arXiv subjects

Wei Li

Publications and source records attributed to Wei Li.

At least 19 recordsLinked to original sources

arXiv2026

PulmoWorld: An Asynchronous Multimodal Medical World Model for Longitudinal Pneumonia Dynamics

Pneumonia changes across respiration, inflammation, and organ function, while the measurements that describe it arrive at different times and are often incomplete. We introduce PULMOWORLD, a world model for longitudinal pneumonia dynamics that connects asynchronous state estimation, recursive simulation, and evidence-constrained planning. History-aware filtering separates state propagation from observation-driven correction, so a measurement updates the shared belief only when it is observed. That belief supports recursive forecasts of clinical values and modality representations, and summaries of future prognosis. Cap-MBRL searches imagined continuations inside an authorized action set and replaces a frozen fallback only when a candidate meets the risk budget and a predicted-cost margin. On the pneumonia cohort MASS, recursive forecasts are competitive with task-trained sequence models, and simulated future states retain prognostic information. Trained separately on public MIMIC-IV-CXR chest radiographs and clinical trajectories, this model improves prognostic discrimination over latest-value fusion in development and is competitive on long-horizon error on the locked pneumonia panel. In five known-mechanism MASS-Syn environments, Cap-MBRL lowers selection regret relative to direct fallback execution.

cs.CV↗

arXiv2026

VGIQA: Fine-Grained Image Quality Assessment with Visual Grounding Rewards

Image quality assessment (IQA) aims to evaluate the perceptual visual quality of images. Recently, the rapid development of Multi-modal Large Language Models (MLLMs) has significantly advanced the IQA field. By integrating chain-of-thought (CoT) reasoning and reinforcement learning (RL), these models can generate more accurate and interpretable results while relying on less training data. However, existing MLLM-based methods are predominantly constrained to holistic, image-level evaluations and lack the capability to produce fine-grained, localized assessments, which substantially limits their practical applicability. To address this limitation, we propose VGIQA, a novel IQA model endowed with both local and global quality assessment capabilities. VGIQA is trained via a two-stage paradigm comprising a cold-start phase followed by reinforcement learning. Driven by carefully designed local quality assessment reward functions, our model exhibits exceptional image quality understanding on VGSR-2K, a new dataset we propose specifically for local quality evaluation, significantly outperforming existing open-source MLLMs such as Qwen2.5-VL-72B-Instruct and Qwen3-VL-32B-Instruct. Furthermore, the local quality evaluation task can effectively enhance the model's performance on the degradation perception task.

cs.CV↗

arXiv2026

Kernel Performance Engineering under Physical Non-Idealities

Physical non-idealities in integrated circuit (IC) design and fabrication are concealed by conventional hardware abstractions, yet they can limit kernel performance and obscure optimization opportunities. Using distributed shared memory (DSMEM) on NVIDIA H100 GPUs as a case study, we show how floorsweeping restricts hardware resources available for clustered execution, potentially offsetting the benefits of reduced data movement enabled by DSMEM. Guided by these insights, we develop a wave-aware clustering strategy that accounts for these resource constraints to improve kernel performance. Evaluations using a global-load-and-broadcast microbenchmark, GEMM, and the Hopper version of FlashAttention 4 demonstrate the performance implications of these constraints, with wave-aware clustering improving FlashAttention 4 throughput by up to 5.30% over its baseline. We further show that, under thermal design power (TDP) limits, reduced data movement enables higher Streaming Multiprocessor (SM) frequencies through Dynamic Voltage and Frequency Scaling (DVFS), explaining otherwise unexpected throughput improvements. Extending this analysis beyond GPUs, we show that power constraints also limit throughput scaling on Tenstorrent's Wormhole accelerator, highlighting the broader importance of accounting for physical non-idealities when analyzing and optimizing kernel performance.

cs.AR↗

arXiv2026

PatentAgentBench: Benchmarking LLM Agents with Professional Patent Task Suites

Patent practice lies at the intersection of technical, legal, and linguistic work, making it challenging yet crucial to technological innovation. Recently, rapid advances in large language models (LLMs) and agentic systems have demonstrated the potential to complete realistic business workflows and even patent works. This raises a question: can current AI systems perform patent work reliably? Existing benchmarks cover a limited range of patent tasks and simplify evidence acquisition by providing reference materials or candidate documents. In this paper, we introduce PatentAgentBench, a benchmark comprising 510 instances across eight task suites: novelty search, freedom-to-operate (FTO) search, design FTO, invalidity search, claim charting, patent drafting, office-action response, and patent translation. Curated from real patent records and professional materials, the benchmark combines task-specific metrics with expert-reviewed annotations and rubrics. Retrieval tasks such novelty search, FTO, etc., require agents to search and compare patent collections without a supplied candidate pool. The drafting and OAR tasks assess the quality of agents' work given technical disclosure and office action notice. We evaluate seven LLMs across four agent harnesses on PatentAgentBench, finding that current systems still struggle to identify novelty, assess risk, and understand practitioner intent throughout patent workflows.

