Searcharxiv⌕ Search

arXiv subjects

Yuheng Zhang

Publications and source records attributed to Yuheng Zhang.

At least 19 recordsLinked to original sources

Must You See Fakes to Catch Fakes? Language Says What to Look For, Real Images What to Ignore

Must a detector see generated images in order to catch them? Detection of AI-generated images has long faced two problems. The cues a detector learns from the few generators in its training set carry traces specific to those generators and transfer poorly to new ones, which limits generalization. The real and generated training images also differ in format and content, so the detector takes these differences as evidence of authenticity, and its decisions change with factors that have nothing to do with authenticity, which harms robustness as well as generalization. Both problems stem from training on generated images. A detector never trained on them can largely avoid both at the source. Vision foundation models have already learned the concept of "AI-generated" during pretraining, and text can point to it directly. Building on this, we propose SaySee, a training-free detector: Say builds a generation direction from sets of paired descriptions that differ only in authenticity and scores an image by the projection of its embedding onto it, and See uses real images to correct the score offsets caused by format, image type and content. On 24 evaluation sets that span standard benchmarks, in-the-wild data and the latest generators, SaySee reaches a mean AUC of 98.2% with MoE-ViE, 4.2 points above the best existing detector. When format is decoupled from authenticity, it remains the best of all methods. Code will be released.

cs.CV↗

VoxSAOR: Semantic-Aware Occupancy Robustification via Aleatoric Uncertainty Modeling

Vision-based 3D occupancy prediction provides a dense semantic representation of scenes for autonomous driving and embodied perception, yet remains vulnerable to occlusion, depth ambiguity, and projection errors that induce spatially varying reliability in voxel features. Existing methods typically map voxel representations to semantic predictions without explicitly modeling such observation-induced feature variations. To address this limitation, we propose VoxSAOR, a semantic-aware occupancy robustification framework that models aleatoric uncertainty through feature-space proxies. It first quantifies Local Semantic Consistency (LSC) to derive a voxel-wise structural stability prior and uses cross-view semantic consensus to filter unreliable semantic correspondences. Within each semantic category, spatially matched stable and ambiguous voxels form stable-to-ambiguous feature transitions, which are modeled by a class-conditional diagonal Gaussian distribution, capturing both feature displacement and dispersion. Guided by these proxy distributions, Aleatoric Regime Ensemble Learning (AREL) trains independently parameterized prediction heads using feature perturbations at progressively increasing scales, with perturbation strength modulated by structural reliability. At inference, proxy modeling and sampling are disabled, and all heads process the same observed features; their output probabilities are then averaged to produce a deterministic consensus prediction without requiring sampling. Experiments across multiple datasets and occupancy backbones demonstrate the effectiveness of VoxSAOR in improving semantic occupancy prediction, long-tail recognition, and predictive calibration. The source code will be made publicly available at https://github.com/MengfeiD/VoxSAOR.

cs.CV↗

OpticalRAG: Pixel-Space Compression for Token-Efficient Retrieval-Augmented Generation

Retrieval-augmented generation (RAG) is bottlenecked by the token budget of large language models: feeding long documents into the context window is expensive, while aggressive textual compression discards fine-grained information that is irreversibly lost in a one-dimensional token sequence. We argue that this bottleneck is not fundamental to information density but to the choice of compression \emph{modality}. Modern visual encoders can pack the contents of a rendered text page into roughly a hundred visual tokens with little semantic loss, opening a new compression-fidelity tradeoff that purely textual methods cannot reach. We present \textbf{OpticalRAG}, a framework that renders documents as page images, encodes them with a frozen vision encoder, and distills the retrieved visual tokens into as few as $16$ injected tokens before passing them to the LLM. Two ingredients make this work without expensive cross-modal training: (i) \emph{Encoder-Bridged Retrieval}, which projects text-query states into the visual feature space of the frozen encoder, through a projection shared with the compressor, to retrieve visual tokens from a text query, with no contrastive training; and (ii) \emph{Query-Driven Distillation}, a lightweight cross-attention compressor that condenses the retrieved visual tokens into a small set of query-conditioned prefix tokens. On LongBench-v2, BAMBOO, and LooGLE-v2, OpticalRAG achieves the best compression-aware accuracy (CAP) on all three benchmarks while injecting only $16$ tokens, and improves the closed-book Qwen2.5-7B-Instruct backbone by $5.17$ points on LongBench-v2. Code is available at https://github.com/brighterr/OpticalRAG.

