Prediction
-
DisasterRoute: Evidence-Guided Dual-System Routing for Visual Disaster Question Answering
2026-10-21 05:41:44Tan Zhang, Xing Deng, Lu Zhang, Xiaofeng Zhu, Ping HuAbstract
Visual question answering from low-altitude UAV imagery must deliver timely and accurate judgments to support disaster response under limited edge computing resources. Routine questions may be answered through direct decisions, whereas more challenging questions may benefit from step-by-step reasoning. However, existing research primarily focuses on improving models' visual understanding and generative reasoning capabilities, with limited exploration of efficient direct decision-making and its adaptive coordination with generative reasoning. We develop Disaster-S1 based on a vision-language model adapted to the disaster domain. Trained for option discrimination, it outputs a probability distribution over answer options in a single multimodal forward pass for fast decision-making. We further propose DisasterRoute, a dual-system routing framework that combines prediction confidence with disaster evidence to decide whether to accept the initial answer or invoke the disaster reasoning model DisasterVL for review. By allocating inference computation on demand, the framework retains the efficiency of direct decision-making and improves overall question-answering accuracy through selective review. Experiments on DisasterBench show that Disaster-S1 achieves superior accuracy with low latency, while DisasterRoute further improves accuracy through selective review at substantially lower latency than full reasoning.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29231
PDF
-
Adaptive Steering and Remasking for Safe Generation in Diffusion Language Models
2026-10-21 04:56:49Yejin Lee, Ungsik Kim, Yo-Sub HanAbstract
Diffusion Language Models(DLMs) provide a promising alternative to autoregressive language models through iterative denoising and bidirectional generation. However, their iterative generation process introduces distinct safety vulnerabilities because harmful content can emerge at arbitrary positions and persist across subsequent denoising steps. Existing defenses rely on fixed interventions or aggressive remasking, which limits adaptive control over denoising trajectories and can degrade generation quality. We propose an inference-time defense framework that combines adaptive safety steering with safety-aware remasking. Our method uses a gating direction to continuously adjust steering strength from the current denoising state and applies a steering direction to masked positions to guide subsequent predictions toward safer trajectories. Our method further employs a lightweight response detector after the first generation block to identify unsafe trajectories at an early stage. The detector triggers targeted remasking over generated content and part of the conditioning prompt, and the model regenerates the selected positions under adaptive safety steering. This design combines continuous trajectory control with explicit correction of unsafe content while requiring no modification of model parameters. Experiments on LLaDA and Dream demonstrate that our method improves robustness against diverse jailbreak attacks while preserving benign generation quality and general model capability. Our code is available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2605.13043
PDF
-
ROCOCO: Robust Outline Completion of Occluded Curvilinear Objects
2026-10-21 04:09:17Ken Zheng, Jaimyn DrakeAbstract
We propose a learning-based generative model for completing curvilinear features under occlusion by environmental obstacles. To this end, we adopt a flow matching-based approach that supports the inpainting of occluded regions with the appropriate curvilinear features. To eliminate the reliance on privileged information about the occlusion a priori, we also train a UNet-based mask prediction model, which can identify regions for inpainting given only the occluded observation. The components of our approach are validated via experiments using procedurally-generated test cases, and the complete pipeline is validated using large public datasets for curvilinear objects. Against off-the-shelf inpainting baselines, ROCOCO is the strongest method by a wide margin when no occlusion mask is given, and is competitive with mask-supervised inpainting when one is, achieving best on full-image error and second only to LaMa on the occluded region, while using a completer of $7.7$M parameters against LaMa's $51$M and Stable Diffusion's $860$M.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29177
PDF
-
Carving Radiology Reports at Their Evidential Joints: Anatomical Region Sets as Generation Units
2026-10-21 03:55:22Jinwei Yang, Junhao Chen, Jianwei Zhao, Zhonghan Xu, Hong Cheng, Ling WangAbstract