cs.AI↗

arXiv2026

MedVTok: A General-Purpose Medical Visual Tokenizer

Multimodal medical AI requires a common visual interface for encoding heterogeneous imaging evidence for patient condition modeling. However, existing medical visual tokenizers are often tied to a specific dimension and imaging modality, forcing multimodal systems to rely on fragmented and poorly reusable representations. We introduce MedVTok, a general-purpose visual tokenizer that maps 2D images and 3D volumes from diverse imaging modalities into a unified token space. Building such a tokenizer poses two key challenges: preserving anatomical structure across dimensions and capturing clinical semantics across modalities. To this end, MedVTok introduces (i) slice differential regularization, which explicitly models inter-slice anatomical coherence missing in previous per-pixel/per-voxel supervision, and (ii) multi-expert representation alignment, which integrates knowledge from multiple medical expert encoders while retaining modality-specific complementary knowledge. Trained on 28M images and 70K volumes across 9 imaging modalities, MedVTok supports classification, retrieval, segmentation, synthesis, visual question answering, and medical report generation through a single tokenizer. Across 24 benchmarks, MedVTok achieves state-of-the-art performance, demonstrating a scalable token interface that unifies medical image dimension, modality, and task usage.

eess.IV↗

arXiv2026

Training-Free Inference Acceleration for Video Large Language Models: Methods, Trade-offs, and Evaluation

Video Large Language Models (VLMs) have demonstrated strong capabilities in video understanding and reasoning, but processing long visual sequences incurs substantial computational, latency, and memory overhead. This survey examines training-free inference acceleration, focusing on reducing inference costs while preserving task-relevant spatiotemporal evidence. We organize existing methods by their intervention stages in the inference pipeline, covering input, visual encoding, projector, prefill, and decoding. Post-training quantization (PTQ) is discussed separately as a cross-stage extension with varying calibration requirements. We analyze how intervention position affects evidence preservation and computational savings, and why local efficiency improvements may not translate into end-to-end acceleration. We complement this analysis with quantitative evidence from prior studies and our own experiments under a shared evaluation setting to assess accuracy-latency trade-offs.

cs.CV↗

arXiv2026

The ascending chain condition for radical well-mixed difference ideals

We establish a finite-basis theorem for radical well-mixed difference ideals in a difference polynomial ring with one endomorphism over an arbitrary difference field. The result extends to every finitely difference-generated difference algebra over such a field and is equivalent to the ascending chain condition for radical well-mixed difference ideals, hence resolving a long-standing open problem in difference algebra. The proof proceeds in two stages. Over an inversive difference coefficient field, we establish a uniform inverse-transform bound relating the perfect closure of a finite system to its radical well-mixed closure. This uses Noetherian induction in a fixed truncated polynomial ring, together with generic freeness and going-down for flat ring homomorphisms. We then prove that the radical well-mixed closure of a system over a difference field $K$ is the contraction of its radical well-mixed closure over the inversive closure of $K$. This contraction result follows from successive prime lifting along flat extensions and removes the inversiveness assumption without requiring the coefficient-field extension to be finitely generated.

math.AC↗

arXiv2026

Approximation algorithms for identical parallel machine scheduling with assignment costs

We study non-preemptive scheduling on identical parallel machines in which assigning job $j$ to machine $M_i$ incurs the cost $c_{ij}=a_iw_j+b_j$. The processing times $p_j$ and the cost weights $w_j$ are independent, and the number of machines is part of the input. For every fixed $\varepsilon\in(0,1]$ and every deadline $T$ for which a schedule with makespan at most $T$ exists, we give a polynomial-time algorithm that computes a schedule with makespan at most $(1+\varepsilon)T$ whose total assignment cost does not exceed the minimum under the deadline $T$. Using this algorithm as a subroutine, we obtain a polynomial-time approximation scheme for minimizing the makespan under a total cost budget whose schedules always satisfy the budget; this answers an open question for costs $c_{ij}=l_ic_j$ with $c_j$ independent of $p_j$. For the lexicographic problem, in which the total assignment cost is minimized among all schedules of minimum makespan, we obtain in polynomial time a schedule with makespan at most $1+\varepsilon$ times the minimum and total assignment cost at most that of a lexicographically optimal schedule; this improves the makespan guarantees of earlier algorithms for additive and multiplicative costs. For costs $c_{ij}=s_i+a_iw_j$, where $s_i$ is charged for each job processed on $M_i$, the same guarantees are obtained in quasi-polynomial time.