cs.CL↗

Beyond State-Wise Mirror Descent: Offline Policy Optimization with Parametric Policies

We investigate the theoretical aspects of offline reinforcement learning (RL) under general function approximation. While prior works (e.g., Xie et al., 2021) have established the theoretical foundations of learning a good policy from offline data via pessimism, existing algorithms that are computationally tractable (often in an oracle-efficient sense), such as PSPI, only apply to finite and small action spaces. Moreover, these algorithms rely on state-wise mirror descent and require actors to be implicitly induced from the critic functions, failing to accommodate standalone policy parameterization which is ubiquitous in practice. In this work, we address these limitations and extend the theoretical guarantees to parameterized policy classes over large or continuous action spaces. When extending mirror descent to parameterized policies, we identify contextual coupling as the core difficulty, and show how connecting mirror descent to natural policy gradient leads to novel analyses, guarantees, and algorithmic insights, including a surprising unification between offline RL and imitation learning.

cs.LG↗

Probing faint radio activity from cores to filaments in two $z=2.24$ protoclusters with uGMRT

We present deep 1.26 GHz radio continuum observations of the two $z=2.24$ protoclusters BOSS1244 and BOSS1542 with the upgraded Giant Metrewave Radio Telescope (uGMRT) to investigate their members' radio emission and its possible environmental dependence. Among 187 and 105 H$α$ emitter (HAE) members in BOSS1244 and BOSS1542, we identify ten and two robust radio counterparts, respectively, together with five potential counterparts from forced photometry. Two HAEs and the massive quiescent galaxy QG1 in BOSS1244 show strong radio excesses relative to their H$α$-based star formation rates (SFRs) and are identified as strong radio-excess AGN candidates. Median stacking of the remaining 173 and 102 radio-undetected HAE members yields positive signals at $2.8σ$ and $3.2σ$ in BOSS1244 and BOSS1542, respectively. The stacked radio luminosities are consistent with those expected from the median H$α$-based SFRs, indicating that star formation can account for the average faint radio emission. After excluding the robust radio counterparts, we divided the HAE members by the projected distance from QG1 in BOSS1244 and from the filament spine in BOSS1542. The largest stacked radio luminosities are found toward the denser regions in both systems, but the differences are not statistically significant, and stellar mass shows no significant dependence on environmental distance. These results identify strong radio-excess AGN candidates among the individually detected members, while the faint HAE population is on average consistent with radio emission from star formation. The similar environmental ordering is suggestive but remains tentative. Deeper radio observations at multiple frequencies and higher angular resolution are required to determine whether this apparent trend reflects weak AGN activity or variations in radio emission associated with star formation.

astro-ph.GA↗

Improving Sampling Efficiency in RLVR through Adaptive Rollout and Response Reuse

Group relative policy optimization (GRPO) trains large language models with verifiable rewards by normalizing rewards within response groups. When every response in a group receives the same reward, the advantages vanish and the group provides no policy gradient signal. We propose Adaptive Rollout and Response Reuse Policy Optimization (AR3PO) to obtain informative groups with fewer generations. Adaptive rollout samples responses to each prompt in stages and stops sampling once the model generates a correct response. If all sampled responses are incorrect, response reuse replaces one of them with a past correct response to the same prompt, when available, and recomputes the group advantages. Policy-recomputed reuse trains on the updated groups using probabilities recomputed under the rollout policy; advantage-only reuse uses historical responses only to compute advantages. Across four mathematical benchmarks with 7B and 8B models, AR3PO consistently improves over GRPO and achieves performance comparable to DAPO while requiring up to $4.2\times$ fewer generated responses. Experiments with a 32B model also show comparable performance to DAPO at a similar training stage with substantially fewer generated responses.

