Papers

Vision-Language-Action (VLA) models have emerged as general-purpose robotic manipulation policies. However, in dexterous hand manipulation, contact forces are typically provided as separate states or force-specific representations, making it difficult to explicitly represent the spatial correspondence between force and their corresponding visual locations. In this work, we propose VisForce, which visually grounds the current and desired forces at their corresponding fingertip locations. VisForce renders current and desired visual force cues on the current wrist image and a task-specific goal image, and combines the two representations through goal-conditioned cross-attention to generate force-aware actions. We evaluate VisForce using a real UR10 robot equipped with an RH56F1 dexterous hand through force-conditioned grasping and three multi-stage manipulation tasks. In force-conditioned grasping experiments, VisForce exhibited a consistent grip-force response as the desired force increased, and achieved grasp-and-lift success rates of 70% and 80% for an egg and a toothpaste tube, respectively. It further achieved final success rates of 70%, 55%, and 40% on cup insertion/bottle pouring, tong-assisted bread transfer, and slip-modulated peg-in-hole, respectively. These results show that fingertip-aligned visual force representations can be effectively used for force-aware conditioning in VLA-based dexterous hand manipulation.

Large language model (LLM) serving is increasingly agentic, with multi-turn sessions that idle between actions yet must retain their full context. Limited GPU memory capacity forces inactive KV states to be evicted, so resuming a session incurs either costly recomputation or slow interconnect transfers. To address this, high bandwidth flash (HBF)-an on-package 3D-NAND memory offering orders-of-magnitude greater capacity than high bandwidth memory (HBM) at comparable read bandwidth-has emerged as a strong candidate. However, its high read energy and limited write endurance make it impractical to serve all KV traffic. Fortunately, our analysis shows that agentic KV states exhibit distinct access patterns: a small hot set is read for every decoding step, while a large cold pool is read only when a paused session resumes. Exploiting this, we place the hot set in HBM and the cold pool in HBF, forming a hot-cold KV hierarchy within the GPU memory tier. On agentic workloads with Qwen3-Coder-30B-A3B, our design delivers 14 ms time-between-tokens (TBT) and adds only $\approx$0.1 ms of resume latency on top of prefill, while hosting $24\times$ more concurrent sessions per GPU. By confining HBM to the hot set, our design also cuts read power by 7.6 kW per 8-GPU node relative to serving all KV from flash-establishing HBF as a cold-tier complement to HBM rather than its replacement.

Graph Incremental Learning has garnered increasing attention as dynamic graph data continues to emerge across diverse fields. Conventional approaches primarily address catastrophic forgetting by preserving node-related knowledge through replay or distillation techniques; however, they often incur high computational costs and inefficiency. This issue is further exacerbated in real-world scenarios where labeled data for new classes is scarce. In this paper, we propose a novel lightweight plastic-memory framework specifically designed for few-shot incremental learning on graphs. The core idea of our framework is the construction of a plastic-memory module that evolves over time, continuously updating and expanding its memory to accommodate new classes while retaining previously learned knowledge. In contrast to existing techniques, our memory module is both lightweight and effective, featuring an innovative evolving micro-clustering structure that dynamically updates representations of class prototypes, sub-prototypes, and their interaction weights. Building on this memory module, we introduce a memory-driven meta-learning framework that enhances adaptability to new tasks in its inner loop while maintaining stability for earlier tasks in the outer loop. Extensive experiments on four benchmark datasets demonstrate the framework's superior performance in balancing stability for old knowledge and adaptability to new knowledge.

Memory constraints in artificial intelligence motivate accurate function approximation with fewer parameters. We study looping, which repeatedly composes one update function with shared parameters; each output becomes the next input. A looped Transformer, for example, reuses one block, whereas its conventional untied counterpart uses separately parameterized blocks. We compare their parameter requirements for a given worst-case approximation accuracy, or equivalently, their approximation accuracy under a common budget limiting distinct trainable coefficients. We then ask whether this representational parsimony improves statistical accuracy. For general likelihood models, we establish an upper squared Hellinger risk bound for looped sieve maximum likelihood and a minimax lower bound for the jointly tuned untied family. Further loop iterations improve the approximation bound without adding parameters, while increasing computation and the fitted-class complexity bound. For targets of known Hölder smoothness, looped residual feedforward networks and post-layer-normalized Transformers attain the minimax polynomial rate up to logarithmic factors with a fixed number of bounded real parameters. At sufficiently large fixed budgets, looped worst-case risk vanishes while optimal worst-case untied risk remains bounded away from zero. The loop-to-untied risk ratio also tends to zero under specified growing-budget conditions. Regression, binary response, and energy-based generative models illustrate the theory.

