Searcharxiv⌕ Search

arXiv subjects

Tao Wang

Publications and source records attributed to Tao Wang.

At least 163 records · Page 9Linked to original sources
arXiv2026

Distilling Latent Manifolds: Resolution Extrapolation by Variational Autoencoders

Variational Autoencoder (VAE) encoders play a critical role in modern generative models, yet their computational cost often motivates the use of knowledge distillation or quantification to obtain compact alternatives. Existing studies typically believe that the model work better on the samples closed to their training data distribution than unseen data distribution. In this work, we report a counter-intuitive phenomenon in VAE encoder distillation: a compact encoder distilled only at low resolutions exhibits poor reconstruction performance at its native resolution, but achieves dramatically improved results when evaluated at higher, unseen input resolutions. Despite never being trained beyond $256^2$ resolution, the distilled encoder generalizes effectively to $512^2$ resolution inputs, partially inheriting the teacher model's resolution preference.We further analyze latent distributions across resolutions and find that higher-resolution inputs produce latent representations more closely aligned with the teacher's manifold. Through extensive experiments on ImageNet-256, we show that simple resolution remapping-upsampling inputs before encoding and downsampling reconstructions for evaluation-leads to substantial gains across PSNR, MSE, SSIM, LPIPS, and rFID metrics. These findings suggest that VAE encoder distillation learns resolution-consistent latent manifolds rather than resolution-specific pixel mappings. This also means that the high training cost on memory, time and high-resolution datasets are not necessary conditions for distilling a VAE with high-resolution image reconstruction capabilities. On low resolution datasets, the distillation model still could learn the detailed knowledge of the teacher model in high-resolution image reconstruction.

cs.CV↗
arXiv2026

Let It Flow: Agentic Crafting on Rock and Roll, Building the ROME Model within an Open Agentic Learning Ecosystem

Agentic crafting requires LLMs to operate in real-world environments over multiple turns by taking actions, observing outcomes, and iteratively refining artifacts. Despite its importance, the open-source community lacks a principled, end-to-end ecosystem to streamline agent development. We introduce the Agentic Learning Ecosystem (ALE), a foundational infrastructure that optimizes the production pipeline for agentic model. ALE consists of three components: ROLL, a post-training framework for weight optimization; ROCK, a sandbox environment manager for trajectory generation; and iFlow CLI, an agent framework for efficient context engineering. We release ROME, an open-source agent grounded by ALE and trained on over one million trajectories. Our approach includes data composition protocols for synthesizing complex behaviors and a novel policy optimization algorithm, Interaction-Perceptive Agentic Policy Optimization (IPA), which assigns credit over semantic interaction chunks rather than individual tokens to improve long-horizon training stability. Empirically, we evaluate ROME within a structured setting and introduce Terminal Bench Pro, a benchmark with improved scale and contamination control. ROME demonstrates strong performance across benchmarks like SWE-bench Verified and Terminal Bench, proving the effectiveness of ALE.

cs.AI↗
arXiv2026

Origin and Propagation of Spin-orbit Torques in Pt/Co/Cu/NiFe/Capping Multilayers

Spin-orbit torque (SOT) enables efficient current-driven control of magnetization, offering a promising pathway toward low-power spintronic devices. However, the origin and propagation of both damping-like (DL) and field-like (FL) SOTs in complex multilayers remain unclear. Here, we investigate NiFe thickness-dependent SOT efficiencies in Ta/Pt/Co/Cu/NiFe/Cu/Capping multilayers (x = 15 nm; Capping = Pt, Al, and SiO2). By employing a spin rotation geometry, the perpendicularly magnetized Pt/Co/Cu stacks serve as a spin source introducing unconventional spin polarization orthogonal to the Oersted field, eliminating its contribution and enabling unambiguous separation of SOTs using planar Hall and polar MOKE measurements. To distinguish bulk and interfacial contributions, we introduce a sample-area-normalized moment m = mNiFe/S, accounting for thickness-dependent magnetization and eliminating uncertainties arising from nominal thickness scaling and magnetic dead layers. We find that DL-SOT follows nearly linear 1/m scaling, consistent with rapid spin absorption at the Cu/NiFe interface but exhibits finite beta_SOT when 1/m approaches zero in both Pt- and Al-capped samples, indicating additional interfacial spin-current contributions at Cu/Pt and Cu/Al interfaces. In contrast, SiO2-capped samples show negligible interfacial contributions. Furthermore, FL-SOT deviates markedly from 1/m scaling and exhibits a significantly longer spin dephasing length (about 1.7 nm) compared to DL-SOT, implying extended propagation across NiFe. Comparative capping-layer studies further corroborate this behavior through interface-dependent spin transport. Our findings clarify the origin and distinct propagation characteristics of DL and FL torques, providing guidelines for engineering interfacial spin-orbit functionalities in ultrathin metallic heterostructures.

