arXiv subjects

Yuhao Zhang

Publications and source records attributed to Yuhao Zhang.

At least 73 records · Page 4Linked to original sources

arXiv2026

Breaking Semantic Hegemony: Decoupling Principal and Residual Subspaces for Generalized OOD Detection

While feature-based post-hoc methods have made significant strides in Out-of-Distribution (OOD) detection, we uncover a counter-intuitive Simplicity Paradox in existing state-of-the-art (SOTA) models: these models exhibit keen sensitivity in distinguishing semantically subtle OOD samples but suffer from severe Geometric Blindness when confronting structurally distinct yet semantically simple samples or high-frequency sensor noise. We attribute this phenomenon to Semantic Hegemony within the deep feature space and reveal its mathematical essence through the lens of Neural Collapse. Theoretical analysis demonstrates that the spectral concentration bias, induced by the high variance of the principal subspace, numerically masks the structural distribution shift signals that should be significant in the residual subspace. To address this issue, we propose D-KNN, a training-free, plug-and-play geometric decoupling framework. This method utilizes orthogonal decomposition to explicitly separate semantic components from structural residuals and introduces a dual-space calibration mechanism to reactivate the model's sensitivity to weak residual signals. Extensive experiments demonstrate that D-KNN effectively breaks Semantic Hegemony, establishing new SOTA performance on both CIFAR and ImageNet benchmarks. Notably, in resolving the Simplicity Paradox, it reduces the FPR95 from 31.3% to 2.3%; when addressing sensor failures such as Gaussian noise, it boosts the detection performance (AUROC) from a baseline of 79.7% to 94.9%.

cs.CV↗

arXiv2026

VMF-GOS: Geometry-guided virtual Outlier Synthesis for Long-Tailed OOD Detection

Out-of-Distribution (OOD) detection under long-tailed distributions is a highly challenging task because the scarcity of samples in tail classes leads to blurred decision boundaries in the feature space. Current state-of-the-art (sota) methods typically employ Outlier Exposure (OE) strategies, relying on large-scale real external datasets (such as 80 Million Tiny Images) to regularize the feature space. However, this dependence on external data often becomes infeasible in practical deployment due to high data acquisition costs and privacy sensitivity. To this end, we propose a novel data-free framework aimed at completely eliminating reliance on external datasets while maintaining superior detection performance. We introduce a Geometry-guided virtual Outlier Synthesis (GOS) strategy that models statistical properties using the von Mises-Fisher (vMF) distribution on a hypersphere. Specifically, we locate a low-likelihood annulus in the feature space and perform directional sampling of virtual outliers in this region. Simultaneously, we introduce a new Dual-Granularity Semantic Loss (DGS) that utilizes contrastive learning to maximize the distinction between in-distribution (ID) features and these synthesized boundary outliers. Extensive experiments on benchmarks such as CIFAR-LT demonstrate that our method outperforms sota approaches that utilize external real images.

cs.CV↗

arXiv2026

SN 2024igg: A Super-Chandrasekhar/03fg-like SN exhibiting C II-dominated spectra after explosion

We present and analyze photometric and spectroscopic observations of the Type Ia supernova (SN Ia) 2024igg, another ``super-Chandrasekhar'' (or 03fg-like) SN whose strong C II $\lambda6580$ feature was initially misidentified as H$α$, thereby constraining its progenitor system, explosion parameters, and physical scenario. SN 2024igg shows many characteristics in common with other 03fg-like objects, such as high ultraviolet flux, slowly declining light curves ($Δm_{15}(B)=0.90\pm0.08$ mag), low expansion velocities, along with strong and persistent C II absorption. Meanwhile, this SN exhibits some remarkable properties within this subgroup, including a moderately low optical luminosity ($M_{\rm max}(B)=-18.99\pm0.15$ mag), a short rise time less than 18.5 days, and strong C II $\lambda6580$. The bolometric analysis yields a $^{56}$Ni mass of $M_{\rm Ni}=0.547\pm0.082$ $M_{\rm \odot}$ and an ejecta mass of $1.54^{+0.22}_{-0.19}$ $M_{\rm \odot}$, marginally exceeding the Chandrasekhar mass. Our TARDIS result indicates that most of the features in the earliest spectrum could be attributed to C II, which is consistent with a model where a supernova explodes within a carbon-rich circumstellar medium (CSM). The CSM interaction would produce a density peak in the ejecta, offering a natural explanation for the slowly evolving line velocities near $-$8000 km s$^{-1}$. The CSM may stem from the debris of a secondary white dwarf in a white-dwarf merger or the envelope of an asymptotic giant branch star. Combined with the unshifted forbidden lines in the spectrum taken at $t\approx\ +$135 days, we suggest that SN 2024igg comes from a symmetric explosion on a secular timescale after the merger.