Accurate multi-step traffic forecasting remains challenging because observed traffic signals contain heterogeneous temporal dynamics with different characteristics and levels of predictability. Existing approaches typically model these dynamics within a unified representation or rely on predefined decomposition rules, which may limit their ability to flexibly separate persistent patterns from rapidly varying fluctuations. To address this issue, we propose the Adaptive Decomposition Network (ADNet), a component-specific forecasting framework that adaptively disentangles traffic dynamics into dominant and residual components. ADNet introduces a learnable complementary spectral decomposition mechanism that determines the contribution of each frequency bin to the two components. Unlike hard frequency partitioning, every frequency bin can contribute to both components with different learned proportions, allowing the decomposition to be optimized jointly with the forecasting objective. The reconstructed components are then modeled by two dedicated spatiotemporal forecasting branches, and their predictions are integrated to generate the final multi-step forecast. Experiments on the Alameda and Orange regions of the TraffiDent dataset show that ADNet achieves the best performance in 20 of the 24 reported region-horizon-metric comparisons, with particularly clear gains at longer forecasting horizons. Capacity-controlled ablation experiments further show that the learnable decomposition substantially outperforms a fixed decomposition and provides additional improvements beyond the dual-branch architecture alone. These results demonstrate the effectiveness of adaptive decomposition and component-specific modeling for multi-step traffic forecasting.

Nonlinear activations can create equivariant interactions between irreducible representations that linear maps cannot. We use the Gaussian degree decomposition to extend ordinary polynomial degree to such nonlinear maps, and prove that for a fixed coordinatewise equivariant layer each degree factors into a polynomial determined by the linear maps and a scalar determined by the activation. This separates three distinct obstructions, coming from symmetry, coordinates, and activation.

Recent advances in generative video models have enabled the synthesis of visually realistic content, posing significant challenges to synthetic video detection. Existing detectors often rely on appearance artifacts, semantic inconsistencies, and temporal patterns that may be generator-specific, limitating generalization to unseen synthesis models. We investigate whether responses to a pretrained generative model provide more transferable forensic cues. Our key observation is that real and AI-generated videos exhibit distinct velocity responses under a pretrained Flow Matching video model. This distinction persists when different pretrained video-generation backbones are used as probes, suggesting that velocity responses offer transferable forensic signals beyond visual artificts. Motivated by this observation, we propose TRACE (\emph{\underline{T}rajectory \underline{R}epresentation \underline{a}nd \underline{C}onsistency \underline{E}stimation}), a generation-process-aware framework for AI-generated video detection. TRACE leverages a pretrained video DiT as a velocity-field probe to extract representations at multiple flow time points, and models cross-frame consistency through velocity differences between adjacent frames. We further introduce a Real-Centered Trajectory Optimization objective that encourages generator-invariant representation learning. Extensive experiments on AIGVDBench demonstrate that TRACE generalizes effectively across diverse generators, substantially outperforming prior state-of-the-art methods on unseen open- and closed-source video generation models.

Large language models (LLMs) are emerging as scientific assistants, yet their computational demands and limited domain specialization constrain sustainable deployment in environmental engineering. Here, we investigate whether domain-specialized small-scale LLMs can combine scientific capability with sustainable deployment in biological wastewater treatment. We developed a benchmark evaluating three scientific capabilities of LLMs: retrospective cognition, comprehension fidelity, and prospective extrapolation. BioWater (8 billion parameters, fine-tuned on specialized domain knowledge) achieved higher comprehension-fidelity scores than participating human experts and performance comparable to a 397-billion-parameter general-purpose LLM in retrospective cognition and prospective extrapolation. Human-BioWater collaboration generated a scientific hypothesis that was subsequently supported by laboratory experiments, demonstrating its potential to contribute to prospective scientific research. We further evaluated the economic and environmental implications of LLM deployment across global wastewater treatment plants (WWTPs). Locally deployed small-scale LLMs became more sustainable than cloud-based large-scale LLMs as inference demand increased in intelligent WWTPs. These findings highlight domain-specialized small-scale LLMs as a promising pathway towards scientifically capable, computationally efficient, and sustainably deployable artificial intelligence for wastewater treatment.

