The paper explores a new lossy strategy for speculative decoding called verifier skipping, which allows certain draft prefixes to be committed without invoking the verifier. It investigates which confidence signals—raw token confidence, learned marginal survival, or learned conditional survival—best schedule these skips, finding that raw confidence often performs best despite higher positionwise AUROC for learned signals. Experiments on HumanEval with DiffuCoder-7B-Instruct and Qwen3-32B show that all three signals can reduce verifier calls by 9.6% to 13.5% while maintaining the same pass@1 as Strict SDD.
By Haoxuan Luo, Jameson Sandler, Ferdinando Fioretto
SCOPE-4D is an endoscopic 4D geometry foundation model that predicts camera parameters, dense geometry, and 3D tissue trajectories from monocular RGB video in a single forward pass. The authors introduce SCOPE-5K, a curated dataset of about 5,000 real and synthetic gastrointestinal endoscopy and laparoscopy clips, and use it for geometric supervised fine‑tuning. Adding Common–Residual Motion (CRM) constraints and trajectory supervision further improves camera and depth estimation and enables dense 3D tissue tracking, as shown by evaluations on public and new benchmarks and a blinded user study.
By Chaoyi Zhou, Zhongpai Gao, Anwesa Choudhuri, Meng Zheng, Benjamin Planche, Run Wang, Terrence Chen, Siyu Huang, Ziyan Wu
The paper presents ViTok, a multi‑teacher distillation approach that combines SigLIP2 and DINOv3‑L to jointly preserve global recognition and dense semantics. By introducing split adaptor heads, asymmetric losses, teacher reweighting, masked image modeling, and PHI‑S feature balancing, the authors achieve higher ImageNet‑1K kNN accuracy and restore ADE20K segmentation performance to match the teacher. The study also reports negative findings, such as limited benefits from scaling to ImageNet22K and interference from additional teachers.
By Hailun Xu, Kanchan Sarkar
The paper introduces “SpaceConflict”, a benchmark of 23,196 multimodal inputs designed to test whether large language models can not only report spatial facts but also use them in reasoning. Experiments show a gap between a model’s ability to recover an initial spatial state from visual evidence and its ability to apply that state to perform transformations, with the gap narrowing but not closing as model size increases. To address this, the authors propose Operational State Supervision (OSS), which supervises task‑relevant spatial states and their transformation trajectories, improving accuracy on judgments that require state organization and use.
By Jinchang Zhang, Guoyu Lu
The paper introduces Behavior Pack Optimization (BPO), a post‑training method for video multimodal large language models that replaces single‑response rewards with a set of outputs across counterfactual views. BPO enforces stability when interventions are irrelevant, sensitivity when key evidence is removed, and abstention when no evidence remains, using an anchor‑relative advantage to keep the objective stable with small pack sizes. Experiments on datasets such as TempCompass, MVBench, and NExT‑QA show that BPO improves macro accuracy and abstention metrics for models like Qwen2.5‑VL‑7B‑Instruct, with gains that transfer to other benchmarks and models.
By Zhaolu Kang, Shiyu Liu, Tailong Luo, Wei Zhang, Yingjie He, Lei Wei, Guansu Wang, Liang He, Siheng Wang, Guangyuan Dong, Jiaqi Su, Shuang Chen, Haoyu Ji, Qishi Zhan, Kaiyue Zhou
The paper introduces a benchmark for spatially grounded gesture generation, featuring about 2,000 pointing-annotated clips from naturalistic VR dialogue with ground-truth 3D referents. It defines a task where systems must decide when, how, and where to point within conversational speech, and proposes a protocol that separates temporal alignment, spatial grounding, and perceived naturalness. A flow‑matching baseline (MM‑Conv‑Flow) and a retrieval‑based system are evaluated against captured human motion, revealing that geometric grounding can surpass human pointing quality without improving perceived naturalness, underscoring the need to measure referential gesture quality along distinct dimensions.