Organizing radiology reports into independently generated units offers a way to generate fine-grained clinical content and bind it to visual evidence. However, existing methods typically adopt generation-unit structures not directly specified in natural reports or their annotations, introducing additional modeling compromises. We propose RadCarve, which instead defines generation units directly from the anatomical region sets in existing statement-region correspondences, unifying unit formation with visual evidence selection. RadCarve's planner uses composition-specialized query banks to jointly predict unit existence and region composition from the image, iteratively refining both with regional evidence aggregated according to current membership predictions. The generator independently produces each unit's text conditioned on the joint visual evidence of its predicted region set. Compared with strong baselines, RadCarve improves CXRFEScore by 9.4% in unit generation given region sets and achieves 2.4 times the end-to-end binding F1. It also achieves leading report-generation results on Chest ImaGenome and PadChest-GR, using fewer than one-sixth as many parameters as large-scale baselines.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29165
PDF
-
SimPT: An Automated Simulation Framework for Intervention-Based VLA Post-Training
2026-10-21 03:53:00Caicheng Wang, Zipeng Ye, Zhexuan Zhou, Zilong Huang, Xi Zhang, Yuchen XieAbstract
Human-in-the-loop (HIL) post-training improves vision-language-action (VLA) policies through expert corrections, but costly human supervision and repeated real-robot evaluation limit algorithm exploration and development. We introduce SimPT (Simulation-based Post-Training), a simulation framework that automates intervention-based VLA post-training by integrating policy rollouts, corrective data collection, policy training, and standalone evaluation. To automatically determine intervention timing and recovery strategies, we develop an Agent-Compiled Progress Graph that tracks task progress using privileged simulator state and retains stable snapshots during policy execution. After a rollout fails, a CuRobo-based expert attempts recovery from these snapshots, starting at more advanced task stages and falling back to earlier ones as needed to generate corrective trajectories. We implement and compare four post-training methods, RECAP, HG-DAgger, DPO+SFT, and FlowPRO, across seven RoboTwin manipulation tasks. Relative to a fixed-interval snapshot baseline, the Progress Graph reduces recovery attempts by 75.0% and increases recovery success from 93.4% to 98.9%. Using SimPT, we develop RECAP-CF-LPF, a refinement of RECAP that combines cross-fitting and low-pass filtering of value predictions. This refinement increases the overall success rate from 76.50% to 79.36% in held-out simulation trials. In real-robot HIL post-training, it increases the success rate from 72% to 96% across 25 scenarios of an Ethernet cable-insertion task. These results demonstrate the utility of SimPT for exploring and developing VLA post-training algorithms.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29161
PDF
-
PhyMo: Learning Physical Dynamics with Accurate and Continuous Motion from Multi-View Videos
2026-10-21 02:49:38Shangjia Liu, Jinxi Li, Siyuan Zhou, Bo Yang, Bing WangAbstract
Recovering dynamic 3D scene geometry, appearance, and the physical motion states that govern scene evolution from multi-view videos is important yet challenging. A key difficulty is that appearance can be deceptive. Under rendering supervision alone, a model may fit the observed images by changing color or opacity to compensate for inaccurate deformation, rather than recovering the true evolving geometry and motion. This issue is especially severe in scenes with fast or locally complex motion, where inaccurate trajectories can still produce plausible images over the observed frames, but lead to unreliable reconstruction and discontinuous future prediction. In this paper, we propose PhyMo, a framework for learning evolving dynamics from multi-view videos. We introduce a Motion-Aware Trajectory Representation that drives the model to recover accurate motion from deformation, rather than relying on appearance shortcuts to explain observed motion. On top of this, we propose a Continuous Motion Evolution module with bidirectional kinematic continuity constraint to promote temporally consistent evolution of displacement, velocity, and acceleration. Together, these designs yield geometrically faithful interpolation and continuous motion extrapolation. Experiments on four dynamic scene benchmarks demonstrate state-of-the-art performance.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29107
PDF
-
Diffusion Posterior Sampling with Uncertain Forward Models
2026-10-21 02:47:28Wentao Shangguan, Vivek K GoyalAbstract
