让下一篇重要论文，不再错过。

A $360^\circ$ streaming system only needs high quality where the viewer faces, so knowing when a head will turn away matters as much as where it will go. Across four public head-tracking datasets, two used to fix and validate the method and two held out for testing, three patterns emerge on the held-out sets. First, viewers facing the same region turn away within a second of one another far more often than time-shifted traces allow. That estimate is about five times the pairwise threshold for a video timing one departure in ten, mapping to roughly one and a third to over twice that one-in-ten share. The excess co-departing fraction is 0.0422, with a 90% interval from 0.0279 to 0.0565. Second, time since the last large head movement improves held-out departure prediction beyond video and viewer effects. It adds 0.0793 nats per departure, a predictive dependence rather than proof of an internal timer. Third, on the eye-tracked test set, gaze crosses the region boundary after the head starts moving, lagging head onset by 0.30 s. Because headsets record continuous angles, a departure exists only after a rule marks when a settled hold ends once the head swings past a speed and angle threshold. Before touching the test sets, we picked one primary setting for the headline estimates above and tested 18 settings on synthetic traces with planted answers, requiring at least nine passing settings for robustness. Fifteen settings passed for the clock and all 15 valid settings confirm its gain; 8 passed for synchrony and all 8 valid settings agree, one short of the nine-setting bar; and 6 of 7 dual-validated settings agree on gaze lag. Synchrony validation drops from 17 or 18 passing settings under guessed noise to 8 under measured noise because zero-tolerance rules reject a small baseline offset, consistently positive at one dataset size, as datasets grow.

Embedded systems rely on sensor measurements that traverse a complex hardware-firmware stack (e.g. peripheral interfaces, communication buses, memory, and drivers) before reaching the applications that use them. At worst, an attacker with access to this path may successfully intercept, overwrite, fabricate, or replay readings. This may cause systems that act on such data, such as flight controllers and industrial processes, to behave incorrectly or unsafely. Existing defences, such as trusted execution environments and trusted I/O paths, still depend on trusted driver software and correctly configured access control, which remain fragile in practice. In response, we present SensorGuard-RV, a hardware-first approach that delivers sensor samples into processor registers without trusting intermediate software, buses, or memory. After configuration, a write-once ownership switch hands the sensor to a fixed-function controller that performs hardware-level authenticated encryption on sensor samples under a key inaccessible to software. We develop a new RISC-V instruction then instructs the CPU to decrypt the sample and writes it to a register only after verifying its tag and rejecting replays, including across resets. We implement SensorGuard-RV on the NEORV32 core with an ADXL362 SPI accelerometer on an Artix-7 FPGA. At 100 MHz, a protected sample reaches a register in 5,191 cycles (51.9 $μ$s), of which the two Ascon authenticated encryption engines account for 30. Because fixed-function acquisition removes per-byte software polling, the protected loop sustains 19,015 transactions/s, 5.3% above a software-driven SPI baseline. The added logic (3,155 LUTs and 2,215 registers; 5.0% and 1.7% of the device) is dominated by the Ascon engine and preserves the 100 MHz clock, while estimated power rises from 118 to 368 mW.

Linear attention compresses context into a fixed-size key-value state for linear-time training and constant-memory inference. Gated delta-rule models decay this state and erase its prediction for an incoming key before writing a new value; Gated DeltaNet-2 (GDN-2) generalizes this with channel-wise decay and decoupled, full-rank erase and write gates. However, we show that state updates are far less sensitive to erasure than to writes: key activation and unit normalization concentrate key energy onto a sparse subset of channels, whereas value writes scale all channels, so the erase gate can be parameterized at lower rank. Furthermore, slow-decaying state rows accumulate much larger norms than fast rows, dominating unweighted readout and forcing stronger write-time erasure to suppress interference. Guided by this analysis, we introduce Muir, which fuses all gates into a sensitivity-allocated low-rank projection, re-weights state rows with a matched-filter readout gate, and reinvests the saved parameters in a wider feed-forward network. Because standard chunkwise algorithms become numerically unstable at the large chunk sizes needed to saturate wide matrix units, we design a parallel update for stable, efficient training. Across 166M-1.34B parameters, Muir lowers perplexity by 1.8-2.0% over GDN-2 and 2.5-3.6% over Transformer, raises commonsense accuracy by up to 1.9 points, and improves extractive recall over GDN-2. At 1.34B, Muir outperforms both baselines on multi-hop and multi-needle recall, extrapolates 2,560x from its 4K training length to 10M-token streams without perplexity drift, and maintains near-100% passkey retrieval to 1M tokens with static readout. Muir achieves these gains while matching GDN-2 throughput, reaching its 100B-token validation loss 1.2x faster, and training 2.7x faster than Transformer at 32K.