HopChain has shown on still images that multi-hop data synthesis improves vision-language reasoning, because long chain-of-thought reasoning exposes errors that compound across steps, while most data used for reinforcement learning with verifiable rewards (RLVR) rarely demands a chain of visual evidence, so these weaknesses are likely to stay unexposed. We observe the same problem in video, where this framework has not yet been explored. We therefore build Video-HopChain, a dataset of 22,550 multi-hop video questions over 13,378 videos, together with a held-out benchmark of 1,000 questions. Each question chains three to six yes/no questions about moments in one video, and each yields one of two integers depending on its answer. The final answer is the sum of these integers, so an exact match on that sum gives the verifiable reward that RLVR needs. We first train Qwen3-VL-8B with GRPO on a standard video dataset, and a second stage on Video-HopChain then raises the mean over eight video understanding and reasoning benchmarks from 55.4 to 57.9 and improves every one of them. Training on such a dataset, however, exposes a known limitation of GRPO: its learning signal comes from the reward variance within a group, so hard questions whose rollouts are all incorrect and easy questions whose rollouts are all correct both leave the group with no gradient. To recover these groups at the same compute budget, we introduce Confidence-Gated Exploration (CGE). With 8 rollouts per question, CGE samples the first 4 as usual. If these 4 are either all correct or all incorrect, it samples the last 4 with the policy's most confident token masked inside the reasoning span, and removes the masked positions from the loss while all 8 rollouts enter the advantage. With CGE, the mean rises further to 59.3. We release the dataset, the checkpoint, and the data generation and training code.

Integrated sensing and communications (ISAC) spans radiofrequency (RF), visible-light, optical-fiber, power-line, and acoustic systems, yet technology transfer across these domains remains underexplored. This article examines bidirectional knowledge and technology transfers centred on RF-ISAC. It shows how VLC motivates power-domain sensing, fiber enables differential referencing and distributed processing, and PLC inspires topology-aware monitoring and adaptive probing. Conversely, RF-ISAC contributes joint signal design, nuisance suppression, and weak-return recovery. The article highlights transferable principles, required adaptations to improve ISAC, and information potentially lost across physical media.

Vision-language models (VLMs) have achieved strong performance on tasks such as visual question answering, yet small image resizes can turn correct answers into errors. We investigate whether changes in visual configuration, such as image tiling and token arrangement, contribute to this instability. Across seven checkpoints and four benchmarks, equally small resizes cause more correctness flips when they switch configurations. Surprisingly, in over half of these cases, models answer the question incorrectly but can still read the correct answer when told what to read. Furthermore, attention interventions in LLaVA-NeXT suggest that configuration changes can weaken the use of readable information during answering. We therefore guide models using field cues and their own transcriptions. With annotation assistance, these forms of guidance together correct 97.2% of errors with readable information. These findings show that configuration changes can affect how models use information they can still read.

Graphical User Interface (GUI) Agents autonomously interact with software to fulfill user requests, where GUI navigation stands out as the most critical and challenging capability. Mastering this capability demands a complex synergy of step-wise decision-making, state-action alignment, and long-horizon planning. While directly mixing these corresponding navigation tasks seems intuitive to simultaneously acquire these skills, such a direct combination is severely bottlenecked by inconsistent optimization objectives and profound data heterogeneity. To overcome these barriers, we propose the MaP (stands for “Masked Trajectory Prediction”), a unified framework that seamlessly harmonizes divergent GUI navigation tasks. By modeling multi-turn GUI interactions as a trajectory and defining training objectives through component masking and prediction, MaP shifts the optimization from task-specific marginal distributions to a consistent objective. Furthermore, to handle the data heterogeneity across multiple navigation tasks, we design a role-aware adapter learning module that dynamically routes each token to a specialized representation space. Extensive experiments on five representative GUI navigation benchmarks demonstrate that MaP effectively mitigates gradient conflicts and significantly outperforms the direct mixture training, establishing a robust paradigm for multi-task GUI navigation.