Diffusion posterior samplers reconstruct signals by combining a pretrained diffusion prior with a measurement likelihood specified by a forward model and a noise model. Calibration and reference data can characterize uncertainty in the forward model's parameters, such as blur kernels or detector responses. Fixing these parameters at estimated values leaves their uncertainty out of the likelihood, and parameter errors are thus absorbed into the reconstruction. We introduce an uncertainty-aware likelihood (UAL), a Gaussian approximation to the marginal likelihood obtained by integrating over the uncertain parameters. For each candidate signal, UAL propagates parameter uncertainty into a structured measurement covariance, reducing the quadratic penalty along residual directions that parameter variation can explain. Existing samplers can use UAL in their data updates while retaining their pretrained diffusion priors and sampling schedules. Under a fixed Gaussian nuisance model, we bound likelihood error from linearization and average divergence between the exact marginal and UAL posteriors. We compare samplers with and without UAL across six inverse problems spanning Gaussian and motion deblurring, CT, MRI, ultrasound, and fluorescence microscopy, and find improvements in reconstruction and ensemble prediction. UAL reduces geometric mean MSE by 9% to 20% across samplers in the main Gaussian-deblurring setting and by 51% to 92% in motion deblurring when non-blind samplers use the population-mean kernel. In motion deblurring, DAPS and PnP-DM with UAL also give 6.2 to 6.4 dB higher ensemble-mean PSNR than published BlindDPS and Fast Diffusion EM. On the ablation set, UAL outperforms trace-matched isotropic inflation for PnP-DM in both deblurring tasks; in motion deblurring, it also exceeds the best-tested fixed scalar data weighting by 3.3 dB for PnP-DM and 4.2 dB for DAPS in average PSNR.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29102
PDF
-
Dual-Channel Contrastive Encoding of Sperm Whale Coda
2026-10-21 02:46:56Jo\~ao Victor Quintanilha, Emma Virnelli, Shuo ZhangAbstract
Sperm whales (Physeter macrocephalus) communicate through short click sequences called codas, whose structure includes the rhythmic pattern of inter-click intervals (ICI) and the spectral composition of the clicks. In this paper, we propose the Dual-Channel Contrastive Encoder (DCCE), which explicitly encodes rhythm and spectral information with separate encoders before fusing them. We evaluate DCCE on three coda-understanding tasks: social unit, coda type, and individual identification. We compare DCCE against strong baselines from literature under a common linear-probe protocol and analyze the role of each channel and training objective through ablations. On 1,383 codas, shared partitions and validation-only layer selection support three random splits and three recording-date-disjoint folds. DCCE with unit pairing and type supervision achieves mean random-split macro-F1 of 0.901/0.865/0.597 for unit/type/identity. Raw ICI remains strongest for type (0.987), and ICI plus temporal mel statistics is a strong identity reference (0.724). An auxiliary identity objective improves spectral-branch identity prediction but does not improve the mean fused-identity score in this configuration. Date-grouped evaluation and four year holdouts expose generalization limits. The results support task-dependent combinations of timing and spectral information, without establishing independent biological channels or general architectural superiority. Code, notebooks, and complete results are available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29101
PDF
-
Causal world models under physical parameter shifts
Abstract
World models predict how embodied systems evolve under actions, but their accuracy can degrade when physical conditions move outside the training distribution. We study two components of causal modeling: which inputs a predictor may use, and how its mechanism computes their effects. In a synthetic benchmark with shortcut features, a frozen dependency mask derived from paired interventions reduces mean out-of-distribution (OOD) RMSE from 1.769 for a fully connected predictor to 0.697, close to the generator-oracle mask (0.694). For robot link-mass extrapolation in CausalWorld, with disjoint training and test mass ranges, we develop finger-local world models in which link masses enter inertia and dynamics-bias terms through explicit affine coefficients, actions act through either a learned torque network or the known PD controller, and learned residuals capture contact-related effects. Compared with a separately trained control whose mass inputs are fixed at training means, the known-PD model turns out-of-range masses into a consistent predictive gain: median time-averaged joint R2 over 150-step rollouts rises from 0.609 to 0.727, with improvement in all ten task-seed runs (exact two-sided sign test, p = 0.002). In a shared 100-step forecast, the learned-torque model is competitive with the strongest recurrent robotic world model (RWM) configuration (median joint R2 of 0.755 - 0.768 versus 0.752), whereas supplying out-of-range masses sharply degrades the MLP and recurrent predictors. These results indicate that how a physical parameter enters the dynamics mechanism determines whether its out-of-range value helps or harms prediction.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29088