Setting. This is an empirical characterization study; we propose no new method and run no comparisons against alternative objectives. We fine-tune Qwen3.5-4B with sparse LoRA and length-normalized DPO plus an SFT retention term (the RPO objective) on structured financial-filing tasks, and track downstream capability as the implicit reward margin grows to $\sim 7$ $β$-scaled units (nf4 training; $\sim 3$ under fp16 measurement on the unquantized base). Question. Does margin growth degrade capability uniformly across tasks, or selectively? Findings. In a pre-registered study (8 seeds, 6 unseen at freeze) with artifact-aware scoring and TOST equivalence testing, the alignment tax is graded and task-selective: extraction is preserved ($Δ= -0.55$ pts, equivalent at $\pm5$ pts), json loses little ($Δ= -3.52$ pts, equivalence fails on a clustered item-level interval), and router degrades severely ($Δ= -30$ pts). Router's loss concentrates in a margin 1-3 damage window containing a transient format-break spike. The margin is accounted for almost entirely ($\geq 96\%$) by suppression of rejected completions, with chosen reward approximately flat, consistent with the retention term anchoring it; we do not attribute the behavioral damage to any single gradient channel. Gold-logp deflation is task-selective (extraction $-0.15$, json $-1.04$, router $-1.47$ nats/token under fp16), with gold selectively protected relative to incorrect controls on extraction (8/8 seeds). QLoRA reward margins differ up to $3.4\times$ between nf4 training and unquantized fp16 measurement, with an axis-dependent gap. Scope. Length-normalized rewards with SFT retention, sparse LoRA on 8/32 layers, one 4B model, easily-separable synthetic preferences, three structured tasks. A no-retention control and a measured format-alignment metric are left to future work.

Reliable robotic manipulation requires both semantic understanding and precise control in three-dimensional space. However, vision-language-action (VLA) foundations largely inherit visual representations from 2D vision-language pre-training, while spatial supervision is often introduced only during task-specific adaptation. Developing spatial understanding as a transferable foundation-model capability therefore calls for learning from broader robot experience. We present SLA, a Spatial-Language-Action Model that scales implicit spatial alignment to continual pre-training and maintains it during downstream post-training. Initialized from a pretrained pi05 policy, SLA jointly optimizes action prediction and alignment between intermediate visual embeddings and spatial representations from a frozen 3D perception model. We curate and unify heterogeneous robot datasets to support spatial learning across diverse embodiments, tasks, and scenes. To accommodate this diversity, we introduce feature-dependent gates that adaptively aggregate visual embeddings from multiple VLA layers for spatial supervision. The teacher further exploits partially available depth and camera calibration alongside RGB observations, enabling effective use of heterogeneous geometric annotations. Probabilistic state dropout encourages reliance on visual cues and mitigates overfitting to demonstration trajectories. All auxiliary components are used only during training, allowing direct reuse of pretrained VLA weights and preserving the original inference architecture.

Visuomotor policies trained through imitation learning are highly sensitive to changes in camera viewpoint. This is particularly problematic for mobile manipulation (MoMa), where robots rarely execute manipulation from the exact same base pose used during demonstration collection due to navigation uncertainty, localization drift, and environmental constraints. Recent approaches use novel-view synthesis (NVS) to augment demonstrations with unseen viewpoints, but they assume static scenes and therefore fail to re-render the manipulated object during the post-contact interaction phase, leaving the most critical parts of the demonstration unaugmented. We present MOMASPLAT, a novel-view demonstration augmentation framework for mobile manipulation that synthesizes full manipulation trajectories from unseen robot base poses, including dynamic object interactions. Our key insight is that standard MoMa data collection naturally provides the multiview observations needed to reconstruct both the static scene and the manipulated object in 3D. MOMASPLAT combines 3D Gaussian Splatting, feed-forward 3D foundation models, and kinematics-based action retargeting to separately reconstruct the scene and manipulated object, render them from novel viewpoints, and generate geometrically consistent synthetic demonstrations paired with transformed actions. We evaluate MOMASPLAT on five real-world mobile manipulation tasks involving both rigid and articulated objects across three scenes, and validate the fidelity of its synthesized demonstrations against ground truth in simulation. MOMASPLAT substantially improves robustness to unseen base poses, more than doubling the average success rate of the strongest baseline and maintaining non-trivial success at large base displacements where all baselines fail, while reducing the need for largescale real-world data collection.