AI planning studies how an agent can reach a goal by executing a sequence of actions. To plan correctly, the agent needs an action model describing when each action can be executed and how it changes the world. Constructing such models by hand requires domain expertise, and can be costly and error-prone. Action models can instead be learned from available data using existing neurosymbolic approaches, but they currently assume access to complete traces of fully observable images . These approaches fail to learn action models under partial observability where some of the images might not be present or are not fully informative of the current state of the world. Hence, this paper proposes NeSyAM, a novel neurosymbolic modeling paradigm for action model learning under partial observability. In addition, the paper presents a unified variational framework for theoretically analysing the limitations of existing methods compared to our proposed approach. NeSyAM is then tested extensively on six visual planning domains and three observation regimes to show it consistently recovers relevant parts of the true action model under partial observability.

The Remote Point Problem (RPP) is an algorithmic problem that asks, given a linear subspace $L \subseteq \mathbb{F}^n$ of dimension $k$, to deterministically find a vector $v \in \mathbb{F}^n$ far in Hamming distance from $L$. This problem was introduced by Alon, Panigrahy and Yekhanin [APY09], motivated in part by the matrix rigidity approach for proving circuit lower bounds. An algorithm is said to achieve remoteness $d$ if it finds a vector $v$ whose Hamming distance from $L$ is at least $d$. We observe that over the rational numbers, the problem admits a deterministic polynomial-time algorithm that achieves optimal remoteness $n-k$. Over finite fields, we obtain a (modest) improvement of a result of Alon, Panigrahy and Yekhanin [APY09], and give an algorithm that achieves remoteness $\Omega\left(\frac{n}{\max\{k, \log n\}} \log n\right)$.

Using large language models (LLMs) to simulate diverse human populations has the potential to transform many aspects of computational social science, yet many evaluations score the average response rather than the spread of opinion within real groups. Here, we develop a diagnostic framework that measures point accuracy alongside dispersion retention, the ratio of predicted to human standard deviation ($\dr$), on 10{,}000 respondent–question pairs from the World Values Survey (WVS) spanning twelve countries and six continents. We evaluate eleven zero-shot language models and five variants fine-tuned on WVS data with SFT, DPO, and GRPO. We identify a failure mode we term consensus collapse, where alignment training compresses outputs toward one stereotype per group. Along the post-training trajectory from the Llama~3.1 70B base to the Tulu~3 checkpoints, the first stage, supervised instruction tuning, removes half of the spread with minimal accuracy gain ($\dr$ 1.22 to 0.59; accuracy $+0.9$ points), the later stages do not restore it, and a gap opens between WEIRD and non-WEIRD countries that survey fine-tuning then deepens while pursuing higher point accuracy. The most accurate model (Tulu~3 70B-DPO fine-tuned on WVS, 57.9%) keeps half the human spread overall ($\dr = 0.50$) and 11% of it for Nigeria, against 0.70–0.87 for WEIRD countries. Raising the sampling temperature to 1.0 leaves the Wasserstein-1 distance ($\wone$) to human distributions unchanged for both fine-tuned DPO models, and GRPO on Qwen~3.5 9B does not restore the spread under either an accuracy reward or a distribution-shaped reward. Mixing the aligned model with an unaligned prior raises $\dr$ from 0.51 to 0.62 on a held-out split but leaves Nigeria at 0.36. Point accuracy alone therefore misjudges these simulators, and current post-training trades diversity for consensus.

This paper proposes a model-free current control method for permanent magnet synchronous motors (PMSMs) based on disturbance feedforward and residual feedback. An ultra-local current model incorporates motor dynamics, parameter uncertainties, cross-coupling effects, and other nonideal factors into generalized lumped disturbances. An extended state observer (ESO) estimates these lumped disturbances and compensates for them through feedforward action, transforming the original PMSM current-control problem into regulation of a simplified post-compensation residual system. A state-feedback H-infinity controller for the residual system is then learned directly from operating data using off-policy integral reinforcement learning. Owing to the simplified residual dynamics, the value function and control policies are parameterized in quadratic and linear forms, reducing the learning problem to low-dimensional parameter estimation without neural-network approximation. The proposed method requires neither prior knowledge nor online identification of PMSM electrical parameters: input-gain mismatch is incorporated into the ESO-estimated lumped dynamics, while the residual-feedback policy is obtained from operating data. Comparative simulations against deadbeat predictive current control, model-based H-infinity control, and model-free predictive current control show fast current tracking, low current distortion, and strong robustness to large parameter variations. With the learned H-infinity policy fixed and without retraining or retuning, nearly unchanged control performance is maintained when stator resistance, stator inductance, and permanent-magnet flux linkage are simultaneously varied to 20% and 200% of their nominal values.