PDF
-
Single-Image Food Mass Estimators Do Not Transfer Their Gram Scale Across Datasets
2026-10-21 00:59:03Yongsheng LianAbstract
Estimating food mass from a photograph is central to image-based dietary assessment, but whether trained models transfer to new cameras and foods is unclear. We fine-tuned DINOv2-base on 2,253 weighed Nutrition5k dishes to predict mass from one RGB photograph and compared it with a zero-shot vision-language model, Qwen3-VL-32B, on Nutrition5k and three external datasets: ASA24 studio portions, ECUSTFD whole foods and Food Portion Benchmark meal trays. In domain, the fine-tuned model's mean absolute error was 24.8 g, less than half of Qwen's. On external datasets it did not beat Qwen. For a food it had never seen, it ranked portions by size, but no better than a count of food pixels. Its gram scale was wrong in two ways: it changed with how much of the scene the photograph showed, and it was off by a different factor for each food. Training on other foods or trays from the tested dataset cut the external error by 38 to 65%. DINOv2-large, a larger backbone trained on the same dishes, did not reduce the external error. One labeled photograph of the same food cut the model's ASA24 error to 28.3 g, better than that photograph achieved alone. Prediction intervals that held the true mass for 90% of Nutrition5k dishes held it for at most half of external photographs. Nutrition5k often photographed the same plate several times, so a random test split shared near-identical photographs with training and overstated accuracy. Single-image mass estimators therefore need labeled photographs from the setting where they are used.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29025
PDF
-
FlashNext-RL: A System Optimization Study for Long-Horizon Agentic RL of Qwen3.8-Flash-Next
2026-10-20 23:53:09Zhewei Yao, Yuxiong HeAbstract
Long-horizon agentic reinforcement learning (RL) requires both efficient rollout generation and training on long sequences. However, the memory requirements and model-specific operators of Qwen3.8-Flash-Next make this difficult with existing open-source implementations. To resolve these limitations, we introduce FlashNext-RL, a model-specific RL training optimization framework based on prime-rl, with vLLM for software engineering (SWE) rollout generation. Our main focus is BF16 training of the full 48-layer model on a single 8xB200 node, with the first two layers, the n-gram table, the sparse-attention indexer and the multi-token prediction head frozen. We optimize host offload, gradient accumulation and the kernels for mixture-of-experts, sparse attention and the gated residual stream. In the full-model development study, non-padding training throughput increases from approximately 395 tokens/s/GPU in the initial 32K-token configuration to 4,438 tokens/s/GPU in the optimized 131K-token configuration, with hardware matmul throughput (aka without considering the cost of attention components) increasing from approximately 20 to 213 TFLOPs/GPU. A controlled cumulative study on an 8-layer model further quantifies the contributions of individual optimizations. Together with three inference nodes, the system supports a live SWE RL run with a target batch of 1,024 trainable rollouts and a 131,072-token context limit. Over steps 5-15, an iteration takes 22.4 minutes on average, with bounded entropy and trainer-sampler mismatch KL. These results highlight the importance of tailoring RL training to the model's architecture, memory layout and kernel requirements, rather than relying only on general-purpose RL infrastructure.
Abstract (translated)
URL
https://arxiv.org/abs/2610.28998
PDF
-
RoPT: Robust 6DoF Object Pose Tracking in Video
2026-10-20 22:25:47Yongqing Liang, Yuzhong Chen, Sihang Li, Junsuk Kim, Chase Kidder, Hui-ping Wang, Yunzhi Xu, Raihan Haque, Yubiao Zhang, Chen FengAbstract
Accurate 6-DoF object pose tracking is essential for robotic interaction, yet conventional monocular pipelines frequently suffer from catastrophic divergence under rapid object motion. We present RoPT, a robust tracking framework engineered to overcome this limitation by unifying visual perception, temporal kinematics, and global re-localization into a tightly coupled closed loop. Unlike existing open-loop trackers, RoPT delivers three critical capabilities that fundamentally alter system resilience. First, the visual perception component tracks object masks to robustly handle partial and severe occlusions. Second, a kinematic factor graph regularizes frame-to-frame updates, forecasts forward pose predictions, and drives a parallel multi-hypothesis verification scheme to bypass local minima. Finally, an active re-localization module transforms the pipeline into a closed-loop system, making it failure-aware and capable of deterministic self-recovery. We demonstrate the effectiveness of our framework across challenging benchmarks alongside real-world robotic demonstrations, achieving significantly higher tracking survivability under rapid motion.