Current medical benchmarks typically break down clinical competence into isolated factual questions, single-image tasks, or static case vignettes. These tests score individual answers, but reveal less about whether a model can integrate subspecialty knowledge with multimodal evidence and track a patient's condition throughout a course of care. To close this gap, we introduce NPCBench, a clinical apprenticeship benchmark for nasopharyngeal carcinoma (NPC). NPCBench mirrors specialist training, from guideline and foundational knowledge to case planning, longitudinal review, CarePath integration, and open-ended consultation. It contains 4,777 component and consultation items and a 100-patient CarePath cohort with 417 scored diagnostic and care tasks. The episode-level evaluation tests whether diagnostic and care answers are jointly correct within a care pathway. By linking stepwise tasks to patient-level care trajectories, NPCBench evaluates whether models ground decisions in visual evidence, interpret changes in patient state, and meet clinical requirements in consultation. Across 30 evaluated models, we observe a composition gap: strong performance on individual tasks does not ensure jointly correct diagnostic and care answers for the same patient. Among 21 multimodal models, strict CarePath success reaches at most 23%. Paired controls and physician calibration further examine evidence grounding, longitudinal revision, and consultation judging. Overall, NPCBench broadens the evaluation of medical artificial intelligence beyond static question-answering paradigms to an apprenticeship-oriented assessment of clinical workflows, thereby offering a testbed for evidence-based and temporally coherent reasoning in oncology care.

Robotic perception increasingly relies on heterogeneous sensor suites combining cameras with emerging modalities (event, SPAD, iToF/dToF, and thermal). Yet existing simulators idealize these sensors, hiding critical modality-specific failure modes that limit real-world deployment. We present VisionSIM, a unified, physically grounded simulation platform that elevates sensor realism to a first-class concern by modeling the raw physical measurement process for diverse conventional and emerging modalities. We expose fine-grained control over physical sensor parameters, reproducing failure modes overlooked by general-purpose simulators: pileup and glare for LiDAR; phase wrapping, saturation, and motion artifacts for iToF; refractory drops for event cameras; and history-dependent emission for thermal cameras. This enables principled cross-modality comparison on identical scenes and trajectories. We release the platform and a large indoor dataset with synchronized raw measurements and rich ground-truth annotations.

We present the Whiel (a contraction of Whi(le-R)el) system for Lean-certified reasoning about imperative relational programs. The system can encode containment and equivalence problems for Datalog, its extensions with negation, and the core fragment of graph query languages. It can also represent optimized or incremental evaluation algorithms that are not expressible in the original query language, as well as first-order specifications of integrity constraints, schema mappings, and views arising in data integration, exchange, and sharing. Whiel's program states are database instances, and its imperative structure enables compositional Hoare-style reasoning with first-order relational specifications, using prophecy variables and leveled invariants to support inductive arguments. Whiel supports neurosymbolic discovery of loop invariants whose verification conditions can be checked using the Vampire first-order theorem prover. Counterexamples are fed back into the neurosymbolic pipeline, while proposed solutions are translated into Lean proof certificates. To validate the Whiel system, we evaluate its performance on a curated benchmark suite. In summary, we envision the Whiel system as a tool for tackling mathematically demanding, typically undecidable problems involving complex query and constraint languages, with a high degree of trust in its output.