cs.LG↗

Fooling Reward Models with Gibberish: Adversarial Examples Beyond the Semantic Regime

Reward models (RMs) are widely used as optimization targets in reinforcement learning from human feedback (RLHF). We show that leading RMs can assign higher scores to gibberish than to meaningful reference answers. To systematically discover such inputs, we introduce Token Mapping Perturbation Attack (TOMPA), a framework that uses a perturbed token mapping to extend an attack LLM's search beyond natural language. The mapping changes the linguistic meaning of the attack LLM's natural language responses, producing non-linguistic text for the target RM to score. We then use these scalar scores as rewards to train the attack LLM through reinforcement learning to discover non-linguistic adversarial inputs. Experiments against Llama and Qwen RMs demonstrate such attacks using two instantiations: identity token mapping across different tokenizers and random permutation with a shared tokenizer. Against Skywork-Reward-V2-Llama-3.1-8B, ranked first on RewardBench 2, the two mappings achieve mean rewards of +33.64 and +43.78, respectively, compared with +17.48 for GPT-5 reference answers. Their attack success rates are 97.6% and 100%, measured as the percentages of individual responses scoring above their corresponding GPT-5 references. These results show that strong benchmark performance does not ensure robustness to non-linguistic adversarial inputs, exposing a fundamental vulnerability in using RMs as optimization targets.

cs.LG↗

Turing LLM Post-Training: Calibrated Distillation and Efficient Dynamic-MoE Execution

We study post-training for dynamically routed mixture-of-experts (MoE) models, focusing on integrating independently optimized specialists while keeping consolidation efficient. Our approach combines calibrated multi-teacher on-policy distillation (C-MOPD), which matches each selected teacher on a local Top-$K$ support with teacher-wise loss normalization and clipping, with an execution path that uses post-capacity bounded-width dispatch, final-route replay, and asynchronous rollout--training overlap. On Turing-20B-A2B, the selected model achieves a four-domain normalized integration score of 0.9988 and retains at least 90.16\% of the measured specialist gain in every domain. At matched $K=32$, the full calibration objective scores 0.9962, compared with 0.9724 without calibration and 0.9834 without clipping. At a matched 8K response cap, the model exceeds both shared SFT and Qwen3-8B on all four reasoning-enabled mathematics and code benchmarks (unweighted mean: 85.92, versus 80.33 for Qwen3-8B and 74.21 for SFT), and improves over SFT on ten of twelve broader evaluation metrics. In a PPU-based profile, the combined execution configuration reduces consolidation step time from 398.7 to 139.3 seconds, an observed 2.86$\times$ speedup over serial execution. These results demonstrate strong specialist integration alongside more efficient consolidation, while reflecting the complete training and execution framework rather than any single component.

cs.LG↗

Iterative Calibration and Learning: Mitigating Teacher Bias in LLM Distillation without Target-Domain Reward Feedback

Large language model (LLM) distillation can propagate teacher bias under covariate shift when target reward feedback is unavailable. We propose Iterative Calibration and Learning (IterCaL), which alternates source-reward teacher calibration through token-level branching with student learning on target prompts. We prove polynomial convergence in expected average KL to the oracle student, the reference-regularized target-reward maximizer within a potentially misspecified student class; a quantified source-to-target calibration discrepancy gives an explicit additive true-return term. Our separation shows that regularized direct teacher matching retains positive oracle-KL error at its population optimum for every shared reference penalty, even when the teacher is stronger than every student. On the same instance, IterCaL converges to the oracle using source rewards alone.