What is the least recurrent memory needed to reproduce a specified expert under partial observability? The instantaneous requirement is the conditional entropy of the expert's behavioral quotient, but recurrence must also preserve distinctions that future observations will not restore before use. We characterize this minimal recurrent behavioral memory by a compatibility relation: under transitivity its classes attain the exact minimum, while the general case is an entropy minimization over closed compatible state assignments, with exact certificates on finite instances. A sole-carrier measurement protocol separates behavioral sufficiency, excess code rate, and information carried by observations or other memory paths; experimental bit requirements refer to the induced symbolic behavioral model under the stated occupancy. Across manipulation tasks, learned code rates remain near zero- and two-bit requirements as hidden modes grow to $512$, and anticipatory memory follows a $2\to1\to0$ requirement despite zero instantaneous demand during waiting. Learning this representation remains difficult: event-agnostic future-behavior supervision yields $36/40$ sufficient seeds with one frozen configuration and improves the longest-horizon pixel setting from $0/8$ to $6/8$ sufficient held-out seeds (closed-loop success from $0.08$ to $0.57$). On unmodified community benchmarks, the protocol certifies delay-independent requirements, which sufficient codes match at mid-delay. The supervision aids commitment but can induce predictive surplus; annealing it lets imitation and rate training reduce that surplus, separating the information-theoretic target from the ability to learn it.

Precision medical robotics demands adaptive decision-making under strict safety, interpretability, and execution constraints. Although recent Vision-Language-Action (VLA) models show strong multimodal reasoning ability, their continuous action generation paradigm is not well suited for precision medical tasks, where reliable closed-loop operation may also depend on non-action system function calls. To address this gap, we propose MedVLA, a hierarchical framework that couples high-level multimodal reasoning with low-level function-constrained execution. We further introduce a scalable multi-agent pipeline to generate skill-oriented chain-of-thought(CoT) data for structured training. Built on different multimodal large-model backbones, MedVLA consistently improves performance after fine-tuning, demonstrating the effectiveness of the proposed framework across model variants. Under identical initial conditions, we perform 100 closed-loop flexible electrode implantation trials. The results show that MedVLA achieves a 95.0% task success rate, substantially outperforming representative VLA baselines, including OpenVLA (8%) and $\pi_0$ (15%), in accuracy, stability, and safety. These results indicate that structured reasoning with constrained function-level execution is a practical route toward deployable precision medical robotics.

Applying large language models to Traditional Chinese Medicine (TCM) prescription generation reveals three clinically critical gaps: models produce end-to-end mappings without auditable reasoning following the li-fa-fang-yao paradigm (SR Gap), treat each encounter in isolation without follow-up adjustment via sui zheng jia jian (LA Gap), and fail to enforce absolute contraindication rules such as Shi Ba Fan (SC Gap). We propose a progressive four-stage framework (SFT $\to$ PG-CoT $\to$ Dynamic $\to$ K-RL) that addresses each gap: PG-CoT constrains CoT distillation under the li-fa-fang-yao paradigm to produce auditable diagnostic chains, Dynamic SFT models patient trajectories with explicit transition reasoning, and K-RL encodes deterministic pharmacological rules as rule-based DPO preference signals. Across 12 fine-tuned models and 6 zero-shot baselines, our framework substantially improves prescription quality over zero-shot baselines—with a 7B model (Mistral-7B) surpassing zero-shot GPT-5 on all three TCM evaluation metrics.