Abstract (translated)
URL
https://arxiv.org/abs/2610.28962
PDF
-
RAHF-Net: Range-Aware Heterogeneous Fusion for LiDAR-Assisted RGB Semantic Segmentation
2026-10-20 21:00:32R. D. Lin, Pengcheng Weng, Ge Wang, Han Ding, Fei WangAbstract
This paper studies LiDAR-assisted RGB semantic segmentation, where LiDAR point-cloud features improve dense pixel-wise predictions in the RGB image plane rather than 3D point labels or detections. Existing 2D fusion methods often project point clouds into shallow depth/range maps and apply spatially uniform fusion, discarding rich 3D structure and ignoring the range-dependent reliability of outdoor LiDAR. We propose RAHF-Net, a Range-Aware Heterogeneous Fusion Network. Its Cylindrical-to-Perspective (C2P) module projects pretrained cylindrical voxel features into the image plane, preserving structured 3D semantic priors. A Range-Aware Heterogeneous Fusion (RAHF) module then uses a soft depth-derived range mask to perform LiDAR-guided geometric enhancement in reliable near-field regions and visual-context enhancement in sparse far-field regions. On KITTI-360, RAHF-Net achieves 65.76 mIoU, outperforming representative RGB-only and RGB-LiDAR baselines, including CMX by 1.45 points. Experiments on DELIVER further demonstrate improved robustness under adverse weather, illumination degradation, and sensor corruption.
Abstract (translated)
URL
https://arxiv.org/abs/2610.25400
PDF
-
Vector-Quantized World Model: Latent Feedback Motion Planning for Robot Manipulation
2026-10-20 20:08:37Xuezhou Xu, Chung-Pang Wang, Zhenzhen Li, Benjamin RiviereAbstract
Visual world models have shown strong potential for robot planning, yet long-horizon manipulation remains challenging due to compounding prediction errors, large search complexity, and latent plans that the low-level controller cannot reliably reach. In this work, we introduce Vector-Quantized World Model (VQ-WM), a latent feedback motion planning framework that integrates discrete long-horizon search with continuous visual feedback control: a discrete latent predictor models transitions between vector-quantized latent states, allowing the planner to look far ahead and find a shortest path to the goal through discrete search and a continuous learned controller sequentially reaches the planned states through visual feedback, correcting deviations to mitigate compounding errors. Whereas most vector quantization uses Euclidean distance, we use a novel estimator of the first-passage-time as a quasi-metric, which aligns the planning abstraction with the controller's capabilities. We evaluate on the physical robot manipulation of the Tower of Hanoi puzzle: whereas existing state-of-the-art foundation models (VLAs, WAMs) only succeed when the solution trace in their training dataset, our method achieves 100\% success rate with only short-play data, demonstrate its ability to compose long-horizon solutions beyond the short demonstrated trajectories and execute them reliably. Our solution also outperforms strongest world-models by 13.3--85.3 percentage points on simulation benchmarks spanning manipulation and locomotion, demonstrating consistent progress and new capabilities.