cs.DS↗

arXiv2026

Weak type $(1,1)$ estimates for noncommutative maximal truncated rough singular integrals

We prove a weak type $(1,1)$ estimate for maximal truncations of noncommutative rough singular integrals whose angular kernels satisfy the classical $L\log L$ integrability and cancellation conditions. The main ingredient of the proof is a new linearization method for handling the maximal truncations acting on the bad functions arising from the noncommutative Calderón--Zygmund decomposition.

math.FA↗

arXiv2026

Perfect Born Sampling of Symmetric Thermal Tensor Network States

Accurate calculations of correlated fermions at low temperatures constitute a major challenge in quantum many-body physics, most notably in the studies of high-$T_c$ superconductivity and ultracold atom optical lattices. Stochastic sampling of tensor-network states offers a promising route to tackle this problem; however, existing schemes have long faced a fundamental dilemma---sampling efficiency and symmetry acceleration \textit{cannot} be achieved simultaneously. Here we propose a perfect Born sampling approach for thermal tensor networks, which performs importance sampling directly from the purified density matrix via the Born rule and incorporates Abelian and non-Abelian symmetries by sampling symmetry quantum numbers. We benchmark the method, realized as both stochastic matrix product states (stoMPS) and stochastic projected entangled pair states (stoPEPS), on large-scale quantum lattice models. Using stoMPS, we accurately simulate the square-lattice Hubbard model on cylinders up to width $W=10$, and study the triangular-lattice Hubbard model down to $T/t = 1/64$. We reveal scalar chiral order at half filling and kinetic ferromagnetism upon electron doping, with the latter directly compared to ultracold atom experiments. We further extend the stoMPS method to compute finite-temperature quantum dynamics, as demonstrated by the optical conductivity of the Hubbard model. We showcase the generalization of stoPEPS on up to $20\times20$ square-lattice quantum Ising model, where accurate results are obtained near its quantum critical point. Our method combines high sampling efficiency with full symmetry acceleration. The resulting state-of-the-art framework captures both equilibrium and dynamical properties down to ultralow temperatures, thereby providing a powerful theoretical tool for correlated quantum systems, particularly fermion Hubbard quantum gas microscopy.

cond-mat.str-el↗

arXiv2026

OpenHome: From Perception to Insight in Smart-Home Camera Videos

Vision language models (VLMs) have been increasingly adopted in consumer smart-home camera systems to interpret recorded videos and provide insights. However, video clips captured from smart-home cameras are different from curated web videos dominating existing VLM benchmarks. Smart-home video clips often do not contain subtitles or narration and are captured from different viewing angles, under various lighting conditions, and in diverse scenarios. However, the capabilities of the model in this domain, under sparse visual signals and limited context, are not fully assessed by existing VLM benchmarks. To close this gap, we introduce OpenHome, a visual-only, four-option multiple-choice benchmark for evaluating smart-home video understanding. It contains 2,254 questions on 1,118 real-world video clips and organizes 10 tasks into three capability levels. Perception evaluates visible content, Reasoning evaluates causal, spatial, and temporal relationships, and Insight evaluates subject intent, supporting evidence and appropriate homeowner responses. Based on the dataset, we describe the task design, data pipeline and evaluation protocol, and report results for 32 closed-source and open-weight VLMs. The best model, Gemini 4 Argon, reaches 81.2% overall accuracy. Frontier closed-source models outperform open-weight models on every level. The best closed-source model leads the best open-weight model by 16.3% on Perception, 22.4% on Reasoning and 37.4% on Insight.

cs.CV↗

arXiv2026

Forecasting carbon-risk hotspots ahead of the extreme 2026 El Niño

A potentially super El Niño is forecasted in 2026, raising the need to anticipate land carbon-cycle risks before its impacts unfold. We develop a prospective early-warning framework that uses dynamic global vegetation models driven by an ensemble of ECMWF seasonal climate forecasts and machine-learning burned area predictions. Forecasts indicate a $0.23 \pm 0.21$ PgC reduction in global net biome productivity in the second half of 2026, with hotspots of sink weakening in eastern South America and elevated fire risk in Southeast Asia.