Humanoid motion tracking policies rely on dense frame-by-frame references, limiting their use as high-level motion controllers for planning and interactive motion generation. We study Sparse Timed Keyframe Motion Tracking, where a policy receives only sparse future keyframes and their desired arrival times, and must execute stable whole-body motions that reach successive goals. We propose PLAT, a three-stage sparse timed keyframe motion tracking policy learning framework with Privileged LAtent Transition learning. PLAT bridges dense motion tracking and sparse goal-conditioned control by exploiting dense goal sequences as privileged supervision during training while requiring only sparse timed keyframe commands at deployment. A pretrained dense tracking expert first provides robust motion priors. A privileged latent prior is then learned through DAgger-style imitation, followed by latent residual reinforcement learning that refines latent transitions instead of directly optimizing actions. Extensive simulation experiments demonstrate that PLAT maintains accurate and stable sparse timed keyframe tracking across varying planning horizons, with particularly strong performance under long-horizon commands. Successful deployment on a Unitree G1 humanoid robot further demonstrates the effectiveness and practicality of PLAT for sparse humanoid motion control.

Manipulation requires both broad scene awareness and detailed local feedback, yet conventional camera rigs provide them through separate front and wrist cameras. We present Fisheye-VLA, a visual interface that brings these capabilities together using a single passive fisheye. A global view preserves the workspace, while local perspective crops direct detail toward the interaction. The key design question is where this local visual budget should go. We answer it through a controlled re-rendering study, comparing alternative crop directions on the same recorded observations. The study finds that end-effector-centered views capture most of the estimated benefit of a much larger candidate pool, motivating a compact allocation around both hands. Our interface uses calibrated end-effector projection and motion lead to track the crops, while a shared ray encoding preserves their spatial meaning as they move. Integrated with a pretrained VLA, it achieves 84% and 82% success in the two expanded tabletop regions, where some target placements extend beyond the front-camera coverage, and supports shelf and conveyor manipulation. Ablations show that local crops and their viewing directions become more important in the larger workspace regions. The results demonstrate that a single fisheye can support these manipulation tasks without physical wrist cameras.

ICP-based 3D Gaussian Splatting (3DGS) SLAM tracks in real time by registering incoming frames against map Gaussians, using each primitive's covariance for both rendering and registration. These two uses place conflicting demands on one covariance. The mapper shapes it to minimize photometric error, often flattening it against surfaces, while robust registration typically benefits from measurement uncertainty. We propose a dual-covariance parameterization. Each Gaussian keeps a single mean but holds two covariances: a rendering covariance optimized by the mapper, and a tracking covariance derived from an RGB-D sensor noise model. We further use the tracking covariances as Gaussian anchors for image corners, providing constraints in directions where depth geometry is weak. We evaluate on TUM RGB-D, ScanNet, Replica, and two outdoor sequences recorded with a RealSense D435i on wheeled and handheld platforms. We achieve robust tracking performance across multiple scenes and reduced odometry drift, while tracking at $\sim$ 60 FPS.

RandOpt samples weight-perturbed language models and ensembles top-ranked candidates through plurality voting, but its global perturbation scale ignores heterogeneous module geometry. We propose Modular Norm RandOpt, an architecture-aware sampling method using module-wise natural norms and calibrated scales while preserving selection and voting. It outperforms RandOpt using $3\times$ fewer candidates on Countdown and at least $12\times$ fewer on GSM8K, with corresponding wall-clock savings. Evaluations across seven tasks and three Qwen scales ($0.5$B–$3$B) show higher mean accuracy than RandOpt on Countdown, GSM8K, and MATH-500 at every scale. The gains extend to Llama 3.2 $3$B and Gemma 3 $4$B on Countdown and GSM8K. On Qwen2.5-1.5B, our ensembles also achieve higher mean accuracy than iterative baselines on both tasks at comparable main-run evaluation budgets. On GSM8K, a tail-density diagnostic implies only a $1.2$–$1.8\times$ candidate reduction, while most ensemble improvement is associated with more favorable correct-expert support. These results highlight perturbation geometry as a key design choice for population-efficient, gradient-free search around pretrained models.

