arXiv subjects

Da Li

Publications and source records attributed to Da Li.

At least 19 recordsLinked to original sources

arXiv2026

LingGS: Decoupling Feed-Forward Geometry Prediction from Camera Tracking for High-Fidelity Gaussian SLAM

Achieving high-quality reconstruction and responsive camera tracking remains challenging for monocular RGB Gaussian splatting SLAM under a limited optimization budget. We present an online system that addresses this challenge by combining learned geometric priors with explicit geometric camera tracking, which can substantially outperform learned camera prediction in terms of tracking accuracy. A frozen streaming 3D foundation model serves as a virtual RGB-D sensor, providing depth estimation from monocular video. An explicit geometric solver then estimates camera motion through feature reprojection, and a streaming Gaussian mapper integrates the resulting observations into a dense scene representation. We further show that geometric tracking improves camera pose accuracy across representative streaming geometry backbones, demonstrating the broader applicability of our approach. Extensive experiments on indoor and outdoor datasets show that our method significantly improves reconstruction quality over existing approaches, achieving about 7 FPS on a single RTX 4090 and delivering 2-3 times the throughput of the evaluated prior SLAM systems.

cs.CV↗

arXiv2026

SplatScout: Query-Driven Part-Level Referring Segmentation in Gaussian Scenes

Existing referring segmentation methods on Gaussian scenes primarily target whole objects, relying on segmenter-generated candidates from which a vision-language model (VLM) selects. Such predetermined fixed candidates may omit fine-grained parts needed as targets or cues for identifying objects. We introduce SplatScout, a query-driven, closed-loop approach that places perception under VLM control. The model actively selects views, directs segmentation, and inspects rendered 3D results to refine its predictions. For part-level evaluation, we introduce PartRef-ScanNet++, comprising 261 manually verified queries across 12 ScanNet++ scenes with mesh-derived ground truth. On PartRef-ScanNet++, SplatScout achieves 0.538 mIoU, substantially outperforming existing methods. Ablations demonstrate the benefits of iterative refinement, while comparisons across VLM backbones highlight the importance of effective tool-feedback utilization. SplatScout also achieves higher scores on four existing referring benchmarks. Code and the PartRef-ScanNet++ benchmark are available at https://github.com/kaust-violab/SplatScout.

cs.CV↗

arXiv2026

VISTA: Differentiable Viewpoint Planning for Long-Horizon Task Visualization

Long-horizon object-centered tasks in interactive 3D environments involve heterogeneous events that unfold across time and space, including fine-grained manipulations, larger structural motions, and subtle state transitions. Effectively visualizing such processes is essential for helping users understand how actions lead to outcomes over time. However, existing camera control methods in 3D systems are commonly designed for static objects, and often fail to maintain the legibility of task-relevant components and their evolving relationships throughout long-horizon executions. We present VISTA, a differentiable viewpoint planning framework for long-horizon task visualization. We formulate camera control as a visualization policy that optimizes camera parameters under a unified rendering-based quality objective and we propose five quality metrics to satisfy our five design goals. To support both local part inspection and interaction-centric visualization centered on a primary object, our formulation handles targets defined either as object--part pairs or as relational tuples involving another interacting object. This allows the planner to reveal not only what part is important, but also how it functionally relates to surrounding structures during task execution. We demonstrate that the same planner supports two complementary use cases: presentation-driven narrative visualization for communicating how long-horizon tasks unfold to human viewers, and agent-driven active observation within an observe-act loop. Together, these results position differentiable viewpoint planning as a general mechanism for ``how-things-work'' visualization of long-horizon object-centered interactions in 3D environments.

cs.GR↗

arXiv2026

Dynamical coarse-graining of neural-network training

Dynamical coarse-graining of neural networks gives macroscopic observables and enables training to be predicted, and controlled in a state space far smaller than the parameter space. In this paper, we first define the learning operator following the training law of neural networks and establish a descent theorem for the learning operator, which serves to determine whether the operator is sufficient to characterize the learning dynamics. Then the exact closure theorem and the dynamic coarse-graining theorem are presented in sequence. The three theorems established in this paper provide a systematic framework for constructing and assessing dynamical coarse-grained models of neural-network training. Finally, we demonstrate the applicability of the theorems with a finite-width nonlinear ReLU network. Its dynamic coarse-grained model for a fixed pattern is exactly derived and composed of five independent moments. The model exactly predict the training outputs and evolving learning operator. The framework may also guide the development of approximate coarse-grained models for more general neural networks when exact closure is unavailable.