astro-ph.HE↗

arXiv2026

Competitive Non-Clairvoyant KV-Cache Scheduling for LLM Inference

Large Language Model (LLM) inference presents a unique scheduling challenge due to the Key-Value (KV) cache, where a job's memory footprint grows linearly with the number of decoded tokens. This growth couples scheduling decisions with feasibility: a scheduler must minimize latency under a hard memory budget, yet the response lengths of requests are inherently unknown. While recent works have explored this problem either assuming clairvoyance -- exact knowledge of response lengths -- or relying on machine-learned predictions, obtaining robust performance guarantees without any prior knowledge of job sizes remains a theoretically fundamental and practically important open problem. In this work, we propose the Geometric Slicing Algorithm (GSA), the non-clairvoyant policy to achieve the first constant competitive ratio for this problem in the offline batch setting. GSA manages uncertainty through a geometric phase structure that periodically restarts jobs to bound memory exposure, combined with a staggered pipeline mechanism that enables high concurrency by smoothing aggregate memory consumption. We prove that GSA achieves a competitive ratio of at most 61.92 for general instances, improving to 32 in the large-memory regime. Our algorithmic framework also yields a clairvoyant counterpart, the Geometric Batching Algorithm (GBA), which achieves an approximation ratio of 10.67 for general instances and 6.75 in the large-memory regime -- significantly improving upon the best previously known bound of over 9000. Numerical experiments on real request traces demonstrate that our algorithms perform robustly while preserving these worst-case guarantees.

cs.DS↗

arXiv2026

Robust Uncertainty Quantification for Factual Generation of Large Language Models

The rapid advancement of large language model(LLM) technology has facilitated its integration into various domains of professional and daily life. However, the persistent challenge of LLM hallucination has emerged as a critical limitation, significantly compromising the reliability and trustworthiness of AI-generated content. This challenge has garnered significant attention within the scientific community, prompting extensive research efforts in hallucination detection and mitigation strategies. Current methodological frameworks reveal a critical limitation: traditional uncertainty quantification approaches demonstrate effectiveness primarily within conventional question-answering paradigms, yet exhibit notable deficiencies when confronted with non-canonical or adversarial questioning strategies. This performance gap raises substantial concerns regarding the dependability of LLM responses in real-world applications requiring robust critical thinking capabilities. This study aims to fill this gap by proposing an uncertainty quantification scenario in the task of generating with multiple facts. We have meticulously constructed a set of trap questions contained with fake names. Based on this scenario, we innovatively propose a novel and robust uncertainty quantification method(RU). A series of experiments have been conducted to verify its effectiveness. The results show that the constructed set of trap questions performs excellently. Moreover, when compared with the baseline methods on four different models, our proposed method has demonstrated great performance, with an average increase of 0.1-0.2 in ROCAUC values compared to the best performing baseline method, providing new sights and methods for addressing the hallucination issue of LLMs.