physics.ao-ph↗

arXiv2026

AdvFidelity: Unifying Preference Alignment and Few-Step Image Generation within Reinforcement Learning

Practical image generation requires both human preference alignment and few-step efficiency, yet combining reinforcement learning (RL) with distillation is often multi-stage and may erode RL gains We observe that few-step RL progressively improves structural coherence, even with specialized rewards, but leaves low-level artifacts that these rewards may fail to penalize. To address this gap, we propose AdvFidelity, which integrates artifact-sensitive fidelity supervision into RL through an online adversarial reward. This enables joint reward alignment and high-fidelity few-step generation within a single RL stage, without distribution matching distillation (DMD) or other explicit distillation objectives. We train a discriminator to rank high-quality references (positives) above content-matched reconstructions (negatives) produced by the current generator. Reconstructing moderately noised references largely preserves style and scene layout while exposing low-level degradations. This encourages the discriminator to assess artifacts rather than content differences. With negatives reconstructed online, alternating updates adapt fidelity supervision to the generator's evolving failure modes. References can come from existing target-domain data or full-step model samples. Built on LeapAlign, AdvFidelity backpropagates gradients from both rewards through the few-step sampling trajectories used at inference. Across two model backbones and three task settings, AdvFidelity matches full-step LeapAlign in alignment, task performance, and overall generation quality and remains highly competitive with other evaluated full-step RL methods, using only 4 or 8 steps without CFG.

cs.CV↗

arXiv2026

Néel Ordered Magnetic Phases in Bipartite Quasicrystals

Magnetism is a fundamental research area in which the recently proposed altermagnetism (AM) has become an emergent frontier. Very recently, the quasicrystal (QC) was proposed as a possible platform to realize AM. However, the existence of AM in QCs still lacks vigorous evidence. In this work, we adopt the sign-problem-free projector quantum Monte Carlo (PQMC) algorithm to investigate the magnetic phases in the half-filled Hubbard models in various 2D bipartite QCs, and always obtain Néel ordered states. While the Néel states in bipartite crystals are usually antiferromagnetism (AFM), we find it common that those in bipartite QCs can also be AM or ferrimagnetism (FIM). Based on symmetry analysis, combined with our comprehensive PQMC results, we propose a general criterion for determining the magnetism classes of the Néel states in a bipartite QC: According to whether the two sublattices are related by inversion, a non-inversion point-group operation such as mirror or rotation, or no operation about the unique symmetry center, the corresponding Néel state is AFM, AM or FIM / geometry-enforced AM (G-AM), respectively. The G-AM is a special type of $s$-wave AM caused by the unique QC structure formed through projection from high-dimensional crystals to lower dimensions, which is rare in crystals. For example, our results yield symmetry-protected AM for the two $D_4$-symmetric Thue-Morse QCs and G-AM for the $D_5$-symmetric Penrose QC at half-filling. Our results provide a solid foundation for experimental investigations and potential applications of different classes of magnetism in QCs.

cond-mat.str-el↗

arXiv2026

TTS-Q: Speaker-Aware Post-Training Quantization of Text-to-Speech Diffusion Transformers

Zero-shot text-to-speech (TTS) can clone a voice from seconds of audio. Yet the quality comes from billion-parameter diffusion transformers whose size and multi-step sampling put them out of reach of the on-device, assistive, and real-time applications that need them most. Post-training quantization (PTQ) shrinks such models without retraining, but it has been built for language and vision, not speech. At 4 bits, a strong quantizer leaves intelligibility close to full precision but consistently erodes speaker timbre, so error rates stay low while the voice drifts from the target speaker. We ask why quantization hurts speech, where the damage lives, and whether it can be repaired without retraining. To find out, we run the first 4-bit weight-and-activation (W4A4) PTQ study of a TTS diffusion transformer at 1B and 3.5B. The timbre cost comes from both the weight and the activation quantization, but only the activation part has the timestep structure a precision schedule can exploit. We therefore introduce TTS-Q, which selects hard-to-quantize items from task-matched calibration candidates and applies a budget-balanced mixed-precision schedule that assigns higher precision to late-step feed-forward activations. TTS-Q recovers timbre best at both scales, at no cost to intelligibility. Separately, the deployed calibrated-W4A4 reference reduces peak memory on GPU, providing a practical reference for resource-constrained TTS deployment.

cs.SD↗

arXiv2026

Contrasting Crystal-Field and Dipolar Anisotropies in the Triangular Antiferromagnet NaBaEr(BO$_3$)$_2$

