arXiv subjects

Yuxuan Zhang

Publications and source records attributed to Yuxuan Zhang.

At least 91 records · Page 5Linked to original sources

arXiv2025

LENS: Learning to Segment Anything with Unified Reinforced Reasoning

Text-prompted image segmentation enables fine-grained visual understanding and is critical for applications such as human-computer interaction and robotics. However, existing supervised fine-tuning methods typically ignore explicit chain-of-thought (CoT) reasoning at test time, which limits their ability to generalize to unseen prompts and domains. To address this issue, we introduce LENS, a scalable reinforcement-learning framework that jointly optimizes the reasoning process and segmentation in an end-to-end manner. We propose unified reinforcement-learning rewards that span sentence-, box-, and segment-level cues, encouraging the model to generate informative CoT rationales while refining mask quality. Using a publicly available 3-billion-parameter vision-language model, i.e., Qwen2.5-VL-3B-Instruct, LENS achieves an average cIoU of 81.2% on the RefCOCO, RefCOCO+, and RefCOCOg benchmarks, outperforming the strong fine-tuned method, i.e., GLaMM, by up to 5.6%. These results demonstrate that RL-driven CoT reasoning significantly enhances text-prompted segmentation and offers a practical path toward more generalizable Segment Anything models (SAM). Code is available at https://github.com/hustvl/LENS.

cs.CV↗

arXiv2025

Depth-first directional search for nonconvex optimization

Random search methods are widely used for global optimization due to their theoretical generality and implementation simplicity. This paper proposes a depth-first directional search (DFDS) algorithm for globally solving nonconvex optimization problems. Motivated by the penetrating beam of a searchlight, DFDS performs a complete stepping line search along each sampled direction before proceeding to the next, contrasting with existing directional search methods that prioritize broad exploratory coverage. We establish the convergence and computational complexity of DFDS through a novel geometric framework that models the success probability of finding a global optimizer as the surface area of a spherical cap. Numerical experiments on benchmark problems demonstrate that DFDS achieves significantly higher accuracy in locating the global optimum compared to other random search methods under the same function evaluation budget.

math.OC↗

arXiv2025

Heuristic Quantum Advantage with Peaked Circuits

We design and demonstrate heuristic quantum advantage with peaked circuits (HQAP circuits) on Quantinuum's System Model H2 quantum processor. Through extensive experimentation with state-of-the-art classical simulation strategies, we identify a clear gap between classical and quantum runtimes. Our largest instance involves all-to-all connectivity with 2000 two-qubit gates, which H2 can produce the target peaked bitstring directly in under 2 hours. Our extrapolations from leading classical simulation techniques such as tensor networks with belief propagation and Pauli path simulators indicate the same instance would take years on exascale systems (Frontier, Summit), suggesting a potentially exponential separation. This work marks an important milestone toward verifiable quantum advantage, as well as providing a useful benchmarking protocol for current utility-scale quantum hardware. We sketch our protocol for designing these circuits and provide extensive numerical results leading to our extrapolation estimates. Separate from our constructed HQAP circuits, we prove hardness on a decision problem involving generic peaked circuits. When both the input and output bitstrings of a peaked circuit are unknown, determining whether the circuit is peaked constitutes a QCMA-complete problem, meaning the problem remains hard even for a quantum polynomial-time machine under commonly accepted complexity assumptions. Inspired by this observation, we propose an application of the peaked circuits as a potentially quantum-safe encryption scheme~\cite{chen2016report,kumar2020post,joseph2022transitioning,dam2023survey}. We make our peaked circuits publicly available and invite the community to try additional methods to solve these circuits to see if this gap persists even with novel classical techniques.

quant-ph↗

arXiv2025

XDXD: End-to-end crystal structure determination with low resolution X-ray diffraction

Determining crystal structures from X-ray diffraction data is fundamental across diverse scientific fields, yet remains a significant challenge when data is limited to low resolution. While recent deep learning models have made breakthroughs in solving the crystallographic phase problem, the resulting low-resolution electron density maps are often ambiguous and difficult to interpret. To overcome this critical bottleneck, we introduce XDXD, to our knowledge, the first end-to-end deep learning framework to determine a complete atomic model directly from low-resolution single-crystal X-ray diffraction data. Our diffusion-based generative model bypasses the need for manual map interpretation, producing chemically plausible crystal structures conditioned on the diffraction pattern. We demonstrate that XDXD achieves a 70.4\% match rate for structures with data limited to 2.0~Å resolution, with a root-mean-square error (RMSE) below 0.05. Evaluated on a benchmark of 24,000 experimental structures, our model proves to be robust and accurate. Furthermore, a case study on small peptides highlights the model's potential for extension to more complex systems, paving the way for automated structure solution in previously intractable cases.