cs.CL↗

arXiv2025

Sub-Ensemble Correlations as a Covariance Geometry

Conventional practice of spatially resolved detection in diffusion-coupled thermal atomic vapors implicitly treat localized responses as mutually independent. However, in this study, it is shown that observable correlations are governed by the intrinsic spatiotemporal covariance of a global spin-fluctuation field, such that spatial separation specifies only overlapping statistical projections rather than independent physical components. A unified field-theoretic description is established in which sub-ensembles are defined as measurement-induced statistical projections of a single stochastic field. Within this formulation, sub-ensemble correlations are determined by the covariance operator, inducing a natural geometry in which statistical independence corresponds to orthogonality of the measurement functionals. For collective spin fluctuations described by a diffusion-relaxation Ornstein-Uhlenbeck stochastic field, the covariance spectrum admits only a finite set of fluctuation modes in a bounded domain, imposing an intrinsic, field-level limit on the number of statistically distinguishable sub-ensembles. The loss of sub-ensemble independence is formalized through the notion of spatial sampling overlap, which quantifies the unavoidable statistical coupling arising from shared access to common low-order fluctuation modes. While multi-channel atomic magnetometry provides a concrete physical setting in which these constraints become explicit, the framework applies generically to diffusion-coupled stochastic fields.

physics.atom-ph↗

arXiv2025

MM-ACT: Learn from Multimodal Parallel Generation to Act

A generalist robotic policy needs both semantic understanding for task planning and the ability to interact with the environment through predictive capabilities. To tackle this, we present MM-ACT, a unified Vision-Language-Action (VLA) model that integrates text, image, and action in shared token space and performs generation across all three modalities. MM-ACT adopts a re-mask parallel decoding strategy for text and image generation, and employs a one-step parallel decoding strategy for action generation to improve efficiency. We introduce Context-Shared Multimodal Learning, a unified training paradigm that supervises generation in all three modalities from a shared context, enhancing action generation through cross-modal learning. Experiments were conducted on the LIBERO simulation and Franka real-robot setups as well as RoboTwin2.0 to assess in-domain and out-of-domain performances respectively. Our approach achieves a success rate of 96.3% on LIBERO, 72.0% across three tasks of real Franka, and 52.38% across eight bimanual tasks of RoboTwin2.0 with an additional gain of 9.25% from cross-modal learning. We release our codes, models and data at https://github.com/HHYHRHY/MM-ACT.

cs.CV↗

arXiv2025

Ga$_2$O$_3$ TCAD Mobility Parameter Calibration using Simulation Augmented Machine Learning with Physics Informed Neural Network

In this paper, we demonstrate the feasibility of performing automatic Technology Computer Aided Design (TCAD) parameter calibration and extraction using machine learning, with the machine trained solely by TCAD simulation data. The methodology is validated using experimental data. Schottky Barrier Diodes (SBDs) with different effective anode workfunction (WF) are fabricated with emerging ultra-wide bandgap material, Gallium Oxide (Ga2O3), and are measured at various temperatures (T). Their current voltage curves are used for automatic Ga2O3 Philips Unified Mobility (PhuMob) model parameters calibration. Five critical PhuMob parameters were calibrated. The machine consists of an autoencoder and a neural network and is trained solely by TCAD simulation data with variations in WF, T, and the five PhuMob parameters (seven variations in total). Then, Ga2O3 PhuMob parameters are extracted from the noisy experimental curves. Subsequent TCAD simulation using the extracted parameters shows that the quality of the parameters is as good as an expert's calibration at the pre-turned on regime, but not in the on state regime. By using a simple physics-informed neural network, the machine performs as well as the human expert in all regimes.

cs.LG↗

arXiv2025

LLM$^3$-DTI: A Large Language Model and Multi-modal data co-powered framework for Drug-Target Interaction prediction

Drug-target interaction (DTI) prediction is of great significance for drug discovery and drug repurposing. With the accumulation of a large volume of valuable data, data-driven methods have been increasingly harnessed to predict DTIs, reducing costs across various dimensions. Therefore, this paper proposes a $\textbf{L}$arge $\textbf{L}$anguage $\textbf{M}$odel and $\textbf{M}$ulti-$\textbf{M}$odel data co-powered $\textbf{D}$rug $\textbf{T}$arget $\textbf{I}$nteraction prediction framework, named LLM$^3$-DTI. LLM$^3$-DTI constructs multi-modal data embedding to enhance DTI prediction performance. In this framework, the text semantic embeddings of drugs and targets are encoded by a domain-specific LLM. To effectively align and fuse multi-modal embedding. We propose the dual cross-attention mechanism and the TSFusion module. Finally, these multi-modal data are utilized for the DTI task through an output network. The experimental results indicate that LLM$^3$-DTI can proficiently identify validated DTIs, surpassing the performance of the models employed for comparison across diverse scenarios. Consequently, LLM$^3$-DTI is adept at fulfilling the task of DTI prediction with excellence. The data and code are available at https://github.com/chaser-gua/LLM3DTI.

