Searcharxiv⌕ Search

arXiv subjects

Shihao Wang

Publications and source records attributed to Shihao Wang.

At least 19 recordsLinked to original sources
arXiv2026

4K Image Super-Resolution on Mobile NPUs: Mobile AI 2026 Challenge Report

Image super-resolution is widely used on mobile devices for upscaling various types of image data. This creates a need for efficient deep learning-based models that are not only performant but can also be deployed on neural processing units (NPUs) present in the majority of mobile and IoT chipsets. In this Mobile AI challenge, we address this problem and propose the participants to design fast solutions that can upsample images from Full HD to 4K resolution while running on popular mobile NPUs from Qualcomm and MediaTek. For this, the participants were provided with the DIV2K dataset and trained models to do an efficient 3X image upscaling. The competition had two tracks: in the first track, the models had to be fully quantized (INT8), and the runtime of all models was evaluated on the Qualcomm Snapdragon 8 Elite Gen 5 NPU. In the second track, the participants were developing floating-point solutions that were evaluated on the MediaTek Dimensity 9500 NPU. Top solutions proposed in this challenge are able to reconstruct 4K images under 5 milliseconds while reaching a PSNR score of 30 dB. A comprehensive description of the models developed in the challenge is provided in this paper.

cs.CV↗
arXiv2026

Spatial-Mode Selection for Local Mineral Absorption Spectroscopy

Absorption-band depth measures the fractional decrease in mineral reflectance below its local continuum. Recovering this feature strength for a small mineral region supports quantitative comparisons between regions, but diffraction mixes its light with the surroundings. Conventional spectral fitting can separate mean contributions; it cannot remove background photon noise already recorded. We develop a quantum-inspired measurement framework that uses spatial-mode demultiplexing (SPADE) to retain the target absorption response while reducing background acceptance before detection. The framework connects finite mineral-region geometry and reference spectra to estimation precision. A loss-dependent design condition identifies when adding SPADE exposure improves on direct imaging (DI) under fixed-response Poisson measurements with unknown scene parameters. Joint inference from initial DI observations and both mode ports estimates target depth together with an unknown, spatially shared background spectrum. In equal-time simulations that include localization uncertainty, alignment time, calibration, and detector costs, this acquisition reduces depth root-mean-square error (RMSE) by 14\%--26\% across five absorption strengths relative to all-pixel DI with the same background freedom. These results use a supplied target feature shape and geometry, with target continuum slope fitted and curvature fixed. The analysis separates optical selectivity from parameter ambiguity and acquisition cost, providing a quantitative basis for spatial-mode receiver design in local mineral spectroscopy.

physics.optics↗
arXiv2026

Beyond Direct Imaging: Quantum Measurement Design for Small-Target Detection in UAV Hyperspectral Sensing

Small mineral patches can be spectrally distinct yet difficult to detect when diffraction mixes their light with the surrounding rock. Conventional hyperspectral detectors analyze recorded intensities, leaving an opportunity to recover target evidence through the optical measurement itself. We develop a quantum-assisted framework for exploiting this opportunity in prospective unmanned aerial vehicle (UAV) hyperspectral sensing. A common radiometric and diffraction model describes direct imaging and spatial-mode measurements of the same light. The theory identifies target changes that survive background fitting and explains when mode measurements retain more detection information. It quantifies losses from mode mismatch and unresolved outputs, linking the quantum benchmark to finite receiver design. A common-basis result identifies when material spectra can share a spatial measurement at fixed geometry. An adaptive receiver uses an imaging pilot to select candidate modes and combines both stages in a full-region presence test. Simulations based on measured mineral spectra and published UAV acquisition scales demonstrate gains over complete direct imaging at a shared active-exposure budget, including pilot cost and optical loss. At matched empirical 1\% region-level false alarms, detection increases from 62.95\% to 69.65\% for a representative target. Paired analysis supports gains with registration uncertainty and, separately, under controlled nonuniform backgrounds. Calibrated mode errors reveal where measurement losses weaken or reverse the benefit. These results connect background identifiability to receiver choices and their operating conditions. They provide a framework for designing spatial measurements alongside spectral channels to detect small material features.