cond-mat.mtrl-sci↗
arXiv2026

Slow-Fast Inference: Training-Free Inference Acceleration via Within-Sentence Support Stability

Long-context autoregressive decoding remains expensive because each decoding step must repeatedly process a growing history. We observe a consistent pattern during decoding: within a sentence, and more generally within a short semantically coherent span, the dominant attention support often remains largely stable. Motivated by this observation, we propose Slow-Fast Inference (SFI), a training-free decoding framework that decouples generation into frequent low-cost fast steps and occasional dense-attention slow steps. Fast steps reuse a compact sparse memory for efficient decoding. Slow steps are triggered near semantic boundaries. At slow steps, the model revisits the broader context and uses the Selector to refresh the selected memory for subsequent fast steps. Across the evaluated context lengths, SFI delivers approximately $1.6\times$--$14.4\times$ higher decoding throughput while generally maintaining quality on par with the full-KV baseline across long-context and long-CoT settings. Because SFI is training-free and applies directly to existing checkpoints, it offers a practical path to reducing inference cost for contemporary autoregressive reasoning models in long-context, long-horizon, and agentic workloads.

cs.LG↗
arXiv2026

A systematic study of AGN feedback in a disk galaxy I: global overview

This is the first paper in a series using our MACER framework to investigate the evolution of a disk galaxy, which emphasizes the role of active galactic nucleus (AGN) feedback and incorporates cosmological inflows. This paper presents the model setup and the overall results. The predicted AGN duty cycle of approximately 0.49% is consistent with observations. Analysis of the AGN luminosity and star formation rate (SFR) light curves reveals a positive correlation between the two. We find that cold filaments condense in the circumgalactic medium (CGM) region due to radiative cooling and subsequently fall onto the galaxy, significantly enhancing both the SFR and AGN activity. The galaxy is then quenched over a timescale of approximately 1 Gyr by the strong feedback from the enhanced AGN activity. This indicates that a positive correlation between SFR and AGN luminosity does not preclude AGN feedback from acting as the quenching mechanism for the galaxy. Notably, models without AGN feedback exhibit significantly lower peak SFRs than those with it. We attribute this difference to cumulative AGN feedback, which drives gas from the galaxy into the CGM, facilitating the formation of more massive cold filaments and ultimately promoting more intense starburst episodes.

astro-ph.GA↗
arXiv2026

Lightening the Load: A Cluster-Based Framework for A Lower-Overhead, Provable Website Fingerprinting Defense

Website fingerprinting (WF) attacks remain a significant threat to encrypted traffic, prompting the development of a wide range of defenses. Among these, two prominent classes are regularization-based defenses, which shape traffic using fixed padding rules, and supersequence-based approaches, which conceal traces among predefined patterns. In this work, we present a unified framework for designing an adaptive WF defense that combines the effectiveness of regularization with the provable security of supersequence-style grouping. The scheme first extracts behavioural patterns from traces and clusters them into (k,l)-diverse anonymity sets; an early-time-series classifier (adapted from ECDIRE) then switches from a conservative global set of regularization parameters to the lighter, set-specific parameters. We instantiate the design as Adaptive Tamaraw, a variant of Tamaraw that assigns padding parameters on a per-cluster basis while retaining its original information-theoretic guarantee. Comprehensive experiments on public real-world datasets confirm the benefits. By tuning k, operators can trade privacy for efficiency: in its high-privacy mode Adaptive Tamaraw pushes the bound on any attacker's accuracy below 30%, whereas in efficiency-centred settings it cuts total overhead by 99% compared with classic Tamaraw.

