arXiv subjects

Wei Li

Publications and source records attributed to Wei Li.

At least 163 records · Page 9Linked to original sources

arXiv2026

Octopus: History-Free Gradient Orthogonalization for Continual Learning in Multimodal Large Language Models

Continual learning in multimodal large language models (MLLMs) aims to sequentially acquire knowledge while mitigating catastrophic forgetting, yet existing methods face inherent limitations: architecture-based approaches incur additional computational overhead and often generalize poorly to new tasks, rehearsal-based methods rely on storing historical data, raising privacy and storage concerns, and conventional regularization-based strategies alone are insufficient to fully prevent parameter interference. We propose Octopus, a two-stage continual learning framework based on History-Free Gradient Orthogonalization (HiFGO), which enforces gradient-level orthogonality without historical task data. Our proposed two-stage finetuning strategy decouples task adaptation from regularization, achieving a principled balance between plasticity and stability. Experiments on UCIT show that Octopus establishes state-of-the-art performance, surpassing prior SOTA by 2.14% and 6.82% in terms of Avg and Last.

cs.LG↗

arXiv2026

GraphThinker: Reinforcing Temporally Grounded Video Reasoning with Event Graph Thinking

Video reasoning requires a fine-grained understanding of the temporal dependencies and event-level relations between objects and events in videos. Current Multimodal Large Language Models (MLLMs) are prone to severe temporal hallucinations in video reasoning. An underlying cause of these hallucinations is weak visual-temporal grounding and the lack of explicit structure for modelling event relations. Models often rely on auxiliary text, such as dense captions, rather than explicitly anchoring their reasoning in actual visual evidence. However, these textual representations are inherently unstructured and fail to provide explicit causal constraints needed to guide the model's reasoning. In this work, we propose GraphThinker, a reinforcement finetuning method that constructs a structured event representation of a video and enforces visual grounding to jointly reduce reasoning hallucinations. Specifically, we employ an MLLM to construct an Event-based Video Scene Graph (EVSG) that captures both intra- and inter-event relations, guiding a structured video reasoning process. Moreover, we address the weak grounding issue by introducing a novel visual attention reward during reinforcement finetuning that encourages the model to actively attend to reliable visual cues. On the RexTime dataset, GraphThinker achieves an over 4% improvement in IoU=0.3 for moment localisation. On the VidHalluc dataset, GraphThinker achieves a 9.8% improvement in reducing temporal sequence hallucination and a 7.6% gain in Binary QA in reducing action hallucination, compared to the state-of-the-art methods.

cs.CV↗

arXiv2026

NTIRE 2026 The Second Challenge on Day and Night Raindrop Removal for Dual-Focused Images: Methods and Results

This paper presents an overview of the NTIRE 2026 Second Challenge on Day and Night Raindrop Removal for Dual-Focused Images. Building upon the success of the first edition, this challenge attracted a wide range of impressive solutions, all developed and evaluated on our real-world Raindrop Clarity dataset~\cite{jin2024raindrop}. For this edition, we adjust the dataset with 14,139 images for training, 407 images for validation, and 593 images for testing. The primary goal of this challenge is to establish a strong and practical benchmark for the removal of raindrops under various illumination and focus conditions. In total, 168 teams have registered for the competition, and 17 teams submitted valid final solutions and fact sheets for the testing phase. The submitted methods achieved strong performance on the Raindrop Clarity dataset, demonstrating the growing progress in this challenging task.

cs.CV↗

arXiv2026

MMRareBench: A Rare-Disease Multimodal and Multi-Image Medical Benchmark

Multimodal large language models (MLLMs) have advanced clinical tasks for common conditions, but their performance on rare diseases remains largely untested. In rare-disease scenarios, clinicians often lack prior clinical knowledge, forcing them to rely strictly on case-level evidence for clinical judgments. Existing benchmarks predominantly evaluate common-condition, single-image settings, leaving multimodal and multi-image evidence integration under rare-disease data scarcity systematically unevaluated. We introduce MMRareBench, to our knowledge the first rare-disease benchmark jointly evaluating multimodal and multi-image clinical capability across four workflow-aligned tracks: diagnosis, treatment planning, cross-image evidence alignment, and examination suggestion. The benchmark comprises 1,756 question-answer pairs with 7,958 associated medical images curated from PMC case reports, with Orphanet-anchored ontology alignment, track-specific leakage control, evidence-grounded annotations, and a two-level evaluation protocol. A systematic evaluation of 23 MLLMs reveals fragmented capability profiles and universally low treatment-planning performance, with medical-domain models trailing general-purpose MLLMs substantially on multi-image tracks despite competitive diagnostic scores. These patterns are consistent with a capacity dilution effect: medical fine-tuning can narrow the diagnostic gap but may erode the compositional multi-image capability that rare-disease evidence integration demands.