quant-ph↗
arXiv2026

Compatibility of Perverse Truncation of Ngô Fibrations

We study compatibility of perverse truncations for morphisms between Ngô fibrations with rationally smooth total spaces. We show that, under an effective rank inequality, equivariant pullbacks preserve the normalized perverse truncations. Applied to the diagonal, this gives multiplicativity of perverse truncations for Hitchin systems with rationally smooth total spaces.

math.AG↗
arXiv2026

BitMTP: When Multi-Token Prediction Meets Low-Bit Large Language Models

For efficient deployment of large language models (LLMs), methods of ultra-low-bit backbones reduce weight storage, memory traffic, and linear-layer execution cost, while multi-token prediction (MTP) methods reduce sequential decoding steps by predicting multiple future tokens in parallel. Although complementary, the direct combination of these methods cannot accelerate decoding at comparable quality: masked drafting lacks intermediate token input supplied in next-token training, while wider computation and causal key-value (KV) updates can offset the savings from fewer steps. To address these challenges, we propose BitMTP, an algorithm-system co-design framework for efficient MTP on ultra-low-bit LLM backbones. A joint next-token prediction and MTP objective equips a single model with both autoregressive (AR) generation and block-causal drafting. Confidence-only decoding, speculative-only decoding, and hybrid decoding provide different quality-throughput operating points without using separate draft and verifier models. Accepted-prefix alignment keeps the committed prefix, materialized causal KV boundary, and CUDA Graph replay state consistent for continued generation. BitMTP uses this contract to select width-specialized low-bit execution and reuse captured graphs over shared KV storage. On NVIDIA H20, BitMTP with hybrid decoding achieves five-task mean speedups of 2.2x on BitNet b1.58-2B and 2.1-2.4x on 1.7B-4B Bonsai models at comparable task quality. On RTX 3090 and Jetson Orin, it further achieves up to 2.0x in five-task mean throughput and 3.3x on MBPP+, respectively. These results demonstrate that BitMTP accelerates ultra-low-bit LLM generation across backbones, model sizes, and GPU types. The source code is released at https://github.com/PolyU-VCLab/BitMTP

cs.CL↗
arXiv2026

OmniCompass: A Holistic Benchmark for Omni-Modal Video Grounding and Understanding

With the rapid development of multimodal large language models (MLLMs), evaluating their video understanding capabilities has become increasingly important. Existing video understanding benchmarks cover only part of the evaluation dimensions, limiting systematic diagnosis of omni-modal video understanding. We introduce OmniCompass, a holistic benchmark for omni-modal video grounding and understanding. OmniCompass defines two evaluation tracks: OmniCompass-VL assesses vision-language models on tasks requiring visual evidence, while OmniCompass-AV assesses omni-modal models on tasks requiring joint audio-visual evidence. Each track includes three complementary tasks: temporal grounding (TG), spatial grounding (SG), and question answering (QA), assessing when a referenced target appears, where it is, and what happens. QA covers five perception and three reasoning categories in OmniCompass-VL, and three perception and two reasoning categories in OmniCompass-AV. OmniCompass-VL contains 820 videos and 2,460 instances across domains and three duration types: short, medium, and long. A narrative- and audio-rich subset forms OmniCompass-AV with 240 videos and 720 instances. Each instance is evaluated under two paired settings: text-only and text+image, specifying the same target through text and visual-language guidance. All instances are manually annotated with multi-round quality verification. Evaluations of 50 MLLMs show that open-source models still lag behind proprietary counterparts, particularly in audio-visual understanding. Precise temporal and spatial grounding remains challenging even for leading models, and their predictions still exhibit inconsistencies across the paired target specifications.