By Anna Deichler, Rishabh Dabral, Fethiye Irmak Dogan, Anindita Ghosh, Jonas Beskow
The paper introduces VDOT++, a unified distillation framework for few-step video generation across text-to-video, image-to-video, and condition-based tasks. It extends the previous VDOT method by incorporating an asymmetric unbalanced optimal transport distillation that allows student tokens to carry less mass, and an α1 ground cost that uses a weighted median to preserve modes. The approach combines distribution matching with adversarial refinement and cross-scale distillation, achieving competitive performance on multiple video benchmarks with only four sampling steps.
By Yutong Wang, Xingtong Ge, Enhuai Liu, Yunke Wang, Tianfan Xue, Yu Qiao, Yaohui Wang, Xinyuan Chen, Chang Xu
PocketSplat is a feed‑forward framework that builds mobile Gaussian assets within a specified budget. It organizes dense, geometry‑aware latent candidates in world space, allocates integer capacity to local latent cells, and decodes Gaussian attributes only for retained candidates. The method uses cell‑conditioned latent fusion and spatial responsibility decoding to aggregate multi‑view evidence and adapt Gaussian support, achieving a strong quality–budget trade‑off on DL3DV and out‑of‑distribution benchmarks, and running directly on an iPhone faster than streamed MVSplat variants.
By Wenzhi Guo, Xianda Chen, Dongxuan Chen, Guangchi Fang, Bing Wang
The paper introduces SalTempto, a new video saliency benchmark featuring 224 highly dynamic clips from HACS‑Segments, each with event context and gaze data from up to 16 subjects. It shows that static baselines perform poorly on SalTempto compared to LEDOV, while fine‑tuned temporal models achieve significant gains, yet still leave a large portion of explainable gaze information unaccounted for. The dataset highlights the need for better temporal modeling in video saliency research.
By Susmit Agrawal, Rebecca Wanner, Juliane Verwiebe, Matthias Tangemann, Matthias Bethge, Matthias K\"ummerer
The paper introduces DyCPO, a co‑evolutionary framework that jointly optimizes token selection and adaptive counterfactual intervention for video reasoning. It builds a multi‑role dependence metric to balance visual exploration with answer‑relevance mining, suppressing filler tokens and spurious visual noise. By deriving counterfactual signals from the model’s own successful and failed rollouts, DyCPO enables self‑diagnostic analysis and co‑evolution of the optimization objective with the policy, leading to consistent performance gains on complex video reasoning benchmarks.
By Yudong Han, Yong Wang, Zaiquan Yang, Liang Lin, Chongyang Tao, Xiangxiang Chu, Liyuan Pan
The paper introduces Event-guided Neural Video Codec (ENVC), a neural video compression method that incorporates event streams—capturing brightness changes between frames—into both motion and frame coding stages. By using event-guided motion priors and event-conditioned predictors, ENVC improves RGB compression efficiency, achieving significant BD-rate savings across six benchmarks. The authors also synthesize paired RGB-event data for training and demonstrate that the gains persist on large-motion sequences, highlighting events as a valuable complementary modality for video coding.
By Jiyun Kong, Jungwoo Kim, Enes Eray Demirtas, Touradj Ebrahimi, Jong-Seok Lee
GeoScaffold introduces a geometric supervision framework that embeds 3D geometry into vision‑language navigation policies during training, eliminating the need for depth sensors or 3D encoders at inference. The method learns a compact depth tokenizer from training trajectories, then fine‑tunes the policy with geometry query tokens that reconstruct depth, connectivity, and traversability, turning these states into compact geometric latents for action decoding. After training, the tokenizer, generators, and reconstruction heads are discarded, leaving a lightweight backbone that outperforms vision‑only navigators on continuous VLN benchmarks.
By Yixuan Jiang, Wentong Li, An Liu, Zihao Xin, Fulin Tang, Cong Leng, Yang Gao, Jian Cheng
PatchScene is a diffusion-based framework that completes large-scale LiDAR scenes using a patch-based voxel diffusion paradigm, generating fine-grained geometry in localized 3D regions. It incorporates a confidence-guided spatio-temporal fusion mechanism to merge overlapping patches and adjacent frames, and introduces an Annular-Flow diffusion strategy that propagates high-fidelity information from near to far ranges. Experiments on SemanticKITTI show state-of-the-art performance, with the model trained on 20 m LiDAR ranges generalizing to 50 m scenes without retraining.