cond-mat.mtrl-sci↗

arXiv2025

Protenix-Mini+: efficient structure prediction model with scalable pairformer

Lightweight inference is critical for biomolecular structure prediction and downstream tasks, enabling efficient real-world deployment and inference-time scaling for large-scale applications. While AF3 and its variants (e.g., Protenix, Chai-1) have advanced structure prediction results, they suffer from critical limitations: high inference latency and cubic time complexity with respect to token count, both of which restrict scalability for large biomolecular complexes. To address the core challenge of balancing model efficiency and prediction accuracy, we introduce three key innovations: (1) compressing non-scalable operations to mitigate cubic time complexity, (2) removing redundant blocks across modules to reduce unnecessary overhead, and (3) adopting a few-step sampler for the atom diffusion module to accelerate inference. Building on these design principles, we develop Protenix-Mini+, a highly lightweight and scalable variant of the Protenix model. Within an acceptable range of performance degradation, it substantially improves computational efficiency. For example, in the case of low-homology single-chain proteins, Protenix-Mini+ experiences an intra-protein LDDT drop of approximately 3% relative to the full Protenix model -- an acceptable performance trade-off given its substantially 90%+ improved computational efficiency.

q-bio.QM↗

arXiv2025

Complexity and hardness of random peaked circuits

Near-term feasibility, classical hardness, and verifiability are the three requirements for demonstrating quantum advantage; most existing quantum advantage proposals achieve at most two. A promising candidate recently proposed is through randomly generated peaked circuits. In this work, we study an explicit construction for random peaked circuits: first selecting a random circuit $C$ of polynomial size, which forms a $k$-design. Subsequently, a second random circuit $C'$ is chosen from the same architecture, subject to a postselection criterion: $C'$ must exhibit a high overlap with $C$ in one of their rows. Utilizing unitary design properties, we demonstrate that the circuits generated by this method are non-trivial; specifically, $C'$ is provably far from $C^\dagger$. Indeed, with overwhelmingly high probability, a random peaked circuit generated this way is non-compressible and is of circuit complexity $\tilde Ω(nk)$. This resolves an open problem posed by Aaronson in 2022. Secondly, we analytically establish that estimating the peakedness of a random peaked circuit to within a $2^{-\text{poly}(n)}$ additive error, is average-case \#P-hard. When the additive error is relaxed to $1/\text{poly}(n)$, we note that the worst-case scenario for this problem is BQP-complete. Under widely accepted assumptions on random quantum circuits, we identify a regime where no classical polynomial-time sequential simulator attains inverse-polynomial additive accuracy on the peak on a non-negligible fraction of instances. Thirdly, we study using peaked circuits as a practical attempt for a verifiable quantum advantage protocol. While the postselection method for generating peaked circuits could be costly, we demonstrate that numerical search for $C'$ with randomized initialization successfully returns a random peaked circuit, achieving the properties as theoretically predicted.

quant-ph↗

arXiv2025

VideoScore2: Think before You Score in Generative Video Evaluation

Recent advances in text-to-video generation have produced increasingly realistic and diverse content, yet evaluating such videos remains a fundamental challenge due to their multi-faceted nature encompassing visual quality, semantic alignment, and physical consistency. Existing evaluators and reward models are limited to single opaque scores, lack interpretability, or provide only coarse analysis, making them insufficient for capturing the comprehensive nature of video quality assessment. We present VideoScore2, a multi-dimensional, interpretable, and human-aligned framework that explicitly evaluates visual quality, text-to-video alignment, and physical/common-sense consistency while producing detailed chain-of-thought rationales. Our model is trained on a large-scale dataset VideoFeedback2 containing 27,168 human-annotated videos with both scores and reasoning traces across three dimensions, using a two-stage pipeline of supervised fine-tuning followed by reinforcement learning with Group Relative Policy Optimization (GRPO) to enhance analytical robustness. Extensive experiments demonstrate that VideoScore2 achieves superior performance with 44.35 (+5.94) accuracy on our in-domain benchmark VideoScore-Bench-v2 and 50.37 (+4.32) average performance across four out-of-domain benchmarks (VideoGenReward-Bench, VideoPhy2, etc), while providing interpretable assessments that bridge the gap between evaluation and controllable generation through effective reward modeling for Best-of-N sampling. Project Page: https://tiger-ai-lab.github.io/VideoScore2/