cs.LG↗

arXiv2025

UNILocPro: Unified Localization Integrating Model-Based Geometry and Channel Charting

In this paper, we propose a unified localization framework (called UNILocPro) that integrates model-based localization and channel charting (CC) for mixed line-of-sight (LoS)/non-line-of-sight (NLoS) scenarios. Specifically, based on LoS/NLoS identification, an adaptive activation between the model-based and CC-based methods is conducted. Aiming for unsupervised learning, information obtained from the model-based method is utilized to train the CC model, where a pairwise distance loss (involving a new dissimilarity metric design), a triplet loss (if timestamps are available), a LoS-based loss, and an optimal transport (OT)-based loss are jointly employed such that the global geometry can be well preserved. To reduce the training complexity of UNILocPro, we propose a low-complexity implementation (called UNILoc), where the CC model is trained with self-generated labels produced by a single pre-training OT transformation, which avoids iterative Sinkhorn updates involved in the OT-based loss computation. Extensive numerical experiments demonstrate that the proposed unified frameworks achieve significantly improved positioning accuracy compared to both model-based and CC-based methods. Notably, UNILocPro with timestamps attains performance on par with fully-supervised fingerprinting despite operating without labelled training data. It is also shown that the low-complexity UNILoc can substantially reduce training complexity with only marginal performance degradation.

eess.SP↗

arXiv2025

GeoFM: Enhancing Geometric Reasoning of MLLMs via Synthetic Data Generation through Formal Language

Multi-modal Large Language Models (MLLMs) have gained significant attention in both academia and industry for their capabilities in handling multi-modal tasks. However, these models face challenges in mathematical geometric reasoning due to the scarcity of high-quality geometric data. To address this issue, synthetic geometric data has become an essential strategy. Current methods for generating synthetic geometric data involve rephrasing or expanding existing problems and utilizing predefined rules and templates to create geometric images and problems. However, these approaches often produce data that lacks diversity or is prone to noise. Additionally, the geometric images synthesized by existing methods tend to exhibit limited variation and deviate significantly from authentic geometric diagrams. To overcome these limitations, we propose GeoFM, a novel method for synthesizing geometric data. GeoFM uses formal languages to explore combinations of conditions within metric space, generating high-fidelity geometric problems that differ from the originals while ensuring correctness through a symbolic engine. Experimental results show that our synthetic data significantly outperforms existing methods. The model trained with our data surpass the proprietary GPT-4o model by 18.7\% on geometry problem-solving tasks in MathVista and by 16.5\% on GeoQA. Additionally, it exceeds the performance of a leading open-source model by 5.7\% on MathVista and by 2.7\% on GeoQA.

cs.AI↗

arXiv2025

Ferrohydrodynamic Microfluidics for Bioparticle Separation and Single-Cell Phenotyping: Principles, Applications, and Emerging Directions

Ferrohydrodynamic microfluidics relies on magnetic field gradients to manipulate diamagnetic particles in ferrofluid-filled microenvironments. It has emerged as a promising tool for label-free manipulation of bioparticles, including their separation and phenotyping. This perspective reviews recent progress in the development and applications of ferrofluid-based microfluidic platforms for multiscale bioparticle separation, ranging from micron-scale cells to submicron extracellular vesicles. We highlight the fundamental physical principles for ferrohydrodynamic manipulation, including the dominant magnetic buoyancy force resulting from the interaction of ferrofluids and particles. We then describe how these principles enable high-resolution size-based bioparticle separation, subcellular bioparticle enrichment, and phenotypic screening based on physical traits. We also discuss key challenges in ferrohydrodynamic microfluidics from the aspects of ferrofluid biocompatibility, system throughput, and nanoparticle depletion. Finally, we outline future research directions involving machine learning, 3D printing, and multiplexed detection. These insights chart a path for advancing ferrofluid-based technologies in precision biomedicine, diagnostics, and cellular engineering.