Abstract (translated)
URL
https://arxiv.org/abs/2610.25600
PDF
-
Robot Learning with Visual Predicted Force
2026-10-20 19:20:24Haonan Chen, Feiyang Wu, Yuxiang Ma, Mustafa Mete, Pengfei Ye, Junxuan Shen, Cheng Zhu, Aurora Ruggeri, Kelvin Cheung, Jiayuan Mao, Edward Adelson, Jiajun Wu, Robert D. Howe, Yilun DuAbstract
Force-aware manipulation typically relies on specialized force or tactile sensors. We show that force-aware manipulation can instead be achieved through visual force prediction from the deformation of a compliant Fin Ray gripper. Our approach trains two models. First, we train a visual force estimator on calibration data and use it to annotate task demonstrations with force estimates. Second, we train an action--force proposal policy on these force-augmented demonstrations to jointly generate candidate robot actions and their associated forces. At test time, we sample candidate actions and the forces they are expected to produce, then execute the action whose predicted force is closest to a target from the demonstrations. We evaluate our approach on berry picking, empty-can grasping, in-hand reorientation, and plug insertion. Our results show that visual force prediction can guide inference-time action selection for contact-rich manipulation without requiring force or tactile sensors at deployment.
Abstract (translated)
URL
https://arxiv.org/abs/2610.04741
PDF
-
T-WLA: Bridging Slow Vision and Fast Touch
2026-10-20 18:36:13Kaichen Zhou, Yuzhen Chen, Hannah Yao, Chonghe Jiang, Rao Fu, Xihang Yu, Wei Dai, Bashar Kabbarah, Haoyu Chen, Ishita Goluguri, Zhiyang Dou, Haonan Chen, Fangneng Zhan, Mengyu Wang, Wojciech Matusik, Antonio Torralba, Paul Pu LiangAbstract
In contact-rich manipulation, contact conditions can change before a visual policy updates its action plan. Recomputing the plan through the visual policy adds latency, while tactile feedback alone does not specify how the ongoing motion should be revised for the task. We introduce T-WLA, a world-language-action architecture that uses shared visual memory to guide both action generation and tactile correction. At each visual update, the visual branch integrates the current image, task instruction, and retrieved history into a visual memory that conditions base-action generation. The same memory is cached for a lightweight tactile editor, which combines it with the active action chunk and incoming touch to predict residual corrections to unexecuted actions without rerunning the visual branch. Future-image and future-pressure prediction provide training-only supervision, encouraging the visual memory and tactile features to capture interaction dynamics relevant to action generation and correction. We evaluate T-WLA on tactile-conditioned hand-motion prediction using EgoTouch and Tachin, and on four real-world dexterous manipulation tasks. T-WLA reduces finger-joint MPJPE by 30% relative to the strongest reported baseline and achieves an average success rate of approximately 50.0% across real-robot manipulation tasks. Videos and code are available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.28814
PDF
-
SAIL: A Learnable Generalisation of Spatial Autocorrelation over Graph-Structured Data
2026-10-20 18:28:15Piotr Keller, Fayyaz MinhasAbstract
In vision and scientific domains involving spatial data, a key question is whether node feature values vary independently across space or form coherent spatial structures. Classical spatial autocorrelation statistics such as Moran's $I$ and Geary's $C$ quantify this phenomenon but operate on a single scalar variable with fixed neighbourhood weights, limiting their use for high-dimensional graph-like data. We introduce SAIL (Spatial Autocorrelation Index Learner), a differentiable framework that generalises spatial autocorrelation into a learnable operator by jointly learning feature projections defining candidate scalar fields and inter-node neighbourhood affinities. SAIL produces interpretable local and global spatial coherence descriptors, enabling supervised prediction and unsupervised discovery of spatial autocorrelation modes and hotspots. We show that SAIL exactly recovers classical spatial statistics under appropriate parametrisations and, for tasks driven by latent spatial autocorrelation, can approximate the Bayes-optimal predictor with an inductive bias matched to such tasks. Experiments on synthetic data and large-scale biomedical graphs demonstrate statistical correctness and discovery of spatial coherence with applications across vision, geospatial analysis, and biomedicine. The code repository is available at: this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.28806
PDF
-
Motion Equivariant Convolutional LSTM: When Motion Becomes a Gauge Symmetry
2026-10-20 18:08:45Vahid Hosseinzadeh, T. Anderson Keller, Michael PuthawalaAbstract
Objects move, cameras pan, and weather drifts with the wind: much of what changes in sequential data is motion. Motion changes where things are, not what they are, and is therefore a symmetry of the data. Yet standard recurrent networks are not equivariant to it: their memory stays put while the input moves on. Flow equivariant recurrent networks first brought this symmetry into sequence models, for motions of constant velocity, or flows, by enumerating a bank of velocities. We extend this idea in three directions: from simple recurrent networks to convolutional LSTMs, from flows to motions whose velocity changes in time, and from enumerating velocities to estimating them. Treating a sequence as a field over time, we show that motion becomes a gauge symmetry: a global symmetry made local in time. Built on this view, our motion equivariant convolutional LSTM (MELSTM) uses its velocity model as the gauge field and is provably equivariant to every motion. Across future-frame prediction, classification of digits visible only through their motion, and action recognition under camera shake, MELSTM outperforms a convolutional LSTM and its flow equivariant counterpart with the same number of parameters: it is more accurate, generalizes to longer horizons and to unseen motions, and carries a fraction of the flow equivariant model's state.