cs.CV↗

arXiv2025

PromptSleuth: Detecting Prompt Injection via Semantic Intent Invariance

Large Language Models (LLMs) are increasingly integrated into real-world applications, from virtual assistants to autonomous agents. However, their flexibility also introduces new attack vectors-particularly Prompt Injection (PI), where adversaries manipulate model behavior through crafted inputs. As attackers continuously evolve with paraphrased, obfuscated, and even multi-task injection strategies, existing benchmarks are no longer sufficient to capture the full spectrum of emerging threats. To address this gap, we construct a new benchmark that systematically extends prior efforts. Our benchmark subsumes the two widely-used existing ones while introducing new manipulation techniques and multi-task scenarios, thereby providing a more comprehensive evaluation setting. We find that existing defenses, though effective on their original benchmarks, show clear weaknesses under our benchmark, underscoring the need for more robust solutions. Our key insight is that while attack forms may vary, the adversary's intent-injecting an unauthorized task-remains invariant. Building on this observation, we propose PromptSleuth, a semantic-oriented defense framework that detects prompt injection by reasoning over task-level intent rather than surface features. Evaluated across state-of-the-art benchmarks, PromptSleuth consistently outperforms existing defense while maintaining comparable runtime and cost efficiency. These results demonstrate that intent-based semantic reasoning offers a robust, efficient, and generalizable strategy for defending LLMs against evolving prompt injection threats.

cs.CR↗

arXiv2025

A deep learning model for chemical shieldings in molecular organic solids including anisotropy

Nuclear Magnetic Resonance (NMR) chemical shifts are powerful probes of local atomic and electronic structure that can be used to resolve the structures of powdered or amorphous molecular solids. Chemical shift driven structure elucidation depends critically on accurate and fast predictions of chemical shieldings, and machine learning (ML) models for shielding predictions are increasingly used as scalable and efficient surrogates for demanding ab initio calculations. However, the prediction accuracies of current ML models still lag behind those of the DFT reference methods they approximate, especially for nuclei such as $^{13}$C and $^{15}$N. Here, we introduce ShiftML3.0, a deep-learning model that improves the accuracy of predictions of isotropic chemical shieldings in molecular solids, and does so while also predicting the full shielding tensor. On experimental benchmark sets, we find root-mean-squared errors with respect to experiment for ShiftML3.0 that approach those of DFT reference calculations, with RMSEs of 0.53 ppm for $^{1}$H, 2.4 ppm for $^{13}$C, and 7.2 ppm for $^{15}$N, compared to DFT values of 0.49 ppm, 2.3 ppm, and 5.8 ppm, respectively.

physics.chem-ph↗

arXiv2025

On-device Anomaly Detection in Conveyor Belt Operations

Conveyor belts are crucial in mining operations by enabling the continuous and efficient movement of bulk materials over long distances, which directly impacts productivity. While detecting anomalies in specific conveyor belt components has been widely studied, identifying the root causes of these failures, such as changing production conditions and operator errors, remains critical. Continuous monitoring of mining conveyor belt work cycles is still at an early stage and requires robust solutions. Recently, an anomaly detection method for duty cycle operations of a mining conveyor belt has been proposed. Based on its limited performance and unevaluated long-term proper operation, this study proposes two novel methods for classifying normal and abnormal duty cycles. The proposed approaches are pattern recognition systems that make use of threshold-based duty-cycle detection mechanisms, manually extracted features, pattern-matching, and supervised tiny machine learning models. The explored low-computational models include decision tree, random forest, extra trees, extreme gradient boosting, Gaussian naive Bayes, and multi-layer perceptron. A comprehensive evaluation of the former and proposed approaches is carried out on two datasets. Both proposed methods outperform the former method in anomaly detection, with the best-performing approach being dataset-dependent. The heuristic rule-based approach achieves the highest F1-score in the same dataset used for algorithm training, with 97.3% for normal cycles and 80.2% for abnormal cycles. The ML-based approach performs better on a dataset including the effects of machine aging, with an F1-score scoring 91.3% for normal cycles and 67.9% for abnormal cycles. Implemented on two low-power microcontrollers, the methods demonstrate efficient, real-time operation with energy consumption of 13.3 and 20.6 \textmu J during inference. These results ...