cs.CV↗

arXiv2026

Probing up-conversion electroluminescence of decoupled porphyrin molecules in a plasmonic nanocavity

Molecular triplet states can produce significant phosphorescence and act as a relay state for luminescence, such as in up-conversion processes. While this property makes triplet emitters interesting for organic light-emitting diodes (OLEDs), the study of their luminescence at the single molecule level in high resolution scanning tunneling microscopy (STM) is challenging. We investigate individual Pd-octaethylporphyrin (PdOEP) molecules decoupled from Ag(100) and Ag(111) by an ultrathin NaCl layer and observe singlet and triplet emission lines at visible wavelengths, only about 100 nm apart from each other. This is in stark contrast to the metal or free-base phthalocyanines, for which typically the lowest triplet transitions lie in the far red or infrared where the sensitivity of charge coupled device (CCD) detectors decrease significantly. The singlet S1 state of PdOEP emits photons even when the photon energy is higher than the energy provided by a tunneling electron, in an energy up-conversion process. This mechanism requires a relay (or shelving) state in which energy is stored in the molecule for the interval between tunneling electrons. Analyzing the energy levels of different molecular states (S1, D0, and T1 states) and fitting the current dependencies of S1 under up-conversion electroluminescence (UCEL) condition for S1 and T1 emission, we verify the validity of a triplet-mediated up-conversion model.

cond-mat.mes-hall↗

arXiv2026

TileLoom: Automatic Dataflow Planning for Tile-Based Languages on Spatial Dataflow Accelerators

Spatial dataflow accelerators are a promising direction for next-generation computer systems because they can reduce the memory bottlenecks of traditional von Neumann machines such as CPUs and GPUs. They organize computation around explicit, compiler-managed data movement over on-chip networks, allowing operands to be forwarded directly between processing elements and reducing reliance on high-latency, bandwidth-limited global shared memory. However, their performance depends strongly on how workloads are mapped to hardware. Naive mappings can perform poorly, and most users rely on hand-tuned vendor libraries. Thus, despite their potential for high performance, energy efficiency, and cost efficiency, limited programmability remains a major barrier to wider adoption. This paper presents TileLoom, an MLIR-based end-to-end framework that compiles tile-based programs, such as Triton kernels, onto spatial dataflow architectures. Unlike compiler frameworks that focus on optimizing code generation within a single tile, TileLoom distributes tile instances across spatially distributed cores and exploits the on-chip network and distributed memories to increase data reuse and reduce communication. TileLoom introduces a hardware representation that captures interconnect topology, memory hierarchy, and compute capabilities, enabling both architecture-specific optimizations and support for diverse spatial dataflow targets. In experiments on two generations of Tenstorrent systems, TileLoom achieves performance comparable to vendor libraries on various kernels.

cs.DC↗

arXiv2026

GLM-5V-Turbo: Toward a Native Foundation Model for Multimodal Agents

We present GLM-5V-Turbo, a step toward native foundation models for multimodal agents. As foundation models are increasingly deployed in real environments, agentic capability depends not only on language reasoning, but also on the ability to perceive, interpret, and act over heterogeneous contexts such as images, videos, webpages, documents, GUIs. GLM-5V-Turbo is built around this objective: multimodal perception is integrated as a core component of reasoning, planning, tool use, and execution, rather than as an auxiliary interface to a language model. This report summarizes the main improvements behind GLM-5V-Turbo across model design, multimodal training, reinforcement learning, toolchain expansion, and integration with agent frameworks. These developments lead to strong performance in multimodal coding, visual tool use, and framework-based agentic tasks, while preserving competitive text-only coding capability. More importantly, our development process offers practical insights for building multimodal agents, highlighting the central role of multimodal perception, hierarchical optimization, and reliable end-to-end verification.