Abstract (translated)
URL
https://arxiv.org/abs/2610.28771
PDF
-
Adaptive Multi-Token Prediction: How far should we go?
2026-10-20 17:56:32Emile Anand, Mohammed Rehan Parwani, Xinyuan CaoAbstract
Multi-token prediction (MTP) can accelerate language-model decoding by drafting several tokens and verifying them in a single target-model pass. However, the best speculation horizon changes with context: longer drafts offer greater amortization when reliable, but can waste compute on suffixes discarded after a rejection. Unlike autoregressive drafting, where a stopping rule can decide token by token, MTP must fix the horizon before drafting, as all heads fire in the same forward pass. We therefore formulate horizon selection as an online control problem, using verification feedback to learn how far the draft can be trusted in each context, and introduce Adaptive Multi-Token Prediction (AMTP): a lightweight controller that requires no retraining and leaves the exact verifier unchanged. To guide horizon selection, we characterize the expected number of tokens committed at each horizon via acceptance survival, the probability that a draft prefix is fully accepted. Before each round, AMTP predicts how far the draft is likely to survive and picks the horizon with the highest expected tokens per second, learning freely from every verification pass. Theoretically, we show that any online learner's prediction error translates into throughput regret against the best horizon for each context, and in turn into a bound on wall-clock completion time. With online gradient descent, this regret is sublinear. Across five draft-target pairings built on Llama-3.1-8B-Instruct with Medusa heads, Meta's MTP models, and Qwen3-4B-Inst-2507-MTP, evaluated on AIME 2025 and GSM8K, AMTP decodes up to 4.1x faster than standard decoding and matches or outperforms the fixed window in 8 of 10 settings, with throughput gains of up to 10%.
Abstract (translated)
URL
https://arxiv.org/abs/2610.28619
PDF
-
NarraSpace: Multimodal Scene-Grounded Evidence for Picture-Description Dementia Classification
2026-10-20 17:55:44Qingming Li, Ruiyan Sun, Satoshi NakamuraAbstract
Picture-description dementia classification uses a shared visual scene, but a transcript-level prediction alone does not identify the scene evidence underlying its inputs. NarraSpace links words and speech times to named picture elements and retrieves utterances for predefined scene events. Narrative-spatial descriptors summarize coverage and temporal organization. A frozen language model contrasts each event's text under examples from Alzheimer's disease and healthy-control training participants, alongside transcript surprisal and coverage measurements. Retained source records link event scores to their source utterances, speech intervals, and lexical matches. On the official 48-participant test set of the Alzheimer's Dementia Recognition through Spontaneous Speech benchmark, a fixed support vector classifier achieves $80.88\%\pm2.56\%$ mean accuracy over fifty seeds; averaged probabilities yield $81.25\%$ accuracy and $91.32\%$ area under the receiver operating characteristic curve. NarraSpace makes the construction of scene-linked classification inputs explicit and inspectable by connecting event-indexed language-model measurements to their source utterances, speech timing, and extraction records. Such source-linked measurements could support clinician review of picture descriptions by presenting quantitative summaries alongside the original speech evidence.
Abstract (translated)
URL
https://arxiv.org/abs/2610.28616
PDF