The magnetic ground states of rare-earth frustrated magnets can be governed by competing energy scales arising from crystal-field anisotropy, exchange interactions, and long-range dipolar interactions. Here we investigate this interplay in the triangular-lattice antiferromagnet NaBaEr(BO$_3$)$_2$ by combining magnetization, specific heat, and neutron scattering measurements. Neutron powder diffraction resolves rotations of the ErO$_6$ octahedra and supports the lower-symmetry $R\bar{3}$ structure over the previously proposed $R\bar{3}m$ model. Magnetization measurements and inelastic neutron scattering establish an effective spin-1/2 description of its low-temperature magnetism with pronounced easy-axis anisotropy, arising from a well-isolated Kramers ground-state doublet separated from the first excited crystal-electric-field level by approximately 4 meV. Below $T_{\rm N}=0.1$ K, single-crystal neutron diffraction reveals A-type antiferromagnetic order with the ordered moment predominantly oriented within the $ab$ plane, despite the pronounced paramagnetic easy-axis anisotropy. The estimated dipolar interaction scale is comparable to $T_{\rm N}$, and classical dipolar-energy calculations for the observed stacking pattern favor an in-plane moment orientation. These results reveal contrasting crystal-field and dipolar anisotropies in determining the magnetic ground state of NaBaEr(BO$_3$)$_2$.

cond-mat.str-el↗

arXiv2026

Quantized High-Performance High-Order Moment-Encoded LBM

Lattice Boltzmann methods (LBM) are widely used for visually rich fluid simulation, but their high memory bandwidth and storage requirements limit GPU performance and scalability, especially when coupled with complex geometry. Recent high-order moment-encoded LBM (HOME-LBM) reduces memory cost by storing compact velocity moments and performing collision directly in moment space, but its tightly coupled boundary treatment introduces divergent control flow. We present a high-performance, quantized HOME-LBM framework for large-scale turbulent fluid-solid simulations on a single GPU. Our geometry-decoupled computation scheme separates the regular fluid update from a sparse solid-correction stage, isolating geometry-dependent computations to improve GPU efficiency. To further reduce storage and memory traffic, we develop a reduced-precision representation for HOME-LBM by combining von Neumann stability analysis, empirical moment-range estimation, and runtime decomposition of second-order moments to control quantization error. We evaluate our approach on real-time demonstrations and large-scale examples using both D3Q19 and D3Q27 lattices, with additional 2D and 3D numerical experiments assessing accuracy and stability. Across the tested benchmarks, our framework substantially accelerates the original FP32 HOME-LBM implementation, with a maximum speedup of 32.5x on the Hilbert-curve geometry. Moment storage is reduced by 50% in fluid-only simulations and 25% in fluid-solid coupling examples, while maintaining numerical stability, fidelity, and high visual quality. In the static delta-wing benchmark, our solver achieves a 5.2x speedup over Leapfrog Flow Maps (LFM) under the tested configuration.

cs.GR↗

arXiv2026

Vision-Language-Delta-Action: Learning Compact Deltas as Actionable Transition Interfaces for Robotic Manipulation

Robotic manipulation is not merely observation-to-action generation, but the realization of task-specified state changes in the physical world. Existing vision-language-action (VLA) methods directly learn observation-to-action mappings, imposing a substantial learning burden due to the gap between semantic understanding and action execution. Although world-action models (WAMs) improve control through future prediction, modeling complete future states wastes capacity on static appearance and action-irrelevant changes. We propose Vision-Language-Delta-Action (VLDA), a paradigm that reformulates full future prediction as compact actionable transition modeling. Its core is a compact Delta variable that captures latent world-state changes, serving as an interface between vision-language understanding and action generation. VLDA learns Delta through a unified objective composed of the Local Transition Filter (LT-Filter) and Global Transition Anchor (GT-Anchor). LT-Filter captures task-relevant manipulation-centric changes while suppressing static reconstruction biases. GT-Anchor provides coarse future-level regularization to preserve the overall direction of state changes. Delta-Driven Action Refinement (DDAR) further grounds Delta in the action manifold and injects it as a structured seed to guide action denoising. We implement VLDA within LoongForge, an open-source training framework for large-scale foundation and embodied models, using its modular and distributed infrastructure for unified two-stage optimization. Experiments show that VLDA improves long-horizon consistency, robustness, and generalization, validating actionable Delta as an alternative to direct action prediction and full future-state modeling.

cs.RO↗