cs.LG↗

arXiv2026

DELA: Learning Controllable Edits for Diffusion Language Models

Diffusion language models (DLMs) support bidirectional denoising and iterative refinement, but generation remains largely controlled by token-level masking schedules. This leaves draft-level refinement implicit and limits explicit control over when to insert missing content, delete redundant spans, revisit uncertain regions, or stop editing. We introduce Diffusion Editing Language Adapter (DELA), a plug-in framework that enables controllable draft refinement with a frozen DLM. DELA operates on clean, variable-length drafts using four explicit editing actions: Insert, Delete, Remask, and Stop. The controller determines when and where to edit the draft, while the frozen DLM fills masked regions through denoising, separating high-level edit control from token-level generation. A naive single-phase editor would be computationally expensive because arbitrary interior edits can disrupt prefix stability and limit reliable KV-cache reuse. To address this, DELA uses a two-phase decoding procedure: a fast, append-only drafting phase that supports KV-cache reuse, followed by a global refinement phase that enables targeted non-monotonic edits to the full draft. We train the controller using a staged curriculum that begins with learning local edit priors from synthetic recovery tasks and mined actions that improve draft quality. Subsequent stages refine edit selection through rollout-based outcome preferences and alignment with downstream task feedback. Experiments on mathematical reasoning and code generation show that DELA achieves gains of up to 5.5 percentage points over direct diffusion decoding while keeping the backbone frozen. DELA thus provides a modular approach to explicit draft-level control in diffusion language models.

cs.AI↗

arXiv2026

NormLift: From Lifted Features To Semantic Reliability In 3D Gaussian Splatting

Training-free weighted aggregation is widely used to lift 2D semantic features onto 3D Gaussians for open-vocabulary scene understanding, yet its theoretical role remains insufficiently understood. Existing analyses typically justify this operation from the rendering side, treating Gaussian features as linearly composable Euclidean variables for reconstructing 2D feature maps. However, this view does not match downstream 3D usage, where each Gaussian is often queried independently in a cosine-based embedding space. We revisit feature lifting from the 3D side and formulate per-Gaussian assignment as a cosine alignment problem on the CLIP unit sphere. Under this objective, the $\ell_2$-normalized semantic back-projected feature emerges as the closed-form solution, providing a complementary interpretation of the standard lifting rule from the perspective of per-Gaussian semantic assignment. This solution is still a convex combination of the observed directions, so every lifted feature is inherently a mixture. Such mixing is unavoidable for any per-Gaussian estimator, and the question is whether it can be measured. The same formulation answers this: the norm of the back-projected feature decomposes into intra-view and inter-view consistency, so the feature magnitude quantifies how mixed the observations are and serves as a semantic reliability signal. Calibrated by effective multi-view support, this reliability score identifies the Gaussians whose lifted direction has drifted and guides a mode-voting refinement that corrects them by copying a single observed CLIP direction instead of averaging. Experiments on open-vocabulary 3D semantic segmentation show that NormLift is an efficient, training-free framework that achieves strong performance across evaluation protocols. Our code is available at https://github.com/kaust-violab/NormLift_release

cs.CV↗

arXiv2026

DIFFLET: A Fast Inference Engine for Diffusion Transformer Models on AWS Trainium

On AWS Trainium2, every component of a diffusion model runs as a program compiled ahead of time for fixed shapes, while pipelines differ across models, prompt lengths across requests, and caching decisions across denoising steps. We present Difflet, a serving engine in which five model pipelines share one compilation backend, joint attention is split across cores by attention head, and a host controller decides at each denoising step whether to run the denoiser or reuse its earlier output. A small probe program computes the caching signal on the device and returns one number per step, and a resident model process keeps every program loaded across requests. Across five image and video model families, fused attention accelerates denoiser steps by 1.2-4.3x over the same framework using PyTorch/XLA attention, and at 32.8k visual tokens it is the only attention that compiles. On four of the five models, resident serving is 1.5-3.4x faster than standalone runs that reload weights from a warm file cache for each generation.

cs.DC↗

arXiv2026

Beyond Forced Modality Balance: Intrinsic Information Budgets for Multimodal Learning