cs.CV↗
arXiv2026

A Sharp Bound for the Second Betti Number of Hyperkähler Sixfolds

Let $X$ be a smooth compact irreducible hyperkähler manifold of dimension six. We prove that the second Betti number satisfies $b_2(X) \le 23$. Furthermore, we show that the third cohomology vanishes whenever $b_2(X) \ge 19$, and that all odd cohomology vanishes when $b_2(X) = 23$. If the second Chern class lies in the subalgebra generated by $H^2(X,\mathbb{Q})$, then $b_2(X) \in \{5,8,23\}$. The proof uses the Looijenga--Lunts--Verbitsky decomposition, curvature inequalities for Segre classes, and the Hodge--Riemann bilinear relations on the projective cotangent bundle.

math.AG↗
arXiv2026

A Proof of Mongardi's Conjecture on Finite Symplectic Group Actions on Hyperkähler Manifolds of $K3^{[n]}$ Type

Let a finite group act faithfully by symplectic automorphisms on a manifold of $K3^{[n]}$ type for $n\geq2$. We prove Mongardi's conjecture that the strict inequality $\mathrm{rk}(S)+\ell(A_S)<24$ holds for its associated Leech coinvariant lattice $S$. The converse was already proved independently by Huybrechts and Mongardi: any Leech coinvariant lattice $S$ satisfying the strict inequality and of rank at most $20$ is realized by such an action for some $n\geq2$. The key idea is to consider the minimum norms of vectors representing discriminant classes of $S$. The absence of numerical walls gives lower bounds for certain classes, while the structure of Leech lattice implies upper bounds for all discriminant classes. Comparing these bounds reduces the proof to finitely many values of $n$. We exclude the remaining cases using further lattice arguments, together with Höhn--Mason's classification of fixed-point sublattices of Leech lattice, thereby proving Mongardi's conjecture.

math.AG↗
arXiv2026

Finite Symplectic Automorphisms of Hyperkähler Manifolds of $K3^{[n]}$-Type Act Trivially on Discriminant Groups

We give an alternative proof of Mongardi's result that every finite group of symplectic automorphisms of a hyperkähler manifold of $K3^{[n]}$-type acts trivially on the discriminant groups of its second cohomology lattice and its coinvariant lattice. The proof combines sphere-packing bounds, Vinberg's reflection chambers and structure results on even $2$-elementary lattices.

math.AG↗
arXiv2026

QAM: Quadratic-Accurate Checkpoint Merging via Sequential Consistency

Saved checkpoints record states along a training trajectory, but generally do not determine the updates at states that would be visited under a different schedule. We study how accurately these checkpoints can reconstruct the endpoint of a sequential reference with prescribed update strengths. Under a common local transition model, two checkpoint-index moment conditions characterize all convex merges that agree with this reference through second order. We then prove an information limit that for nondegenerate profiles, no algorithm using only a fixed-length gradient-descent (GD) history with step size $h$ can achieve $o(h^3)$ endpoint error uniformly over a fixed class of smooth, strongly convex losses. The lower bound follows from two losses with identical GD checkpoint histories but sequential reference endpoints separated by $Ω(h^3)$. \textbf{Quadratic-Accurate Merging} (QAM) achieves a matching uniform $O(h^3)$ endpoint error bound. Its explicit coefficients also define the unique profile-dependent merge that exactly matches the sequential GD reference across all fixed quadratic objectives. Across two public Adam checkpoint trajectories (SmolLM3-3B and OpenEuroLLM-Prelude-9B), three windows and three profiles per model, and 15 tasks, QAM shows mixed results for short windows and broader advantages over \textbf{Warmup-Stable and Merge} (WSM) for longer windows. Matched-moment GSM8K diagnostics further show that local consistency alone does not fully determine downstream scores. These results characterize the reconstruction limits of saved histories, provide a coefficient rule that attains the optimal rate, and assess its practical utility.

cs.LG↗
arXiv2026

A Risk-Adaptive and Evidence-Constrained Framework for Generative AI Feedback in Programming Education