Interactive segmentation of 3D medical images supports quantitative analysis of anatomical structures and disease while allowing users to specify and refine their targets. Despite substantial progress by nnInteractive and VISTA3D, reliable segmentation across diverse clinical targets remains challenging, particularly for complex anatomical structures and the heterogeneous, long-tailed spectrum of pathology. We present SAMI3D-DW V1 (hereafter SAMI3D-DW), an interactive 3D segmentation model trained on Deepwise's large-scale proprietary medical image datasets. We evaluate the model under simulated user interactions on a CT/MR benchmark comprising 4,326 cases from 219 source datasets, spanning 107 anatomical and pathological categories, organized by a medical taxonomy and evaluated with a category-balanced DSC score. SAMI3D-DW achieves the highest category-macro Dice among evaluated methods in both interaction modes. With one point, it scores 0.5756 versus 0.5316 for nnInteractive, the strongest baseline, rising to 0.7771 versus 0.7495 with five points. With bounding-box initialization, the scores are 0.7129 versus 0.6530. After five corrective clicks, SAMI3D-DW reaches 0.8004 versus 0.7868. For radiologists and clinicians, SAMI3D-DW enables segmentation of complex anatomical structures, including intracranial vessel trees on CT and MR angiography, with a few clicks. In a preliminary in-house comparison involving neurofibromatosis type 1 (NF1), SAMI3D-DW-assisted tumor annotation took minutes per case and approximately one-fifteenth of the time required for manual annotation, highlighting its potential to support volumetric treatment-response assessment.

Generative models have advanced image-conditioned 3D content creation, yet generating controllable and executable 3D scenes from a single image remains challenging. Existing 3D generative approaches can synthesize visually plausible objects and scenes, but their spatial layout estimation is coupled with specific asset generators. They struggle to jointly model object semantics, metric geometry, and scene-level spatial relationships, which are essential for interactive editing, physical simulation, and embodied applications. We propose Fysiverse-3D-Vision, a unified vision-language-geometry framework for generative 3D scene reconstruction and executable asset construction from a single image. We establish a shared representation where spatial reasoning and geometric reconstruction mutually enhance each other, allowing object layouts to be inferred beyond the constraints of individual asset generators. Our model integrates textual supervision, semantic visual cues, and geometric representations within a unified Transformer to capture scene context, metric geometry, and object-level interactions. An object-conditioned layout module performs cross-attention between target object representations and global geometric features to predict object translation, rotation, and scale. Training progressively learns geometry-language alignment, introduces layout reasoning while preserving reconstruction capability, and refines physical consistency through collision-aware optimization. By separating spatial layout reasoning from asset synthesis, Fysiverse-3D-Vision provides an adaptable interface for interactive scene editing, object-level manipulations, and executable 3D content generation. Experiments demonstrate that our framework achieves superior geometric consistency, layout estimation, rendering quality, and physical property understanding compared with existing approaches.

Omni-modal models have expanded multimodal interaction across vision, audio, speech, and language. However, their training is predominantly organized around semantic descriptions and general-purpose objectives, leaving physical attributes, interaction states, and causal mechanisms only partially specified. This gap is not simply a matter of modality coverage: adding more modalities does not by itself provide the supervision needed to connect observations with the physical structure of the world. We present OmniFysics-Nano-V2, a compact omni-modal model for physical-world perception and understanding. The model supports image, video, audio, speech, and text inputs within a shared reasoning framework, together with text and speech generation. To address the lack of explicit physical supervision, we construct a dual-branch physics-aware data pipeline that grounds salient objects in structured physical attributes and aligns visual changes with acoustic events, intermediate responses, and interaction outcomes. To address homogeneous training objectives, we curate reinforcement-learning prompts by reward diversity and adopt a two-stage Group Relative Policy Optimization curriculum that progresses from general task correctness to fine-grained physical perceptual reasoning. Experiments across multimodal, audio-visual, and physical reasoning benchmarks show that the proposed data and training strategy improves physical-world understanding while preserving broad omni-modal competence. The proposed model achieves leading result on 17 of 21 benchmarks against SOTA omni-modal models. By equipping AI systems with both omni-modal and physical-world perception capabilities, OmniFysics-Nano-V2 is poised to become a cornerstone of next-generation Physical AI.

Class-balanced replay controls class frequency but does not determine the interval between successive replay appearances of a class. We study this interval, the rehearsal gap, separately from the class marginal and class co-occurrence, and introduce randomised-pass replay (RPR), which visits each resident class once per shuffled pass. For a fixed set of C resident classes and replay batch size b less than or equal to C, RPR preserves the balanced time-averaged class marginal and bounds every gap by 2*ceil(C/b)-1; a churn-conditional bound applies while the resident set changes. The scheduler uses no future class information and adds no replay examples or forward passes. In a linear-head ER-ACE diagnostic, joint absence from the incoming and replay batches produces a one-sided classifier-bias gradient. Longer absence episodes are associated with larger negative bias displacement, and removing the incoming-loss mask attenuates the scheduling effect. In the primary ER-ACE experiments, RPR improves final average accuracy by 0.72-1.67 percentage points relative to independent class-balanced retrieval under reservoir storage, with positive effects also observed under balanced storage. Pretrained ViTs show positive effects on the tested LT10 streams with small replay batches, while matched larger-batch controls show no material effect. Fixed-cycle and reused-pass controls change more than one temporal statistic, so the experiments do not isolate rehearsal-gap length from all other forms of temporal dependence. The accuracy effects depend on the learner and operating regime.