Multimodal models often converge to a dominant-modality solution, in which a stronger, faster-converging modality overshadows weaker ones. This modality imbalance causes suboptimal performance. Existing methods attempt to balance different modalities by reweighting gradients or losses. However, they overlook the fact that each modality has finite information capacity. In this work, we propose IIBalance, a multimodal learning framework that aligns the modality contributions with Intrinsic Information Budgets (IIB). We propose a task-grounded estimator of each modality's IIB, transforming its capacity into a global prior over modality contributions. Anchored by the highest-budget modality, we design a prototype-based relative alignment mechanism that corrects semantic drift only when weaker modalities deviate from their budgeted potential, rather than forcing imitation. During inference, we propose a probabilistic gating module that integrates the global budgets with sample-level uncertainty to generate calibrated fusion weights. Experiments on three representative benchmarks demonstrate that IIBalance consistently outperforms state-of-the-art balancing methods and achieves better utilization of complementary modality cues. Our code is available at: https://github.com/XiongZechang/IIBalance.

cs.MM↗

arXiv2026

Stochastic MeanFlow Policies: One-Step Generative Control with Entropic Mirror Descent

Combining entropy regularisation with mirror descent in online off-policy reinforcement learning requires expressive policies with efficient sampling and explicit entropy control. Gaussian policies struggle to represent multimodal targets, whereas generative policies often require iterative sampling or lack tractable entropy estimates. We introduce Stochastic MeanFlow Policies (SMFP), a one-step generative policy class that combines MeanFlow-based noise-to-action mappings with Gaussian reparameterisation. The latent mapping captures diverse action modes, while a learnable conditional Gaussian scale provides an analytic entropy lower bound. We use this bound to regulate stochasticity alongside direct value guidance and advantage-weighted MeanFlow regression, enabling entropy-regularised mirror-descent learning with one-step policies. Across seven MuJoCo benchmarks, SMFP achieves strong performance against Gaussian and generative baselines with single-step inference. Additional experiments demonstrate multimodal action modelling and effective exploration in sparse-reward navigation.

cs.LG↗

arXiv2026

Randomization Beyond Deterministic Adaptivity in Parallel Sampling

We study parallel samplers that reveal each coordinate once and draw the coordinates of each batch independently from their exact conditional marginals. Under the same sampling interface and a hard three-round cap, randomization can improve accuracy beyond every deterministic policy that adapts its batches to previously observed values. For an explicit full-support binary family, the optimal deterministic adaptive forward Kullback-Leibler divergence grows linearly with the dimension, while a uniform mixture of four fixed schedules has exponentially small divergence. Only two independent fair bits are needed to select the schedule. We determine the asymptotic additive constant in the deterministic optimum and show that its optimal total variation tends to one quarter. An outward-rounded finite-sum certificate establishes divergence separation at dimension 389. The mechanism combines an irreversible first-batch dependence loss with posterior concentration in the randomized components. A density-coverage lemma describes how their mixture restores the target. A complementary noisy-matching family exhibits a growing round gap for nonadaptive schedules, although uniform random coordinate order still has linear mixture error. These results concern known finite structures and exact conditional marginals.

cs.IT↗

arXiv2026

Geometry-Preserving Human-to-Robot Upper-Body Motion Retargeting from Monocular Video

Monocular RGB video provides an accessible source of human demonstrations for upper-body robot motion, yet video-driven human-to-robot transfer remains challenging because body and hand motion are recovered at different spatial scales, human and robot kinematics differ substantially, and fine distal motion is difficult to preserve across embodiments. We present a geometry-preserving motion-retargeting framework that integrates unified body--hand reconstruction with morphology-independent geometric transfer. Frame-wise body estimates, video-level observations, and detailed hand evidence jointly constrain a single differentiable Momentum Human Rig (MHR) state, while transient hand artifacts are repaired in parameter space. The reconstructed motion is represented by arm-segment directions, elbow configuration, relative palm orientation, and bilateral wrist relations, and is realized on the target robot through multi-stage inverse kinematics and robot-specific hand adaptation. Within the broader system, Across-VAM provides video generation, whereas Across-WAM performs human-to-robot motion mapping. The method is evaluated on 16 monocular videos comprising 1,769 source frames, including 10 signing and six reach-to-grasp sequences. Unified reconstruction reduces mean hand reprojection error from 22.36 to 7.21 pixels relative to SAM 3D Body. All 16 retargeted trajectories completed kinematic simulation playback, and representative signing and reach-to-grasp motions were further demonstrated on a physical robot. The results demonstrate a unified pipeline from monocular human video to coordinated upper-body motion on a dual-arm dexterous robot.