cs.CV↗

arXiv2026

Probability of Root Cause: A Counterfactual Definition and Its Identification

Attributing an observed outcome to its root cause is a central task in domains ranging from medical diagnosis to engineering fault diagnosis. Existing approaches either equate the root cause with a root node of the causal graph, as in causal-discovery-based root cause analysis, or target causes more broadly and thereby favour proximate ones, as with the probability of causation and posterior causal effects. We argue that this issue stems from the absence of a formal definition of a root cause, which has led to methods designed for other purposes being applied to root cause attribution by default. We address this by giving a formal, individual-level definition of a root cause within the potential outcomes framework, based on the notion of an individual cause and a counterfactual root condition motivated by mediation analysis. Building on this definition, we propose the probability of root cause (PRC), which quantifies how probable it is that a candidate variable set is the root cause of a given outcome, conditional on observed evidence. Under standard assumptions, we establish the identifiability of the PRC and derive an explicit identification formula. Two numerical examples illustrate the approach.

stat.ME↗

arXiv2026

On Capacity and Delay of Wireless Networks with Node Failures

One key challenge in designing resilient large-scale wireless ad hoc networks is to understand how random node failures affect fundamental network performance. In this work, we show that both network capacity and delay scale as \scalebox{0.65}{$\textstyle Θ\left(\sqrt{\frac{n(1-q)}{\log n}}\right)$}, where $n$ is the total number of nodes and $q$ is the node failure probability. The network capacity degenerates to the classical result given by P. Gupta and P. R. Kumar when $q=0$. Based on these results, we find that even with the same number of non-faulty nodes, a network with $n$ nodes and node failure probability $q$ has lower network capacity than a failure-free network with $n(1-q)$ nodes. To compensate for the network capacity loss caused by random node failures, at least $ε(n,q) nq$ redundant nodes are required, where $ε(n,q)>1$. We further prove that the optimal trade-off between network capacity and delay remains $O(1)$ regardless of node failures, implying that high network capacity and low delay cannot be achieved simultaneously. These results demonstrate robustness against stochastic variations in wireless channels.

cs.IT↗

arXiv2026

Capacity Scalability of LEO Constellations With Dynamic Link Failures

Dynamic link failures disrupt the connectivity and geometric symmetry of the constellation structure, thereby increasing protocol overhead and degrading the effective capacity for traffic transport. The fundamental relationship between constellation size and effective capacity under protocol overhead constraints remains unclear. To this end, we define capacity scalability as the ratio of constellation capacity under non-failure conditions to protocol overhead. Specifically, if ISL states follow a two-state discrete Markov chain and the maintenance period is $k \geq 1$, the upper bound of capacity scalability under the uniform traffic pattern is $O(1/n)$, where $n$ is the number of satellites. With perfect information about the constellation topology, the upper bound can be achieved via shortest-path routing. For any given protocol, there exists an optimal constellation deployment scale in terms of capacity scalability. When the constellation size is below this optimum scale, capacity scalability increases with constellation size, thereby improving effective capacity. Increasing the maintenance period $k$ can improve capacity scalability, but it does not change the fact that the capacity scalability converges to zero when the constellation size exceeds the optimal scale.

cs.IT↗

arXiv2026

Data-Asymmetric Latent Imagination and Reranking for 3D Robotic Imitation Learning

Robotic imitation learning typically assumes access to optimal demonstrations, yet real-world data collection often yields suboptimal, exploratory, or even failed trajectories. Discarding such data wastes valuable information about environment dynamics and failure modes, which can instead be leveraged to improve decision-making. While 3D policies reduce reliance on high-quality demonstrations through strong spatial generalization, they still require large-scale data to achieve high task success. To address this, we propose DALI-R, a Data-Asymmetric Latent Imagination and Reranking framework for 3D robotic imitation learning from mixed-quality trajectories. It learns a Latent World Model over 3D point clouds for imagined rollouts and a Task Completion Scorer that reranks candidate action chunks, improving decision-making without additional high-quality demonstrations. We instantiate DALI-R with both diffusion and efficient flow-matching policies and evaluate it on Adroit and MetaWorld benchmarks. Across the two evaluated 3D base policies, DALI-R achieves an average $6.8$\% improvement in success rate while incurring less than $0.7\times$ additional inference overhead.