cs.LG↗

arXiv2025

Adaptive Noise Resilient Keyword Spotting Using One-Shot Learning

Keyword spotting (KWS) is a key component of smart devices, enabling efficient and intuitive audio interaction. However, standard KWS systems deployed on embedded devices often suffer performance degradation under real-world operating conditions. Resilient KWS systems address this issue by enabling dynamic adaptation, with applications such as adding or replacing keywords, adjusting to specific users, and improving noise robustness. However, deploying resilient, standalone KWS systems with low latency on resource-constrained devices remains challenging due to limited memory and computational resources. This study proposes a low computational approach for continuous noise adaptation of pretrained neural networks used for KWS classification, requiring only 1-shot learning and one epoch. The proposed method was assessed using two pretrained models and three real-world noise sources at signal-to-noise ratios (SNRs) ranging from 24 to -3 dB. The adapted models consistently outperformed the pretrained models across all scenarios, especially at SNR $\leq$ 18 dB, achieving accuracy improvements of 4.9% to 46.0%. These results highlight the efficacy of the proposed methodology while being lightweight enough for deployment on resource-constrained devices.

cs.SD↗

arXiv2025

Biorthogonal Neural Network Approach to Two-Dimensional Non-Hermitian Systems

Non-Hermitian quantum many-body systems exhibit a rich array of physical phenomena, including non-Hermitian skin effects and exceptional points, that remain largely inaccessible to existing numerical techniques. In this work, we investigate the application of variational Monte Carlo and neural network wavefunction representations to examine their ground-state (the eigenstate with the smallest real part energy) properties. Due to the breakdown of the Rayleigh-Ritz variational principle in non-Hermitian settings, we develop a self-consistent symmetric optimization framework based on variance minimization with a dynamically updated energy estimate. Our approach respects the biorthogonal structure of left and right eigenstates, and is further strengthened by exploiting system symmetries and pseudo-Hermiticity. Tested on a two-dimensional non-Hermitian transverse field Ising model endowed with a complex longitudinal field, our method achieves high accuracy across both parity-time symmetric and broken phases. Moreover, we propose novel optimization routines that address the challenges posed by exceptional points and provide reliable convergence to the ground state in regimes where standard variational techniques fail. Lastly, we show, through extensive numerical evidence, that our method offers a scalable and flexible computational tool to investigate non-Hermitian quantum many-body systems, beyond the reach of conventional numerical techniques such as the density-matrix renormalization group algorithm.

quant-ph↗

arXiv2025

VL-Explore: Zero-shot Vision-Language Exploration and Target Discovery by Mobile Robots

Vision-language navigation (VLN) has emerged as a promising paradigm, enabling mobile robots to perform zero-shot inference and execute tasks without specific pre-programming. However, current systems often separate map exploration and path planning, with exploration relying on inefficient algorithms due to limited (partially observed) environmental information. In this paper, we present a novel navigation pipeline named "VL-Explore" for simultaneous exploration and target discovery in unknown environments, leveraging the capabilities of a vision-language model named CLIP. Our approach requires only monocular vision and operates without any prior map or knowledge about the target. For comprehensive evaluations, we designed a functional prototype of a UGV (unmanned ground vehicle) system named "Open Rover", a customized platform for general-purpose VLN tasks. We integrated and deployed the VL-Explore pipeline on Open Rover to evaluate its throughput, obstacle avoidance capability, and trajectory performance across various real-world scenarios. Experimental results demonstrate that VL-Explore consistently outperforms traditional map-traversal algorithms and achieves performance comparable to path-planning methods that depend on prior map and target knowledge. Notably, VL-Explore offers real-time active navigation without requiring pre-captured candidate images or pre-built node graphs, addressing key limitations of existing VLN pipelines.

cs.RO↗

arXiv2025

Controllable Weather Synthesis and Removal with Video Diffusion Models

Generating realistic and controllable weather effects in videos is valuable for many applications. Physics-based weather simulation requires precise reconstructions that are hard to scale to in-the-wild videos, while current video editing often lacks realism and control. In this work, we introduce WeatherWeaver, a video diffusion model that synthesizes diverse weather effects -- including rain, snow, fog, and clouds -- directly into any input video without the need for 3D modeling. Our model provides precise control over weather effect intensity and supports blending various weather types, ensuring both realism and adaptability. To overcome the scarcity of paired training data, we propose a novel data strategy combining synthetic videos, generative image editing, and auto-labeled real-world videos. Extensive evaluations show that our method outperforms state-of-the-art methods in weather simulation and removal, providing high-quality, physically plausible, and scene-identity-preserving results over various real-world videos.

cs.GR↗

arXiv2025

Protenix-Mini: Efficient Structure Predictor via Compact Architecture, Few-Step Diffusion and Switchable pLM