cs.LG↗

Near-Optimal First-Order Regret for Bandits with Feedback Graphs

We study first-order regret guarantees for adversarial bandits with a fixed, known directed feedback graph. For strongly observable graphs, [ACBDK15] establishes a minimax regret of $\widetildeΘ(\sqrt{αT})$, where $α$ denotes the independence number of the graph and $T$ denotes the horizon. On the other hand, the best existing first-order bound scales as $\widetilde{O}(\sqrt{(κ+1) L^\star})$ [LLZ20], where $κ$ denotes the clique partition number, satisfying $κ+1\geqα$, and $L^\star$ denotes the cumulative loss of the best arm, leaving a gap between the graph dependence in worst-case and first-order guarantees. In this paper, we bridge this gap by obtaining an expected regret bound of $\widetilde{O}(\sqrt{αL^\star}+α)$. Our first result is an algorithm for the special case of self-aware graphs, in which every vertex has a self-loop. The algorithm combines the update of EXP3 [ACBFS02] with a sampling rule inspired by exploration by optimization [LS20]. More specifically, we construct a map whose fixed points yield sampling distributions that control the one-round regret remainder in terms of the learner's loss and the graph's independence number, which is crucial in deriving optimal first-order guarantees. We further show that a sufficiently accurate fixed point can be computed in polynomial time by a geometrically convergent iteration. Finally, we extend the algorithm to general strongly observable graphs and achieve the same $\widetilde{O}(\sqrt{αL^\star}+α)$ guarantee by adapting the two-level aggregation approach of [LLZ20].

cs.LG↗

PI-MLLM: Psychophysics-Inspired Multimodal Large Language Model for Oddly Satisfying Video Emotion Analysis

Oddly Satisfying Videos (OSVs) can evoke strong viewer responses through repetitive motion, visual tactility, and ASMR-like sounds despite containing few explicit semantic emotion cues. This makes OSVs a challenging setting for Video Emotion Analysis (VEA), since most existing methods focus on semantic video content and under-model the physical sensory dynamics that shape viewer responses. To support Oddly Satisfying Video Emotion Analysis (OSVEA), we construct the Oddly Satisfying Video Dataset (OSVD), comprising 12,000 annotated 5-second clips. Each clip provides viewer-emotion labels under video-only, audio-only, and audio-visual conditions, together with explanations grounded in visual and acoustic sensory evidence. Building on OSVD, we propose PI-MLLM, a psychophysics-inspired multimodal large language model that integrates Modality-Specific Dynamics Alignment (MSDA) and Coalition-Based Audio-Visual Attribution (CAVA). MSDA uses auxiliary prediction objectives to align modality representations with visual motion and acoustic energy-onset dynamics. CAVA evaluates emotion scores across modality coalitions and applies Shapley attribution to identify the dominant modality and determine whether visual and auditory evidence is synergistic or conflicting. Experiments show that PI-MLLM improves both viewer-side emotion prediction and sensory-grounded explanation quality, suggesting the value of physical sensory cues to video emotion analysis.

cs.CV↗

Panoramic Out-of-Distribution Segmentation

Panoramic imaging enables capturing 360° images with an ultra-wide Field-of-View (FoV) for dense omnidirectional perception, which is critical to applications such as autonomous driving and augmented reality. However, current panoramic semantic segmentation methods fail to identify outliers, and pinhole out-of-distribution segmentation (pinhole-OoS) models perform unsatisfactorily in the panoramic domain due to pixel distortions and background clutter. To address these issues, we introduce a new task, Panoramic Out-of-Distribution Segmentation (PanOoS), with the aim of achieving comprehensive and safe scene understanding. Furthermore, we propose the first solution, POS, which adapts to the characteristics of panoramic images through text-guided prompt distribution learning. Specifically, POS integrates a disentanglement strategy designed to materialize the cross-domain generalization capability of CLIP. The proposed Prompt-based Restoration Attention (PRA) optimizes semantic decoding by prompt guidance and self-adaptive correction, while Bilevel Prompt Distribution Learning (BPDL) refines the manifold of per-pixel mask embeddings via semantic prototype supervision. Besides, to compensate for the scarcity of PanOoS datasets, we establish two benchmarks: DenseOoS, which features diverse outliers in complex environments, and QuadOoS, captured by a quadruped robot with a panoramic annular lens system. Extensive experiments demonstrate the effectiveness of POS, with AuPRC increasing and FPR95 decreasing by 34.25 and 21.42 percentage points, respectively, compared with RbA on DenseOoS, outperforming state-of-the-art pinhole-OoS methods. Moreover, POS achieves leading closed-set segmentation capabilities and advances the development of panoramic understanding. Code and datasets will be available at https://github.com/MengfeiD/PanOoS.