cs.CR↗
arXiv2026

Homogenization of the Stetson Photometry with the BEST Database

As one of the most widely recognized high-quality standard stars, the Stetson standards have been extensively used as a photometric reference for calibrating other surveys. In this work, we present an independent validation and re-calibration of the Stetson standard star photometry using the BEST database. Based on typically 30,000-70,000 calibration stars per band, we find that the original Stetson photometry achieves field-to-field zero-point precisions of approximately 10--40\,mmag in the $UBVRI$-band. In addition, significant spatially dependent magnitude offsets are detected within individual Stetson fields for all bands, with magnitudes exceeding 1\%, probably caused by the calibration errors in the Stetson photometry. After correcting those systematic errors, the agreement between the Stetson and BEST photometry is improved to $\sim$5\,mmag for individual fields for $BVRI$-band. The re-calibrated photometry is further validated using the SCR standards, yielding agreement better than 10\,mmag for individual stars in the $BVRI$ bands and confirming zero-point precisions of 2--4\,mmag in the $BVI$ band. The precisions is further confirmed by checks using Gaia DR3 broad band colors. These results highlight the power of the BEST database for improving photometric calibration and suggest that, if feasible, it be incorporated into the calibration process of future releases of the Stetson standard catalog.

astro-ph.SR↗
arXiv2026

Reality Check for Tor Website Fingerprinting in the Open World

Website fingerprinting (WF) attacks on Tor can infer user destinations from encrypted traffic metadata. However, their real-world effectiveness remains debated due to laboratory settings that fail to capture network fluctuations, evaluate noise, and create a representative open world. In this work, we re-examine WF from a guard-relay vantage point using a novel, privacy-preserving methodology that builds an open-world background from real, unlabeled Tor traffic paired with synthetic monitored traces. Using this methodology, we collect a large-scale dataset of over 800,000 traces. We then benchmark state-of-the-art WF attacks under a cross-network setting and show that WF remains highly effective against real Tor open-world traffic: the best-performing attack achieves 0.956 precision and 0.922 recall at a 9% base rate. We further present results that demonstrate robustness to small training sets, network jitter, and concept drift. Moreover, we show that timing-independent classifiers are significantly more robust to network variability than others. Finally, we provide the first systematic study of Tor's Conflux traffic-splitting, where we show that a guard node with a latency advantage can maintain high attack effectiveness even when traffic is split.

cs.CR↗
arXiv2026

Dr. Seg: Revisiting GRPO Training for Visual Large Language Models through Perception-Oriented Design

Following the success of Group Relative Policy Optimization (GRPO) in foundation LLMs, an increasing number of works have sought to adapt GRPO to Visual Large Language Models (VLLMs) for visual perception tasks (e.g., detection and segmentation). However, much of this line of research rests on a long-standing yet unexamined assumption: training paradigms developed for language reasoning can be transferred seamlessly to visual perception. Our experiments show that this assumption is not valid, revealing intrinsic differences between reasoning-oriented and perception-oriented settings. Using reasoning segmentation as a representative case, we surface two overlooked factors: (i) the need for a broader output space, and (ii) the importance of fine-grained, stable rewards. Building on these observations, we propose Dr.~Seg, a simple, plug-and-play GRPO-based framework consisting of a Look-to-Confirm mechanism and a Distribution-Ranked Reward module, requiring no architectural modifications and integrating seamlessly with existing GRPO-based VLLMs. Extensive experiments demonstrate that Dr.~Seg improves performance in complex visual scenarios while maintaining strong generalization. Code, models, and datasets are available at https://github.com/eVI-group-SCU/Dr-Seg.

cs.CV↗
arXiv2026

Robust and cost-effective quantum network using Kramers-Kronig receiver

The quantum internet holds the potential to facilitate applications that are fundamentally inaccessible to the classical internet. Among its most prominent applications is quantum key distribution (QKD) networks, which connect two distant nodes to establish a secure key based on the principles of quantum mechanics. However, the subsequent extensive reliance on interferences in existing QKD protocols leads to the weak robustness of the system and the corresponding network. In this work, we propose a robust and cost-effective quantum network using the Kramers-Kronig receiver. We first propose a continuous-variable QKD protocol based on direct detection without interference, which achieves the recovery of quadrature components through the Kramers-Kronig relation. Subsequently, we have extended this protocol to continuous-variable quantum access networks, further highlighting the robustness and cost advantages of interference-free detection. The experimental results show that each user can achieve a secret key rate at 50 kbit/s within the access network range by using only one photodetector without interference structures. This scheme opens up new possibilities in establishing a robust and cost-effective quantum network, serving as a foundational element in the progress toward establishing a large-scale quantum internet.

quant-ph↗
arXiv2026

AnimeAgent: Is the Multi-Agent via Image-to-Video models a Good Disney Storytelling Artist?