Generative artificial intelligence can turn learning analytics into personalized support, but feedback systems must decide when to intervene, which evidence to use, and how much assistance to provide. We developed a risk-adaptive, evidence-constrained framework for introductory programming using 2993 failed-submission states from 215 students. Student-disjoint models predicted persistent failure and related outcomes; four matched feedback conditions were generated for 136 cases; and calibrated risk informed capacity-limited intervention policies. The validation-selected logistic regression model achieved a test precision-recall area under the curve of 0.550 and a receiver operating characteristic area under the curve of 0.681. Broader student histories improved prediction of unmodified resubmission. After standardized repair and evidence gating, 519 of 544 newly generated messages contained all required components. A fixed-threshold sequential policy selected 17.8% of eligible test states and captured 25.2% of observed persistent failures. These findings support an evidence-gated progressive assistance strategy: calibrated risk guides intervention timing, recorded evidence constrains feedback content, and assistance progresses from self-checks to localized hints when warranted. The framework connects prediction, decision-making, and grounded generation while keeping their evaluation outcomes distinct.

cs.AI↗
arXiv2026

Non-symplectic Indices of Automorphism Groups of Smooth Cubic Fourfolds

We study the full automorphism groups of smooth cubic fourfolds with prescribed symplectic automorphism group. Our starting point is the classification of symplectic automorphism groups by Laza and Zheng. We focus on the non-symplectic index, namely, the index of the symplectic automorphism group in the full automorphism group. We prove general restrictions on this index. We also compute bounds by group-theoretic and lattice-theoretic methods. In several cases, we determine all possible indices. For coinvariant lattices of rank 19, we classify all possible pairs consisting of the symplectic automorphism group and the full automorphism group.

math.AG↗
arXiv2026

Classification of Automorphism Groups of Smooth Cubic Threefolds and Fourfolds

We classify the automorphism groups of smooth cubic threefolds and fourfolds. We also show that there are $156$ (respectively, $40$) connected families of smooth cubic fourfolds (respectively, threefolds) with specified automorphism group action. For each family, the group action and defining equations are also given. The classification combines both representation theory (based on GAP) and lattice theory (based on SageMath).

math.AG↗
arXiv2026

ArtChart: Faithful Artistic Chart Generation with Integrated Text Rendering

Artistic charts combine data visualization with expressive marks, textures, and typography, but they are difficult for image generators: an output is useful only when its stylization preserves chart geometry, exact in-image text, and the semantic binding between labels and marks. We introduce ArtChart, a framework for faithful artistic chart generation with integrated text rendering. Given a structured chart specification and an artistic prompt, ArtChart first renders a text-free grayscale layout that encodes the target chart geometry, then trains a chart-specific control module to preserve mathematical structure. To address the remaining text and layout errors, we further refine the generation policy through GRPO-based reinforcement learning with OCR-based text rewards, VLM-based layout rewards, and aesthetic rewards. A multi-expert distillation stage reconciles these objectives by distilling single-reward experts into one balanced generation policy. We also construct ArtChart-Bench, a bilingual 2K-prompt benchmark covering four chart types, controlled value distributions, diverse label/value formats, and 15 artistic styles, together with ArtChart-Eval, a six-axis evaluation protocol measuring mathematical logic, text accuracy, text layout, aesthetics, instruction following, and readability. Experiments on ArtChart-Bench show that ArtChart consistently outperforms prompt-only, image-editing, and generic ControlNet baselines, with the largest gains on mathematical fidelity and label-layout binding while maintaining competitive visual quality. These results suggest that artistic chart generation should be evaluated as reliable visual communication rather than as generic stylized image synthesis.

cs.CV↗
arXiv2026

Vesta: A Generalist Embodied Reasoning Model

Robots operating in open-world environments must seamlessly integrate localization, spatial reasoning, navigation, and long-horizon planning. While specialist models excel at individual tasks, deploying a multi-model stack is computationally expensive and prone to cascading errors. We present Vesta, a unified embodied generalist that consolidates these capabilities into a single foundation model. Our approach combines a diverse and massive curated corpus designed to induce spatial grounding and a simple multimodal memory harness that enables reasoning over extended time horizons. Across diverse benchmarks, Vesta on average beats individual SOTA baselines by >$20\%$ and beats an ensemble of per-category-best baselines by $>10\%$ -- thus demonstrating that a generalist model can match or exceed specialists. On real-world robotic tasks requiring memory and reasoning, Vesta improves task success by >35\%. Our work thus demonstrates that a single generalist is a feasible, scalable, and arguably preferable alternative to combining specialists.