cs.RO↗

arXiv2026

From Time Series Analysis to Question Answering: A Survey in the LLM Era

Recently, Large Language Models (LLMs) have introduced a novel paradigm in Time Series Analysis (TSA), leveraging strong language capabilities to support tasks such as forecasting and anomaly detection. However, these analysis tasks cannot adequately cover temporal language tasks, such as interpretation and captioning. A fundamental gap remains between TSA and LLMs: LLMs are pre-trained to optimize natural language relevance for question answering rather than objectives specialized for TSA. To bridge this gap, TSA is evolving toward Time Series Question Answering (TSQA), shifting from expert-driven and task-specific analysis to user-driven and task-unified question answering. TSQA depends on flexible exploration rather than predefined TSA pipelines. In this survey, we first propose a taxonomy that reflects the evolution from TSA to TSQA, driven by a shift from external to internal alignment. We then organize existing literature into three alignment paradigms: Injective Alignment, Bridging Alignment, and Internal Alignment, and provide practical guidance for flexible, economical, and generalizable selection of alignment paradigms. We finally analyze datasets across domains and characteristics, identify challenges, and highlight future research directions.

cs.LG↗

arXiv2026

Outlier Smoothing with Closed-Form Rotations for W4A4 Large Language Model Quantization

Large Language Models (LLMs) quantization facilitates deploying LLMs in resource-limited settings, but existing methods that combine incompatible gradient optimization and quantization truncation lead to serious convergence pathology. This prolongs quantization time and degrades LLMs' task performance. Our studies confirm that Straight-Through Estimator (STE) on Stiefel manifolds introduce non-smoothness and gradient noise, obstructing optimization convergence and blocking high-fidelity quantized LLM development despite extensive training. To tackle the above limitations, we propose SingleQuant, a single-pass quantization framework that decouples from quantization truncation, thereby eliminating the above non-smoothness and gradient noise factors. Specifically, SingleQuant constructs Alignment Rotation Transformation (ART) and Uniformity Rotation Transformation (URT) targeting distinct activation outliers, where ART achieves smoothing of outlier values via closed-form optimal rotations, and URT reshapes distributions through geometric mapping. Both matrices comprise strictly formulated Givens rotations with predetermined dimensions and rotation angles, enabling promising LLMs task performance within a short time. Experimental results demonstrate SingleQuant's superiority over the selected baselines across diverse tasks on 7B-70B LLMs. To be more precise, SingleQuant enables quantized LLMs to achieve higher task performance while necessitating less time for quantization. For example, when quantizing LLaMA-2-13B, SingleQuant achieves 1,400$\times$ quantization speedup and increases +0.57\% average task performance compared to the selected best baseline.

cs.LG↗

arXiv2026

ConsisVLA-4D: Advancing Spatiotemporal Consistency in Efficient 3D-Perception and 4D-Reasoning for Robotic Manipulation

Current Vision-Language-Action (VLA) models primarily focus on mapping 2D observations to actions, but exhibit notable limitations in spatiotemporal perception and reasoning: 1) spatial representations often rely on additional sensors, introducing substantial computational overhead; 2) visual reasoning is typically limited to future-frame prediction, lacking alignment with the instruction-grounded scene and thus compromising spatiotemporal consistency. To address these challenges, we propose ConsisVLA-4D, a unified and efficient framework that enhances spatiotemporal consistency in 3D perception and 4D reasoning. Specifically, we design: 1) CV-Aligner, which ensures cross-view object semantic consistency by filtering instruction-relevant regions and aligning object identities across multiple viewpoints; 2) CO-Fuser, which guarantees cross-object spatial geometric consistency by eliminating spatial relation ambiguities between objects across views using compact latent representations. Building upon these, we introduce 3) CS-Thinker to achieve cross-scene spatiotemporal consistency as actions unfold. It learns implicit knowledge of local dynamics from object-semantic tokens of CV-Aligner and global depth from geometric tokens of CO-Fuser, thereby enhancing efficient visual reasoning under scene variations. Extensive experiments demonstrate that, benefiting from its efficient spatiotemporal consistency design, ConsisVLA-4D achieves 21.6% and 41.5% performance improvements, along with 2.3-fold and 2.4-fold inference speedups compared to OpenVLA on the LIBERO benchmark and real-world platforms, respectively.ConsisVLA-4D is open-sourced and publicly available at