eess.SY↗

arXiv2025

A Multi-dimensional Semantic Surprise Framework Based on Low-Entropy Semantic Manifolds for Fine-Grained Out-of-Distribution Detection

Out-of-Distribution (OOD) detection is a cornerstone for the safe deployment of AI systems in the open world. However, existing methods treat OOD detection as a binary classification problem, a cognitive flattening that fails to distinguish between semantically close (Near-OOD) and distant (Far-OOD) unknown risks. This limitation poses a significant safety bottleneck in applications requiring fine-grained risk stratification. To address this, we propose a paradigm shift from a conventional probabilistic view to a principled information-theoretic framework. We formalize the core task as quantifying the Semantic Surprise of a new sample and introduce a novel ternary classification challenge: In-Distribution (ID) vs. Near-OOD vs. Far-OOD. The theoretical foundation of our work is the concept of Low-Entropy Semantic Manifolds, which are explicitly structured to reflect the data's intrinsic semantic hierarchy. To construct these manifolds, we design a Hierarchical Prototypical Network. We then introduce the Semantic Surprise Vector (SSV), a universal probe that decomposes a sample's total surprise into three complementary and interpretable dimensions: conformity, novelty, and ambiguity. To evaluate performance on this new task, we propose the Normalized Semantic Risk (nSR), a cost-sensitive metric. Experiments demonstrate that our framework not only establishes a new state-of-the-art (sota) on the challenging ternary task, but its robust representations also achieve top results on conventional binary benchmarks, reducing the False Positive Rate by over 60% on datasets like LSUN.

stat.ML↗

arXiv2025

Online MMS Allocation for Chores

We study the problem of fair division of indivisible chores among $n$ agents in an online setting, where items arrive sequentially and must be allocated irrevocably upon arrival. The goal is to produce an $α$-MMS allocation at the end. Several recent works have investigated this model, but have only succeeded in obtaining non-trivial algorithms under restrictive assumptions, such as the two-agent bi-valued special case (Wang and Wei, 2025), or by assuming knowledge of the total disutility of each agent (Zhou, Bai, and Wu, 2023). For the general case, the trivial $n$-MMS guarantee remains the best known, while the strongest lower bound is still only $2$. We close this gap on the negative side by proving that for any fixed $n$ and $\varepsilon$, no algorithm can guarantee an $(n - \varepsilon)$-MMS allocation. Notably, this lower bound holds precisely for every $n$, without hiding constants in big-$O$ notation, thereby exactly matching the trivial upper bound. Despite this strong impossibility result, we also present positive results. We provide an online algorithm that applies in the general case, guaranteeing a $\min\{n, O(k), O(\log D)\}$-MMS allocation, where $k$ is the maximum number of distinct disutilities across all agents and $D$ is the maximum ratio between the largest and smallest disutilities for any agent. This bound is reasonable across a broad range of scenarios and, for example, implies that we can achieve an $O(1)$-MMS allocation whenever $k$ is constant. Moreover, to optimize the constant in the important personalized bi-valued case, we show that if each agent has at most two distinct disutilities, our algorithm guarantees a $(2 + \sqrt{3}) \approx 3.7$-MMS allocation.

cs.GT↗

arXiv2025

SPFT-SQL: Enhancing Large Language Model for Text-to-SQL Parsing by Self-Play Fine-Tuning