cs.CV↗

Fast Last-Iterate Convergence in Zero-Sum Markov Games with Bandit Feedback

We study last-iterate convergence in unknown two-player zero-sum discounted Markov games with bandit feedback. The players learn independently along a single trajectory without observing each other's actions. We develop Adaptive Regularized TD Learning (ARTD), which achieves a $\widetilde{\mathcal{O}}(t^{-1/4})$ duality gap bound for the current policies under a uniform hitting time assumption, with high probability simultaneously over all rounds and starting states. This improves the $\widetilde{\mathcal{O}}(t^{-1/(9+ν)})$ rate of Cai et al. (2023), for any fixed $ν>0$, under the same feedback model and hitting time assumption. Our algorithm requires no knowledge of the hitting time bound, the time horizon, or the confidence level. To stabilize policy learning as value estimates change, we separate fast temporal difference averaging from bounded value updates. We adapt log-barrier regularization to the progress of value estimation, controlling both policy and value errors throughout learning. Together, these mechanisms enable fast convergence of the policies actually played, even when the players learn independently from bandit feedback.

cs.LG↗

EmbPASS: Towards Cross-Embodiment Open Panoramic Segmentation

Panoramic images provide a complete 360-degree field of view, enabling comprehensive scene understanding for embodied perception. However, heterogeneous embodied platforms exhibit substantial differences in observation viewpoints and spatial layouts, giving rise to cross-embodiment observation shifts that pose additional challenges to consistent and reliable panoramic perception, while systematic studies of this problem remain limited. To bridge this gap, we introduce a new task, termed Cross-Embodiment Open Panoramic Segmentation. Meanwhile, we establish EmbPASS, a multi-platform panoramic semantic segmentation benchmark spanning Vehicle, Drone, Wearable, and Quadruped platforms under a unified semantic taxonomy, providing a testbed for systematically studying cross-embodiment panoramic perception. We further propose EPONet, an open-vocabulary panoramic semantic segmentation network that integrates Relation-Aware Metric Adapter (RAMA) and Content-Adaptive Semantic Transfer (CAST) to enhance spatial modeling and semantic transfer under heterogeneous embodied observations. Extensive experiments show that EPONet achieves the best platform-balanced performance on EmbPASS with 35.82% mIoU, outperforming the strongest baseline by 1.10%, while remaining competitive on existing panoramic segmentation benchmarks. The source code and EmbPASS benchmark will be made publicly available at https://github.com/guopj1/EmbPASS.

cs.CV↗

Where Speech Enhancement Hurts Recognition: An Inference Time Polar Projection Diagnosis

Speech enhancement (SE) can substantially improve perceptual quality, yet enhanced speech does not necessarily improve automatic speech recognition (ASR). Common explanations such as enhancement artifacts and over-suppression remain qualitative and do not localize which enhancement component affects recognition. We study inference-time polar projection, which transforms an STFT mask $M=Ae^{jϕ}$ into $M_{α,γ}=A^αe^{jγϕ}$ and independently varies magnitude strength and estimated phase correction with frozen SE and ASR models. The resulting response curves separate effects that are coupled in ordinary enhanced-vs-noisy comparisons. On VoiceBank+DEMAND with FRCRN, wav2vec2.0 improves from 17.13 to 10.20 WER near full magnitude correction ($α=.85$), while Whisper reaches its best WER under much weaker correction, improving from 6.44 to 5.57 at $α=.25$ and returning to 6.46 at full magnitude. A second enhancer, a robustness-oriented ASR model, and a reverberant WSJ0 mixture set generalize the pattern and show that the preferred magnitude region depends on the recognizer and acoustic condition. Estimated phase correction is not a stable ASR lever in these evaluations: its effect is smaller, setting-dependent, and provides no consistent gain once magnitude strength is fixed.