cs.RO↗

arXiv2026

Only Say What You Know: Calibration-Aware Generation for Long-Form Factuality

Large Reasoning Models achieve strong performance on complex tasks but remain prone to hallucinations, particularly in long-form generation where errors compound across reasoning steps. Existing approaches to improving factuality, including abstention and factuality-driven optimization, follow a \emph{coupled exploration-commitment} paradigm, in which intermediate reasoning is unconditionally propagated to the final output, limiting fine-grained control over information selection and integration. In this paper, we propose an \textbf{Exploration-Commitment Decoupling} paradigm that disentangles knowledge exploration from final commitment, enabling models to explore with awareness while answering cautiously. We instantiate the paradigm with \textbf{Calibration-Aware Generation (CAG)}, a framework that equips models with end-to-end, calibration-aware generation capabilities, by augmenting intermediate reasoning with calibrated reliability estimates and prioritizing reliable content in final outputs. Across five long-form factuality benchmarks and multiple model families, CAG improves factuality by up to 13%, while reducing decoding time by up to 37%. Overall, our work highlights decoupling as a principled approach for more reliable long-form generation, offering directions for trustworthy and self-aware generative systems.

cs.CL↗

arXiv2026

Probe of Generic Quantum Contextuality and Nonlocal Resources for Qubits

We reveal that the entropic uncertainty relation with a quantum memory is able to intrinsically connect local generic contextuality addressed in the pioneering work by Spekkens and nonlocal quantum resources such as entanglement and Bell nonlocality. Based on the constructed optimal set for any given single-qubit state, we prove rigorously a faithful criterion to witness the generic contextuality in the scenario of local quantum state preparation. Furthermore, within the framework of quantum resource distribution, it is proved that there exist quantitative trade-off relations between local preparation contextuality and bipartite entanglement or Bell nonlocality in a shared quantum system, which are captured by two inequalities where the local and nonlocal quantum resources can coexist. The faithful criterion and quantitative inequalities are all experimentally testable, which are verified through two independent well-designed experiments on the Quafu quantum cloud platform.

quant-ph↗

arXiv2026

Prefill-Time Intervention for Mitigating Hallucination in Large Vision-Language Models

Large Vision-Language Models (LVLMs) have achieved remarkable progress in visual-textual understanding, yet their reliability is critically undermined by hallucinations, i.e., the generation of factually incorrect or inconsistent responses. While recent studies using steering vectors demonstrated promise in reducing hallucinations, a notable challenge remains: they inadvertently amplify the severity of residual hallucinations. We attribute this to their exclusive focus on the decoding stage, where errors accumulate autoregressively and progressively worsen subsequent hallucinatory outputs. To address this, we propose Prefill-Time Intervention (PTI), a novel steering paradigm that intervenes only once during the prefill stage, enhancing the initial Key-Value (KV) cache before error accumulation occurs. Specifically, PTI is modality-aware, deriving distinct directions for visual and textual representations. This intervention is decoupled to steer keys toward visually-grounded objects and values to filter background noise, correcting hallucination-prone representations at their source. Extensive experiments demonstrate PTI's significant performance in mitigating hallucinations and its generalizability across diverse decoding strategies, LVLMs, and benchmarks. Moreover, PTI is orthogonal to existing decoding-stage methods, enabling plug-and-play integration and further boosting performance. Code is available at: https://github.com/huaiyi66/PTI.

cs.CV↗

arXiv2026

SVOM/VT: Flight Model Verification and Pre-launch Testing

This paper presents pre-launch testing and calibration results for the SVOM/VT (Space-based Variable Objects Monitor, Visible Telescope) Flight Model (FM), validating its performance under simulated space conditions through thermal vacuum cycling, energy concentration analysis, stray light suppression, and CCD/electronics calibrations (gain, noise, quantum efficiency). The results confirm full compliance with design requirements: stray light suppression achieves point-source transmittance $<10^{-7}$ at $30^\circ$ off-axis, thermal control maintains stable CCD temperatures ($-75^\circ$C for the red channel, $-65^\circ$C for the blue channel), and detection sensitivity meets the limiting magnitude of 22.50 (SNR $>$ 3 with 300 seconds exposure). Early in-orbit tests further validate performance, yielding limiting magnitudes of 22.70 (V-band, red) and 22.78 (blue), consistent with pre-launch specifications.

astro-ph.IM↗