Ordinal information is central to tasks ranging from conference peer review to Reinforcement Learning from Human Feedback for large language models. In many such settings, there is no objective ground truth against which reports can be directly verified, making incentive-compatible elicitation challenging. Peer-prediction mechanisms have been designed to achieve truthful elicitation despite the lack of ground-truth verification. However, existing peer-prediction methods do not provide incentives for agents to truthfully compress rich private information (e.g. reviews) into coarser ordinal reports (e.g. rankings) requested by the mechanisms. We propose a simple ordinal peer-prediction mechanism that rewards agents based on pairwise agreement, measured by Kendall tau. Under a broad class of Bayesian Stochastically Order Consistent models---which includes classical models such as Mallows and Thurstone---we prove that truthful reporting is stochastically dominant, and strongly truthful under index-equivariant strategies. This provides robust incentives even for agents with non-linear utility functions, such as risk aversion. Our mechanism further accommodates heterogeneous agents, top-K reports, and subset assignments, where each agent is scored against one randomly drawn peer. Experiments on synthetic Mallows and Thurstone data, together with a real-data stress test on the SUSHI preference dataset, show right-shifted truthful-payment distributions and the highest normalized sensitivity among tested baselines.

Online 3D bin packing with a processing buffer requires choosing which box to pack, its orientation, and its placement while keeping the stack stable under unknown future arrivals. A collapse ends the episode and forfeits the remaining volume. Existing methods use geometric rules or hard stability masks, limiting how risk can influence the joint choice of box, orientation, and placement. We propose Stability-Hybrid On-Policy Reinforcement Learning (SHORE), which combines a learned stability probability of a placement with an on-policy estimate of its return conditional on stability. A 3D U-Net learns stability from executed outcomes and simulated alternatives, revealing outcomes of actions underrepresented in early rollouts. A discrete on-policy actor--critic uses stability predictions in both its action values and policy prior, with gates that limit the effect of an untrained or changing model. In deployment, SHORE does not require additional search or simulation overhead. In Isaac Gym, SHORE reaches 73\% space utilization by 0.4M environment steps, while the same learner without stability modeling reaches that level at about 1M. On the expanded 60-box set, SHORE achieves 0.734 space utilization versus 0.588 for the GOPT state-of-the-art baseline, with a higher placement-failure frequency (1.03\% vs. 0.63\%), trading more risk for greater packed volume. The results show faster learning and higher utilization on the more demanding realistic box arrival set. Ablations suggest that simulated alternatives provide useful early supervision, while executed outcomes alone sustain later performance. We also demonstrate the method on a real robot to verify the simplicity of integrating the method to real bin packing pipelines.

Quantization reduces the memory and bandwidth demands of large language model (LLM) inference but can degrade accuracy; offloading higher-precision data can instead stall generation. We introduce StrataFlux, a framework that challenges this tradeoff by recovering accuracy lost to low-bit weight and key-value (KV) cache quantization while retaining quantization's memory and speed advantages. StrataFlux co-designs refinement representation, selection, and delivery to make host RAM and solid-state drives useful sources of additional precision. Its selective precision streaming reduces transfer volume, overlaps delivery with computation, and reuses fetched precision across decoding steps so generation need not wait for every transfer. We evaluate 2-4-bit weights and 2-3-bit KV-cache on dedicated GPUs and unified-memory platforms. For 2-bit Llama-3.1-8B-Instruct weights, StrataFlux reduces WikiText-2 perplexity from 9.46 to 8.40 while maintaining 95.4% of the quantized base's throughput on RTX 5090. With 2-bit weights and KV, StrataFlux raises Qwen3-8B's AIME 2024/2025 accuracy from 11.0% to 47.1% at unchanged GPU memory use. At 128K context, Llama-3.1-8B-Instruct with 3-bit weights and 2-bit KV achieves comparable perplexity and speed to vLLM's 4-bit AWQ weights and 8-bit KV on RTX 5090, using approximately half the GPU memory. These results show that GPU-resident precision need not set the accuracy ceiling, nor offloading-tier latency dictate decoding speed.

Masked diffusion language models offer parallel text generation, but independently sampling multiple tokens can disrupt their dependencies and change the output distribution. Exact continuous-time sampling avoids this approximation, yet typically requires one sequential model evaluation per token. We introduce SpecCTMC, a training-free speculative sampler that reduces this sequential bottleneck without a separate draft model. SpecCTMC proposes successive token updates using previously computed model predictions, evaluates their intermediate states in parallel, and applies an exact acceptance-and-correction procedure. We prove that SpecCTMC preserves the joint distribution of event times, generation order, and token values of the underlying continuous-time Markov chain. It therefore introduces neither time-discretization error nor additional factorization error from parallel decoding. Experiments on LLaDA-8B-Instruct and Dream-v0-Instruct-7B across five benchmarks demonstrate 1.73-4.38x fewer sequential model evaluations than direct continuous-time sampling, while preserving the reference sampling distribution. Our code is available at: https://github.com/kasakh/specctmc