Custom Storyboard Generation (CSG) aims to produce high-quality, multi-character consistent storytelling. Current approaches based on static diffusion models, whether used in a one-shot manner or within multi-agent frameworks, face three key limitations: (1) Static models lack dynamic expressiveness and often resort to "copy-paste" pattern. (2) One-shot inference cannot iteratively correct missing attributes or poor prompt adherence. (3) Multi-agents rely on non-robust evaluators, ill-suited for assessing stylized, non-realistic animation. To address these, we propose AnimeAgent, the first Image-to-Video (I2V)-based multi-agent framework for CSG. Inspired by Disney's "Combination of Straight Ahead and Pose to Pose" workflow, AnimeAgent leverages I2V's implicit motion prior to enhance consistency and expressiveness, while a mixed subjective-objective reviewer enables reliable iterative refinement. We also collect a human-annotated CSG benchmark with ground-truth. Experiments show AnimeAgent achieves SOTA performance in consistency, prompt fidelity, and stylization.

cs.CV↗
arXiv2026

FFINO: Factorized Fourier Improved Neural Operator for Modeling Multiphase Flow in Underground Hydrogen Storage

Underground hydrogen storage (UHS) is a promising energy storage option for the current energy transition to a low-carbon economy. Fast modeling of hydrogen plume migration and pressure field evolution is crucial for UHS field management. In this study, a new neural operator architecture, factorized Fourier improved neural operator or FFINO is proposed as a fast surrogate model for multiphase flow problems in UHS. Experimental relative permeability curves reported in the literature are also parameterized as key uncertainty parameters for the FFINO model. FFINO model performance with the state-of-the-art Fourier-enhanced multiple-input neural operators or FMIONet model are systematically studied through a comprehensive combination of metrics. Our new FFINO model has 38.1% fewer trainable parameters, 17.6% less training time, and 12% less GPU memory cost compared to FMIONet. The FFINO model also achieves a 9.8% accuracy improvement in predicting hydrogen plume in focused areas, and 16.3% higher accuracy in predicting pressure buildup. Sensitivity analysis identifies that the most influential input parameter to models' performance is the injection rate Q, while other parameters show moderate to minor impacts. The inference time of the trained FFINO model is 7,850 times faster than a numerical simulator, which guarantees its superior time efficiency. The novel FFINO model can serve as a fast, accurate, and stable alternative to estimate the temporal and spatial evolution of hydrogen plumes and pressure distributions for real-time UHS applications.

cs.LG↗
arXiv2026

A Domain-Adapted Pipeline for Structured Information Extraction from Police Incident Announcements on Social Media

Structured information extraction from police incident announcements is crucial for timely and accurate data processing, yet presents considerable challenges due to the variability and informal nature of textual sources such as social media posts. To address these challenges, we developed a domain-adapted extraction pipeline that leverages targeted prompt engineering with parameter-efficient fine-tuning of the Qwen2.5-7B model using Low-Rank Adaptation (LoRA). This approach enables the model to handle noisy, heterogeneous text while reliably extracting 15 key fields, including location, event characteristics, and impact assessment, from a high-quality, manually annotated dataset of 4,933 instances derived from 27,822 police briefing posts on Chinese Weibo (2019-2020). Experimental results demonstrated that LoRA-based fine-tuning significantly improved performance over both the base and instruction-tuned models, achieving an accuracy exceeding 98.36% for mortality detection and Exact Match Rates of 95.31% for fatality counts and 95.54% for province-level location extraction. The proposed pipeline thus provides a validated and efficient solution for multi-task structured information extraction in specialized domains, offering a practical framework for transforming unstructured text into reliable structured data in social science research.

cs.CL↗
arXiv2026

Adiabatic Elimination in Relativistic Stochastic Mechanics

We investigate the adiabatic elimination of fast variables in relativistic stochastic mechanics, which is analyzed by using the equation of motion and the distribution function, with relativistic corrections explicitly derived. A new dimensionless parameter is introduced to characterize the timescale. The adiabatic elimination is compared with the path integral coarse graining, which is more general yet computationally demanding.

cond-mat.stat-mech↗
arXiv2026

TrapFlow: Controllable Website Fingerprinting Defense via Dynamic Backdoor Learning

