction. While Signed Graph Neural Networks (SGNNs) have advanced, they face challenges like graph sparsity and unbalanced triangles. The authors propose using data augmentation (DA) techniques to address these issues, although many existing methods are not suitable for signed graphs due to a lack of side information. They highlight that the random DropEdge method, a rare DA approach applicable to signed graphs, does not enhance link sign prediction performance. In response, they introduce the Signed Graph Augmentation (SGA) framework, which includes a structure augmentation module to identify candidate edges and a strategy for selecting beneficial candidates, ultimately improving SGNN training. Experimental results show that SGA significantly boosts the performance of SGNN models, with a notable 32.3% improvement in F1-micro for SGCN on the Slashdot dataset.

cs.LG↗

arXiv2024

Locality-Sensitive Hashing-Based Efficient Point Transformer with Applications in High-Energy Physics

This study introduces a novel transformer model optimized for large-scale point cloud processing in scientific domains such as high-energy physics (HEP) and astrophysics. Addressing the limitations of graph neural networks and standard transformers, our model integrates local inductive bias and achieves near-linear complexity with hardware-friendly regular operations. One contribution of this work is the quantitative analysis of the error-complexity tradeoff of various sparsification techniques for building efficient transformers. Our findings highlight the superiority of using locality-sensitive hashing (LSH), especially OR & AND-construction LSH, in kernel approximation for large-scale point cloud data with local inductive bias. Based on this finding, we propose LSH-based Efficient Point Transformer (HEPT), which combines E$^2$LSH with OR & AND constructions and is built upon regular computations. HEPT demonstrates remarkable performance on two critical yet time-consuming HEP tasks, significantly outperforming existing GNNs and transformers in accuracy and computational speed, marking a significant advancement in geometric deep learning and large-scale scientific data processing. Our code is available at https://github.com/Graph-COM/HEPT.

cs.LG↗

arXiv2024

Angle-Aware Coverage with Camera Rotational Motion Control

This paper presents a novel control strategy for drone networks to improve the quality of 3D structures reconstructed from aerial images by drones. Unlike the existing coverage control strategies for this purpose, our proposed approach simultaneously controls both the camera orientation and drone translational motion, enabling more comprehensive perspectives and enhancing the map's overall quality. Subsequently, we present a novel problem formulation, including a new performance function to evaluate the drone positions and camera orientations. We then design a QP-based controller with a control barrier-like function for a constraint on the decay rate of the objective function. The present problem formulation poses a new challenge, requiring significantly greater computational efforts than the case involving only translational motion control. We approach this issue technologically, namely by introducing JAX, utilizing just-in-time (JIT) compilation and Graphical Processing Unit (GPU) acceleration. We finally conduct extensive verifications through simulation in ROS (Robot Operating System) and show the real-time feasibility of the controller and the superiority of the present controller to the conventional method.

math.OC↗

arXiv2020

Intelligent sampling for multiple change-points in exceedingly long time series with rate guarantees

Change point estimation in its offline version is traditionally performed by optimizing over the data set of interest, by considering each data point as the true location parameter and computing a data fit criterion. Subsequently, the data point that minimizes the criterion is declared as the change point estimate. For estimating multiple change points, the procedures are analogous in spirit, but significantly more involved in execution. Since change-points are local discontinuities, only data points close to the actual change point provide useful information for estimation, while data points far away are superfluous, to the point where using only a few points close to the true parameter is just as precise as using the full data set. Leveraging this "locality principle", we introduce a two-stage procedure for the problem at hand, which in the 1st stage uses a sparse subsample to obtain pilot estimates of the underlying change points, and in the 2nd stage refines these estimates by sampling densely in appropriately defined neighborhoods around them. We establish that this method achieves the same rate of convergence and even virtually the same asymptotic distribution as the analysis of the full data, while reducing computational complexity to O(N^0.5) time (N being the length of data set), as opposed to at least O(N) time for all current procedures, making it promising for the analysis on exceedingly long data sets with adequately spaced out change points. The main results are established under a signal plus noise model with independent and identically distributed error terms, but extensions to dependent data settings, as well as multiple stage (>2) procedures are also provided. The performance of our procedure -- which is coined "intelligent sampling" -- is illustrated on both synthetic and real Internet data streams.

stat.ME↗