cs.RO↗
arXiv2026

DepthMaster: Unified Monocular Depth Estimation for Perspective and Panoramic Images

While monocular depth estimation has achieved significant progress, achieving generalized metric depth estimation for both narrow field-of-view (FoV) perspectives and $360^\circ$ panoramas remains an unsolved challenge. Existing methods are often tailored to specific camera types and struggle to produce accurate metric depth that generalizes across diverse settings. This limitation stems from two key challenges: the inherent geometric discrepancy between perspective and panoramic cameras, and the scarcity of panoramic training data with metric annotations. In this work, we introduce DepthMaster, a unified metric depth estimation framework. Rather than employing specialized networks to learn spherical distortions, we reformulate the problem by decomposing panoramic images into overlapping perspective patches. Crucially, distinct from prior projection-based methods that rely on ad-hoc architectural modifications to handle boundaries, we introduce a novel Correspondence Consistency Loss (CCL) and inject virtual projection cameras as geometric priors, allowing us to seamlessly stitch the patches while avoiding specialized operators and keeping the backbone largely compatible with standard Transformer designs. This strategy also resolves the geometric differences by unifying all inputs into a canonical perspective representation, and effectively circumvents data scarcity by directly unlocking powerful metric priors from vast perspective datasets. Trained on a mixed dataset that contains only one panorama dataset, DepthMaster achieves state-of-the-art zero-shot performance on 13 diverse datasets, outperforming not only universal methods but also leading specialist models in both perspective and panoramic domains.

cs.CV↗
arXiv2026

PRISM: Gauge-Invariant Tangent-Space Differentially Private LoRA

Applying differential privacy (DP) via DP-SGD to Low-Rank Adaptation (LoRA) is a natural approach for privacy-preserving fine-tuning. However, LoRA's low-rank parameterization poses a fundamental challenge. In LoRA, each trainable update is represented as a low-rank matrix $Z = AB^\top$, but this factorization is inherently non-identifiable: many factor pairs $(A,B)$ represent the same update $Z$. As a result, applying DP-SGD directly to the factors induces gauge-dependent perturbations on $Z$, and we show that this naive DP-LoRA can lead to unbounded noise amplification. We propose PRISM, an intrinsic DP mechanism for LoRA that is gauge invariant by construction, avoids bilinear noise amplification, and admits an efficient low-dimensional noise sampler. Moreover, PRISM yields a closed-form characterization of the effective intrinsic noise induced on $Z$, enabling stable privacy-utility trade-offs through bounded, gauge-invariant perturbations. We establish standard $(ε,δ)$-DP guarantees for PRISM and introduce a DP-aware, gauge-invariant adaptive update rule that prevents adaptive optimization from amplifying injected privacy noise, improving numerical stability in practice.

cs.LG↗
arXiv2026

LocateAnything: Fast and High-Quality Vision-Language Grounding with Parallel Box Decoding

Vision-language models (VLMs) commonly formulate visual grounding and detection as a coordinate-token generation problem, serializing each 2D box into multiple 1D tokens that are learned and decoded largely independently. This token-by-token decoding mismatches the coupled structure of box geometry and creates a practical inference bottleneck due to strictly sequential generation. We introduce LocateAnything, a unified generative grounding and detection framework based on Parallel Box Decoding (PBD). By decoding geometric elements such as bounding boxes and points as atomic units in a single step, LocateAnything preserves intra-box geometric coherence and unlocks substantial parallelism. We show that PBD improves both decoding throughput and localization accuracy. We further develop a scalable data engine and curate LocateAnything-Data, a large-scale dataset with more than 138 million training samples, substantially increasing data diversity for high-precision localization. Extensive evaluations show that LocateAnything advances the speed-accuracy frontier, achieving significantly higher decoding throughput while improving high-IoU localization quality across diverse benchmarks. The results highlight the complementary benefits of Parallel Box Decoding and large-scale training data in enabling efficient and precise unified visual grounding and detection.

cs.CV↗