Lightweight inference is critical for biomolecular structure prediction and other downstream tasks, enabling efficient real-world deployment and inference-time scaling for large-scale applications. In this work, we address the challenge of balancing model efficiency and prediction accuracy by making several key modifications, 1) Multi-step AF3 sampler is replaced by a few-step ODE sampler, significantly reducing computational overhead for the diffusion module part during inference; 2) In the open-source Protenix framework, a subset of pairformer or diffusion transformer blocks doesn't make contributions to the final structure prediction, presenting opportunities for architectural pruning and lightweight redesign; 3) A model incorporating an ESM module is trained to substitute the conventional MSA module, reducing MSA preprocessing time. Building on these key insights, we present Protenix-Mini, a compact and optimized model designed for efficient protein structure prediction. This streamlined version incorporates a more efficient architectural design with a two-step Ordinary Differential Equation (ODE) sampling strategy. By eliminating redundant Transformer components and refining the sampling process, Protenix-Mini significantly reduces model complexity with slight accuracy drop. Evaluations on benchmark datasets demonstrate that it achieves high-fidelity predictions, with only a negligible 1 to 5 percent decrease in performance on benchmark datasets compared to its full-scale counterpart. This makes Protenix-Mini an ideal choice for applications where computational resources are limited but accurate structure prediction remains crucial.

cs.LG↗

arXiv2025

GroundingSuite: Measuring Complex Multi-Granular Pixel Grounding

Pixel grounding, encompassing tasks such as Referring Expression Segmentation (RES), has garnered considerable attention due to its immense potential for bridging the gap between vision and language modalities. However, advancements in this domain are currently constrained by limitations inherent in existing datasets, including limited object categories, insufficient textual diversity, and a scarcity of high-quality annotations. To mitigate these limitations, we introduce GroundingSuite, which comprises: (1) an automated data annotation framework leveraging multiple Vision-Language Model (VLM) agents; (2) a large-scale training dataset encompassing 9.56 million diverse referring expressions and their corresponding segmentations; and (3) a meticulously curated evaluation benchmark consisting of 3,800 images. The GroundingSuite training dataset facilitates substantial performance improvements, enabling models trained on it to achieve state-of-the-art results. Specifically, a cIoU of 68.9 on gRefCOCO and a gIoU of 55.3 on RefCOCOm. Moreover, the GroundingSuite annotation framework demonstrates superior efficiency compared to the current leading data annotation method, i.e., $4.5 \times$ faster than GLaMM.

cs.CV↗

arXiv2025

General Modular Harness for LLM Agents in Multi-Turn Gaming Environments

We introduce a modular harness design for LLM agents that composes of perception, memory, and reasoning components, enabling a single LLM or VLM backbone to tackle a wide spectrum of multi turn gaming environments without domain-specific engineering. Using classic and modern game suites as low-barrier, high-diversity testbeds, our framework provides a unified workflow for analyzing how each module affects performance across dynamic interactive settings. Extensive experiments demonstrate that the harness lifts gameplay performance consistently over un-harnessed baselines and reveals distinct contribution patterns, for example, memory dominates in long-horizon puzzles while perception is critical in vision noisy arcades. These findings highlight the effectiveness of our modular harness design in advancing general-purpose agent, given the familiarity and ubiquity of games in everyday human experience.

cs.AI↗

arXiv2025

Stable-Hair v2: Real-World Hair Transfer via Multiple-View Diffusion Model

While diffusion-based methods have shown impressive capabilities in capturing diverse and complex hairstyles, their ability to generate consistent and high-quality multi-view outputs -- crucial for real-world applications such as digital humans and virtual avatars -- remains underexplored. In this paper, we propose Stable-Hair v2, a novel diffusion-based multi-view hair transfer framework. To the best of our knowledge, this is the first work to leverage multi-view diffusion models for robust, high-fidelity, and view-consistent hair transfer across multiple perspectives. We introduce a comprehensive multi-view training data generation pipeline comprising a diffusion-based Bald Converter, a data-augment inpainting model, and a face-finetuned multi-view diffusion model to generate high-quality triplet data, including bald images, reference hairstyles, and view-aligned source-bald pairs. Our multi-view hair transfer model integrates polar-azimuth embeddings for pose conditioning and temporal attention layers to ensure smooth transitions between views. To optimize this model, we design a novel multi-stage training strategy consisting of pose-controllable latent IdentityNet training, hair extractor training, and temporal attention training. Extensive experiments demonstrate that our method accurately transfers detailed and realistic hairstyles to source subjects while achieving seamless and consistent results across views, significantly outperforming existing methods and establishing a new benchmark in multi-view hair transfer. Code is publicly available at https://github.com/sunkymepro/StableHairV2.

cs.CV↗