Website fingerprinting (WF) attacks, which covertly monitor user communications to identify the web pages they visit, pose a serious threat to user privacy. Existing WF defenses attempt to reduce attack accuracy by disrupting traffic patterns, but attackers can retrain their models to adapt, making these defenses ineffective. Meanwhile, their high overhead limits deployability. To overcome these limitations, we introduce a novel controllable website fingerprinting defense called TrapFlow based on backdoor learning. TrapFlow exploits the tendency of neural networks to memorize subtle patterns by injecting crafted trigger sequences into targeted website traffic, causing the attacker model to build incorrect associations during training. If the attacker attempts to adapt by training on such noisy data, TrapFlow ensures that the model internalizes the trigger as a dominant feature, leading to widespread misclassification across unrelated websites. Conversely, if the attacker ignores these patterns and trains only on clean data, the trigger behaves as an adversarial patch at inference time, causing model misclassification. To achieve this dual effect, we optimize the trigger using a Fast Levenshtein like distance to maximize both its learnability and its distinctiveness from normal traffic. Experiments show that TrapFlow significantly reduces the accuracy of the RF attack from 99 percent to 6 percent with 74 percent data overhead. This compares favorably against two state of the art defenses: FRONT reduces accuracy by only 2 percent at a similar overhead, while Palette achieves 32 percent accuracy but with 48 percent more overhead. We further validate the practicality of our method in a real Tor network environment.

cs.CR↗
arXiv2026

Foundations of Top-$k$ Decoding For Language Models

Top-$k$ decoding is a widely used method for sampling from LLMs: at each token, only the largest $k$ next-token-probabilities are kept, and the next token is sampled after re-normalizing them to sum to unity. Top-$k$ and other sampling methods are motivated by the intuition that true next-token distributions are sparse, and the noisy LLM probabilities need to be truncated. However, to our knowledge, a precise theoretical motivation for the use of top-$k$ decoding is missing. In this work, we develop a theoretical framework that both explains and generalizes top-$k$ decoding. We view decoding at a fixed token as the recovery of a sparse probability distribution. We consider \emph{Bregman decoders} obtained by minimizing a separable Bregman divergence (for both the \emph{primal} and \emph{dual} cases) with a sparsity-inducing $\ell_0$ regularization. Despite the combinatorial nature of the objective, we show how to optimize it efficiently for a large class of divergences. We show that the optimal decoding strategies are greedy, and further that the loss function is discretely convex in $k$, so that binary search provably and efficiently finds the optimal $k$. We show that top-$k$ decoding arises as a special case for the KL divergence, and identify new decoding strategies that have distinct behaviors (e.g., non-linearly up-weighting larger probabilities after re-normalization).

cs.AI↗
arXiv2026

Adaptive Image Zoom-in with Bounding Box Transformation for UAV Object Detection

Detecting objects from UAV-captured images is challenging due to the small object size. In this work, a simple and efficient adaptive zoom-in framework is explored for object detection on UAV images. The main motivation is that the foreground objects are generally smaller and sparser than those in common scene images, which hinders the optimization of effective object detectors. We thus aim to zoom in adaptively on the objects to better capture object features for the detection task. To achieve the goal, two core designs are required: \textcolor{black}{i) How to conduct non-uniform zooming on each image efficiently? ii) How to enable object detection training and inference with the zoomed image space?} Correspondingly, a lightweight offset prediction scheme coupled with a novel box-based zooming objective is introduced to learn non-uniform zooming on the input image. Based on the learned zooming transformation, a corner-aligned bounding box transformation method is proposed. The method warps the ground-truth bounding boxes to the zoomed space to learn object detection, and warps the predicted bounding boxes back to the original space during inference. We conduct extensive experiments on three representative UAV object detection datasets, including VisDrone, UAVDT, and SeaDronesSee. The proposed ZoomDet is architecture-independent and can be applied to an arbitrary object detection architecture. Remarkably, on the SeaDronesSee dataset, ZoomDet offers more than 8.4 absolute gain of mAP with a Faster R-CNN model, with only about 3 ms additional latency. The code is available at https://github.com/twangnh/zoomdet_code.

cs.CV↗
arXiv2026

Kelix Technical Report

Autoregressive large language models (LLMs) scale well by expressing diverse tasks as sequences of discrete natural-language tokens and training with next-token prediction, which unifies comprehension and generation under self-supervision. Extending this paradigm to multimodal data requires a shared, discrete representation across modalities. However, most vision-language models (VLMs) still rely on a hybrid interface: discrete text tokens paired with continuous Vision Transformer (ViT) features. Because supervision is largely text-driven, these models are often biased toward understanding and cannot fully leverage large-scale self-supervised learning on non-text data. Recent work has explored discrete visual tokenization to enable fully autoregressive multimodal modeling, showing promising progress toward unified understanding and generation. Yet existing discrete vision tokens frequently lose information due to limited code capacity, resulting in noticeably weaker understanding than continuous-feature VLMs. We present Kelix, a fully discrete autoregressive unified model that closes the understanding gap between discrete and continuous visual representations.

cs.CV↗