cs.RO↗

arXiv2026

LieDiscover: Adaptive Symbolic Library Construction for Explicit Open-form Symmetry Discovery

Discovering underlying symmetries from data has emerged as a crucial challenge in scientific discovery. Existing data-driven methods for symmetry discovery fail to determine the exact number and mathematical form of unknown infinitesimal generators. Recent explicit methods represent generators using a predefined function library and identify them through algebraic optimization, but they often struggle to capture complex symmetries involving high-order polynomials or transcendental functions. To address this limitation, we formulate symmetry discovery as a joint optimization problem over the function library and coefficients. We propose a novel framework that leverages an encoder-decoder architecture to dynamically generate symbolic expressions and expand the library. This generation process is optimized via reinforcement learning, which accelerates the exploration of the symbolic search space through step-wise rewards. Experiments demonstrate that LieDiscover can successfully uncover open-form infinitesimal generators involving high-order polynomials or transcendental functions, which remain intractable for existing methods. The discovered symmetries also improve performance in downstream PDE solving and discovery tasks.

cs.SC↗

arXiv2026

GeoBalance: Geometry-Aware Monitoring and Reconstruction with Asymmetric Optimization for Balanced Multimodal Learning

Multimodal classifiers can converge to modality-dominant solutions in which one modality dominates the joint prediction, suppressing the learning of others. Existing balancing methods mainly adjust losses, gradients, or modality contributions, largely treating modality imbalance as an optimization problem while implicitly treating the weak modality as under-optimized but representationally intact. In this work, we find that this assumption does not always hold, as persistent modality dominance can induce a representation-level collapse of the weak modality, which we term \emph{manifold modality collapse} (MMC). MMC manifests as a coupled geometric degradation in which weak-modality representations collapse onto fewer directions within each class and become less separable across classes. Motivated by this observation, we propose \emph{GeoBalance}, a geometry-aware framework that monitors these two geometric properties and reconstructs the weak modality representation only when it exhibits signs of MMC. Once triggered, GeoBalance uses a fixed Simplex-ETF class scaffold and spectral regularization to restore class separation while preventing collapse onto a few feature directions. To preserve reconstruction during joint training, asymmetric gradient projection removes the joint-gradient component conflicting with reconstruction, leaving non-conflicting optimization unchanged. Extensive experiments across six multimodal benchmarks demonstrate great improvements over competitive balancing methods, validating its effectiveness.

cs.CV↗

arXiv2026

One Skill Does Not Fit All: Automatic Discovery and Taxonomy-Guided Routing of Frame-Selection Skills for Long-Video Question Answering

Long-Video Question Answering (LVQA) requires locating decisive evidence in hour-scale videos under a limited frame budget. Most training-free methods apply the same frame-selection strategy to all questions, despite substantial variation in the evidence required by different question types. Our analysis shows that the relative effectiveness of frame-selection strategies varies across semantic categories and benchmarks, motivating adaptive evidence acquisition. In this paper, we introduce AutoSkill, a source-supervised framework for automatically discovering and routing executable frame-selection skills. Starting from a small labelled source pool, LLM agents iteratively propose, implement, evaluate, and refine candidate skills. For a target benchmark, AutoSkill uses only unlabelled question and option text to induce a shared semantic taxonomy, rewrite labelled source examples into the target style, and estimate a category-to-skill mapping. Neither target videos nor target answers are used in this process. At inference time, each question is assigned one skill, which selects the frames used in a single inference of the frozen video MLLM. Across five long-video benchmark splits, AutoSkill improves Qwen2.5-VL-7B and Qwen3.5-4B by 2.4% and 1.2%, respectively, demonstrating the effectiveness of our AutoSkill.

cs.CV↗

arXiv2026

Finite-Modal Realization and Operator-Norm Convergence of a Source-to-Observation Electromagnetic Scattering Green Operator