eess.AS↗

Minimax Last-Iterate Convergence in Matrix Games with Observed Actions

We study last-iterate convergence in unknown two-player zero-sum matrix games with bandit payoff feedback and observed opponent actions. For games with $d$ actions per player, we develop an algorithm achieving a duality gap of $\widetilde{\mathcal{O}}(\sqrt{d/t})$ with high probability, simultaneously at every round $t$. This improves the dimension dependence of the best previously known guarantee by a factor of $d^{3/2}$. The rate matches a standard bandit lower bound, establishing minimax optimality in both the number of actions and the number of rounds, up to logarithmic factors. The algorithm is computationally efficient, requiring only $\mathcal{O}(d)$ time and memory per round. Our technical contribution is a joint design of adaptive averaging and corrected exponential weights that absorbs estimation variance, together with a potential argument that bounds phase durations.

cs.LG↗

Polylogarithmic Nash Regret in Matrix Games with Bandit Feedback

We study Nash regret minimization in unknown finite matrix games with bandit payoff feedback and observed opponent actions. We develop Optimistic Payoff Balancing (OPB), which achieves instance-dependent $\mathcal{O}(\log^2 T)$ Nash regret against arbitrary adaptive opponents, including games with nonunique equilibria. This resolves the open problem posed by Maiti et al. (2025), extending their polylogarithmic guarantee under bandit feedback from $2\times2$ games to arbitrary finite dimensions. To handle nonunique equilibria, we construct a reference strategy that leaves room for local adjustments. We order independent payoff differences by estimation accuracy and scale these adjustments by uncertainty, allowing the learner to exploit the opponent's imbalance to offset estimation costs. Our result thus shows that observing opponent actions suffices for polylogarithmic Nash regret in general finite matrix games.

cs.LG↗

The Physical Origins of Scatter in the Dust Attenuation Scaling Relation of Star-Forming Galaxies: Star--Dust Geometry

Characterising the physical drivers of dust attenuation in star-forming galaxies (SFGs) is essential for interpreting their spectral energy distributions and star formation histories. The infrared excess (IRX$\equiv$$L_{\rm IR}/L_{\rm UV}$) follows a universal scaling relation with metallicity, star formation rate, galaxy size, and inclination. However, the origin of the scatter around this relation remains poorly understood. We revisit this relation using $\sim$32,000 local SFGs from SDSS, GALEX, and WISE, and investigate why some galaxies deviate systematically from the best-fit relation. We find that the deviations are systematically linked to UV luminosity. UV-faint SFGs ($\log(L_{\rm UV}/{\rm L}_\odot)\le 9$) exhibit a median IRX excess of +0.23 dex, while UV-bright SFGs ($\log(L_{\rm UV}/{\rm L}_\odot)\ge 10$) show a median deficit of $-$0.20 dex relative to the relation defined by the dominant UV-intermediate population (86.38 percent of the sample). These offsets are not driven by metallicity, total infrared luminosity, or specific star formation rate. Instead, the deviations are closely linked to systematic differences in star--dust geometry. UV-faint SFGs are compact (median $R_{\rm e} = 2.90$ kpc) and preferentially viewed edge-on (median $b/a = 0.41$), leading to high effective dust column densities along the line of sight. UV-bright SFGs are extended (median $R_{\rm e} = 5.99$ kpc) and preferentially viewed face-on (median $b/a = 0.74$), allowing UV photons to escape efficiently. Galaxies with significant bulge components ($B/T > 0.4$) populate the envelope of the relation, exhibiting larger scatter. We conclude that the scatter in the IRX relation can be largely explained by variations in the three-dimensional star--dust geometry, with UV luminosity acting as an effective tracer of these geometric and physical differences.

astro-ph.GA↗