Despite the significant advancements of self-play fine-tuning (SPIN), which can transform a weak large language model (LLM) into a strong one through competitive interactions between models of varying capabilities, it still faces challenges in the Text-to-SQL task. SPIN does not generate new information, and the large number of correct SQL queries produced by the opponent model during self-play reduces the main model's ability to generate accurate SQL queries. To address this challenge, we propose a new self-play fine-tuning method tailored for the Text-to-SQL task, called SPFT-SQL. Prior to self-play, we introduce a verification-based iterative fine-tuning approach, which synthesizes high-quality fine-tuning data iteratively based on the database schema and validation feedback to enhance model performance, while building a model base with varying capabilities. During the self-play fine-tuning phase, we propose an error-driven loss method that incentivizes incorrect outputs from the opponent model, enabling the main model to distinguish between correct SQL and erroneous SQL generated by the opponent model, thereby improving its ability to generate correct SQL. Extensive experiments and in-depth analyses on six open-source LLMs and five widely used benchmarks demonstrate that our approach outperforms existing state-of-the-art (SOTA) methods.

cs.CL↗

arXiv2025

Online Makespan Minimization: Beat LPT by Dynamic Locking

Online makespan minimization is a fundamental problem in scheduling. In this paper, we investigate its over-time formulation, where each job has a release time and a processing time. A job becomes known only at its release time and must be scheduled on a machine thereafter. The Longest Processing Time First (LPT) algorithm, established by Chen and Vestjens (1997), achieves a competitive ratio of $1.5$. For the special case of two machines, Noga and Seiden introduced the SLEEPY algorithm, which achieves a tight competitive ratio of $1.382$. However, for $m \geq 3$, no known algorithm has convincingly surpassed the long-standing $1.5$ barrier. We propose a natural generalization of SLEEPY and show this simple approach can beat the $1.5$ barrier and achieve $1.482$-competitive when $m=3$. However, when $m$ becomes large, we prove this simple generalization fails to beat $1.5$. Meanwhile, we introduce a novel technique called dynamic locking to overcome this new challenge. As a result, we achieve a competitive ratio of $1.5-\frac{1}{O(m^2)}$, which beats the LPT algorithm ($1.5$-competitive) for every constant $m$.

cs.DS↗

arXiv2025

CSDformer: A Conversion Method for Fully Spike-Driven Transformer

Spike-based transformer is a novel architecture aiming to enhance the performance of spiking neural networks while mitigating the energy overhead inherent to transformers. However, methods for generating these models suffer from critical limitations: excessive training costs introduced by direct training methods, or unavoidably hardware-unfriendly operations in existing conversion methods. In this paper, we propose CSDformer, a novel conversion method for fully spike-driven transformers. We tailor a conversion-oriented transformer-based architecture and propose a new function NReLU to replace softmax in self-attention. Subsequently, this model is quantized and trained, and converted into a fully spike-driven model with temporal decomposition technique. Also, we propose delayed Integrate-andFire neurons to reduce conversion errors and improve the performance of spiking models. We evaluate CSDformer on ImageNet, CIFAR-10 and CIFAR-100 datasets and achieve 76.36% top-1 accuracy under 7 time-steps on ImageNet, demonstrating superiority over state-of-the-art models. Furthermore, CSDformer eliminates the need for training SNNs, thereby reducing training costs (reducing computational resource by 75% and accelerating training speed by 2-3$\times$). To the best of our knowledge, this is the first fully spike-driven transformer-based model developed via conversion method, achieving high performance under ultra-low latency, while dramatically reducing both computational complexity and training overhead.

cs.CV↗

arXiv2025

Enhancing Speech Large Language Models with Prompt-Aware Mixture of Audio Encoders

Connecting audio encoders with large language models (LLMs) allows the LLM to perform various audio understanding tasks, such as automatic speech recognition (ASR) and audio captioning (AC). Most research focuses on training an adapter layer to generate a unified audio feature for the LLM. However, different tasks may require distinct features that emphasize either semantic or acoustic aspects, making task-specific audio features more desirable. In this paper, we propose Prompt-aware Mixture (PaM) to enhance the Speech LLM that uses multiple audio encoders. Our approach involves using different experts to extract different features based on the prompt that indicates different tasks. Experiments demonstrate that with PaM, only one Speech LLM surpasses the best performances achieved by all single-encoder Speech LLMs on ASR, Speaker Number Verification, and AC tasks. PaM also outperforms other feature fusion baselines, such as concatenation and averaging. Our code would be available at: https://github.com/shanweiqiao/PaM

eess.AS↗