By Qingdong Xu, Jiajun Zhu, Shilin Zhu, Xinjing He, Chao Lu, Huanran Wang, Jiyao Zhang
arXiv:2610.02260v1 Announce Type: new
Abstract: Flow matching models excel at generative modeling, and many downstream applications require their samples to satisfy prescribed constraints, such as ob...
By Yesom Park, Kelvin Kan, Qifan Chen, Thomas Flynn, Hayden Schaeffer. Xihaier Luo
SoTa is a low‑cost, capacitive tactile skin that offers full‑hand coverage for both humans and robots, featuring a shared layout of 202 taxels across finger and palm regions. The multilayer design with fabric electrodes allows in‑house fabrication for under $10 in materials, and the sensor maintains over 97% of its response after 10,000 loading‑unloading cycles and 1,280 tight‑fist folds. Using SoTa, human‑robot co‑training improves in‑distribution success on three contact‑rich manipulation tasks, and adding human demonstrations more than doubles mean success across eight evaluation conditions, from 22.8% to 45.9%, including all out‑of‑distribution scenarios.
By Jingyun Yang, Baiyu Shi, Timothy Yu, Haitian Liu, Alberta Longhini, Weichen Wang, Rika Antonova, Zhenan Bao, Jeannette Bohg
The paper introduces SD‑DPC, a method that learns sparse, interpretable feedback policies for nonlinear systems using data. It first identifies a prediction model via rollout‑based sparse identification of nonlinear dynamics (SINDy), then parameterizes the policy as a sparse combination of dictionary functions and trains it by differentiating a constrained finite‑horizon predictive‑control objective. The resulting explicit feedback law contains only a few terms, satisfies constraints across three benchmark problems, outperforms a distilled policy by up to an order of magnitude, and requires far less memory and online computation than an optimization benchmark while providing explicit sensitivity bounds.
By Ali Reza Daneshvar Garmroodi, Jan Drgo\v{n}a
JOVE is an online framework that jointly assigns large language model (LLM) executors and selects intermediate outputs for paid verification in directed acyclic task graphs. It balances immediate execution costs with future learning by solving per-query mixed‑integer linear programs and updating task‑dependent LLM quality estimates through verification feedback. Experiments on four reasoning benchmarks show that JOVE matches standard inference accuracy while cutting average cost and latency by at least 3.17×.
By Haoran Zhang, Dongjun Kim, Seohyeon Cha, Kevin S Chan, Ananthram Swami, Gustavo De Veciana, Haris Vikalo
This thesis introduces methods for assessing and enhancing the robustness of large language models (LLMs) against adversarial input variations. It proposes a generative robustness metric, R_stab(f), based on Jensen‑Shannon divergence, and proves bounds for localized attacks. The work also presents adaptive attacks, Trojan detection techniques, defense strategies for multi‑layer systems, and lightweight attestation for Model Context Protocol (MCP) agentic systems, all implemented in the JudgeGuard, TrojanArmor, and MCPSec suites.
By Narek Maloyan
The paper tackles the challenge of compressing diffusion language models (dLLMs) by addressing the mismatch between calibration on clean activations and inference on partially masked states. It introduces a trajectory‑aware low‑rank objective that considers corruption levels and masking realizations, and proposes Traj‑MC, a Monte Carlo method to efficiently estimate the trajectory second moment. Experiments show that trajectory‑aware calibration yields better reconstruction along the generation trajectory and preserves significantly more mathematical reasoning compared to clean calibration.
By Tian Liang, Zishan Shao, Yiran Chen
The paper introduces a paired candidate‑coverage benchmark protocol for typed decision models, evaluating two models—Laya and Jev—on datasets such as AG News, DBpedia, Emotion, and TREC. It reports that Laya detects a high percentage of missing-answer cases but also falsely rejects many valid candidates, whereas Jev shows lower false rejection rates but also lower detection of missing answers. The study highlights the need for separate measurements of classification, score ranking, and rejection policies, noting that the benchmark is descriptive and limited to reference‑label omission.
By Jiawen Lu, Tongtong Wu