Deep neural networks are vulnerable to adversarial attacks. Among black-box attacks, targeted decision-based attacks are particularly difficult: the attacker observes only the target model's top-1 label and aims to make it predict a prespecified target class under a bounded perturbation. Before perturbation refinement, the attacker must discover a direction that reaches the prescribed target region. This initialization step can incur substantial query cost. We propose GuidedRay, a targeted decision-based attack based on diversity-guided direction discovery. GuidedRay builds on two observations: target-class reference samples provide useful target-conditioned direction priors, and diverse candidates increase the probability of discovering a targeted adversarial direction. GuidedRay generates varied candidates from one or multiple target-class references and uses a one-query Fast Test to screen their induced sign directions. Once a feasible direction is found, GuidedRay applies Ray Search to reduce its decision-boundary radius. Experiments on CIFAR-10, CIFAR-100, and ImageNet demonstrate that GuidedRay consistently outperforms five state-of-the-art decision-based attacks at four evaluated query budgets from 500 to 5,000, with particularly pronounced gains in direction discovery during initialization. Against models protected by adversarial training or TRADES, it likewise achieves the highest attack success rate at all four query budgets.

Medium-resolution (10-30 m) burned area mapping is vital for monitoring wildfires and their impacts, but remains difficult to scale. Existing methods require either curated fire-specific imagery or dense time-series analysis. Here, we tested whether annual Earth-observation embeddings retain wildfire disturbance signals sufficiently to map burned areas without either requirement. Using Tessera and AlphaEarth embeddings, we tested individual burn-scar delineation, mapping of all same-year fires within an area, regional wall-to-wall mapping, cross-continental transfer, and intra-annual fire timing. Tessera strongly encoded wildfire disturbance, allowing even linear models to separate burned from unburned pixels; the signal was weaker in AlphaEarth. Models trained on a single Tessera embedding matched or exceeded equivalent models using paired pre- and post-fire HLS imagery, and outperformed post-fire imagery alone. The same approach mapped all same-year fires within benchmark scenes (F1 = 0.90). Applied across California, with no California fire data used for downstream training, it recovered 97% of reference burned area and detected substantially more small and medium-sized fires than GABAM or MCD64A1. Separately, a model trained on 2018-2021 US fires transferred without retraining to 88 European fires from 2024-2025 (F1 = 0.88). For well-detected fires, ignition timing was recovered with a mean absolute error of 13 days. Performance declined for fires ignited near the end of the calendar year, and wall-to-wall deployment produced systematic false positives in some unseen landscapes. Annual embeddings nevertheless achieve high segmentation accuracy while moving the burden of dense time series processing upstream, providing a promising path towards simpler regional burned area mapping.

Self-supervised learning for combinatorial optimization has emerged as a promising paradigm for solving discrete optimization problems with neural networks, but a central challenge remains: handling hard combinatorial constraints within continuous, gradient-based training. Continuously extending combinatorial objectives to convex domains is a powerful technique, yet existing approaches often require projection steps that constrain neural network outputs to lie inside the feasible polytope and rely on ad-hoc and problem-specific constructions. We propose a general framework in which the neural network is allowed to predict arbitrary continuous vectors that could potentially lie outside of the feasible polytope. These predictions are then approximated by sparse convex combinations of feasible solutions using a geometric decomposition algorithm based on Frank–Wolfe methods and approximate Caratheodory results. This decomposition induces an a.e.-differentiable, self-supervised loss defined as the expected value of the discrete objective. The same procedure provides an automatic rounding guarantee at inference time. We demonstrate strong empirical performance across multiple combinatorial problems, including the Quadratic Assignment Problem, Maximum Coverage, and the Traveling Salesperson Problem.