Source-to-observation operators provide reusable environment-level descriptions for multi-query electromagnetic (EM) prediction and communication-mode analysis. However, in practical multiple-scattering models, these operators are represented with finitely many angular modes, and agreement for selected excitations or between successive truncation orders does not establish uniform accuracy of the full map or reliability of its singular channels. To close this gap, we formulate the environment-induced response as a scattering Green operator on fixed continuous source and observation spaces and derive an exact trace-space factorization that reconstructs the Maxwell scattered field. For fixed, pairwise-disjoint enclosing trace spheres and a well-posed collective problem, nested vector spherical wave function (VSWF) realizations converge in operator norm. A structural bound separates external modal tails from collective-resolvent sensitivity, and operator-norm convergence guarantees uniform convergence of the singular values. We further construct a finite metric core that preserves the nonzero singular values of each finite-order operator and reconstructs matched orthonormal source--field channels without introducing external-support discretization degrees of freedom (DoF) into the spectral problem. Full-wave benchmarks verify the finite-order implementation. A controlled near-resonant two-sphere study shows that adjacent-order agreement can precede resolution of the dominant high-order collective direction. It further shows that only part of the internal amplification appears in externally accessible gains and that resonance promotes a distinct high-order channel pair above an otherwise preserved low-order family. The resulting framework provides a convergent, metric-consistent finite-modal representation of multiple-scattering source-to-observation operators and their accessible channels.

cs.IT↗

arXiv2026

Hierarchical Possession-Aware Graph Pointer Network for Pass Receiver Selection

Pass receiver selection is a fundamental task in football analytics, aiming to predict the intended receiver under a given game state. This task is challenging with event-centered freeze-frame observations, a broadcast-like setting that provides only partial and variable player visibility without complete trajectories or stable player identities. The model must therefore reason over anonymous visible candidates, opponent pressure, and recent context under partial observation. To address this setting, we propose a Hierarchical Possession-aware Graph Pointer Network (HPGPN), which formulates pass receiver selection as variable-size candidate prediction over visible teammates. HPGPN jointly models current player interactions, local event context, and possession-level temporal dynamics. It represents the current pass situation with a graph, incorporates fixed event context, and uses dynamic possession history to capture how the attacking sequence evolves. Candidate representations are refined hierarchically by integrating spatial, contextual, and historical evidence, and a glimpse pointer head scores the receiver candidates. Experiments on public football event and freeze-frame data show that HPGPN improves pass receiver selection performance. Ablation studies demonstrate the effectiveness of graph-based interaction modeling, fixed event context, and dual-branch dynamic possession-history modeling.

cs.AI↗

arXiv2026

Can LVLMs Uncover the Truth Behind Visual Illusions? An Analysis of Perceptual and Reasoning Capabilities

Large Vision Language Models have integrated reasoning capabilities, elevating cognitive performance to new levels. However, existing evaluations either focus solely on perception or rely on specific domains such as maths or coding. Evaluation for reasoning capabilities that align with an open-world environment is still required, especially one that considers perception and reasoning jointly. To bridge this gap, we propose to evaluate LVLMs by exploiting visual illusions as a diagnostic tool. Visual illusions are phenomena in which the human visual system misinterprets objective signals, resulting in an understanding that deviates from reality. We constructed Illusion-Reasoning, a benchmark of illusion images collected from the real world, incorporating diverse annotated question-answer pairs. Based on Illusion-Reasoning, we show that the reasoning capabilities of a wide range of LVLMs are not as advanced as claimed. Our work provides new insights into LVLMs and offers future directions for optimisation. Our project is publicly available at https://github.com/zhaoliangjie55/EMNLP2026_Illusion.

cs.CL↗

arXiv2026

PushDualGen: Enabling LLMs to Generate Semantic IDs with Interpretable Copy for Industrial Push Recommendation

Push recommendation in KuaiShou proactively delivers personalized content to nearly one billion users to facilitate their engagement. Recently, generative recommendation has achieved end-to-end user personalization through semantic ID. However, their black- box characteristics make recommendation logics difficult to trace, hindering their deployment. OneRec-Thinking addresses this by incorporating CoT before generating SIDs, but this significantly increases inference cost. To support large-scale industrial applications, we propose PushDualGen, a lightweight generator, which first generates the SID and then produces a copy as a skippable explanation. PushDualGen has been deployed in Kuaishou's push recommendation system. Online A/B tests demonstrate the effectiveness of PushDualGen, delivering significant improvements in both user attraction and satisfaction. The effective play rate for videos recommended to users has relatively increased by 8.50%, while the dissatisfaction rate has relatively fallen by 37.70%. In the long term, PushDualGen optimises the content ecosystem, providing more exposure for long-tail videos.

cs.IR↗