Cooperative localization uses odometry and inter-robot measurements to estimate the motion of a robot team. These measurements describe only relative geometry, but the team's estimated position and orientation can drift together over time, i.e., gauge drift. In this work, we study correction of this gauge drift using a radio attenuation map registered in the same world coordinate frame used for robot poses. Specifically, we propose Radio-to-Pose Information (RPI), an interface that converts power residuals and their pose derivatives into a Gaussian information matrix and vector pair. RPI builds on the insight that received power is sensitive to shared displacement in the world map. The pose estimator uses the interface together with odometry and inter-robot measurements to update poses and minimize uncertainty. We characterize the shared motions detectable from the map and observed links. For an uncertain shared map, RPI keeps correlations across robots and time and reproduces joint Gaussian pose estimates and covariances. Across $1000$ simulated Gaussian trials, RPI matches the results of joint estimation. Across 32 paired trials with a registered map, RPI reduces common-position RMSE from $0.542$m to $0.140$m in a $50\times50$m arena compared with a baseline that uses radio only to weight relative measurements.

Oral cancer and the oral potentially malignant disorders (OPMD) that can precede it are visible on the oral mucosa, which makes smartphone photographs a promising tool for screening. Deep learning models for this task classify individual photographs. However, screening refers patients, not photographs: a patient is usually photographed several times, and the referral decision must miss no more than a set share of patients with disease. We propose OraRefer, a patient-level screening model that reads each photograph with a segmentation-guided foundation model that conditions its class decision on the predicted lesion, combines all photographs of a patient with risk factors, and refers through a threshold with a finite-sample sensitivity guarantee. The model localizes the mouth itself, so inference needs no annotation. On 3000 photographs of 714 patients, patient-level fusion reaches an AUROC of 0.919 and outperforms attention-based multiple-instance learning. At a 90% sensitivity target, OraRefer detects 95% of patients with OPMD or cancer and misses 3 of 61 patients with cancer, at a specificity of 0.610. Because the threshold is calibrated without retraining, each clinic can set its own guaranteed operating point on a local calibration sample.

This paper analyzes a bilevel formulation for learning the regularization parameter of combined local-global (CLG) optical flow models. The lower-level problem computes the flow from a Lucas--Kanade data term and a quadratic or total variation (TV) regularizer, and the upper-level problem selects the parameter that minimizes the squared distance to a ground-truth flow. The setting is finite-dimensional, and a full-rank assumption makes the lower-level solution unique. For the quadratic model we derive sufficient conditions for the existence of an optimal parameter, for every optimal parameter to be positive, and for strict local optimality. For TV regularization an optimal parameter always exists, because the lower-level solution vanishes for large parameters, and we give a sufficient condition for positivity that accounts for the coupling of the two flow components through the local structure tensors. Synthetic examples illustrate the conditions, including cases in which they fail.

At the edge of stability, the iterates of gradient descent oscillate along the sharp directions of the Hessian. This poses a significant challenge for continuous-time models as the usual continuous-time approximation -- the gradient flow -- admits no oscillations. Several continuous-time models have been proposed to capture the behavior at the edge of stability, among them central flow and rod flow. Although both models can be interpreted as performing a coarse-graining in time, the connection between them remains poorly understood. Here we show that both flows can be derived from a third object, the edge coupling: a functional on consecutive iterates whose critical points are the fixed points and period-two orbits of gradient descent. It is revealed that the two flows correspond to coarse-grainings at different time scales. Rod flow coarse-grains over only the fastest time scale, while central flow coarse-grains over both the fast and intermediate time scales. Furthermore, using the edge coupling, we derive rod flows and central flows for momentum methods such as heavy ball and Adam. We verify the derived flows experimentally on representative neural network architectures, where they track the discrete iterates throughout all phases of training -- including the transition from the stable regime to the edge of stability.

Given a fixed straight-line drawing of a complete graph, we seek a two-colouring of its edges that minimizes monochromatic crossing pairs. We prove that the threshold decision problem is NP-complete, even for points in general position with integer coordinates. Thus exact minimization is NP-hard, resolving the complexity question posed at the Crossing Number Workshop in 2019. We reduce from MaxCut on permutation graphs by representing source vertices with classes of geometric edges, whose crossings encode source adjacencies. The target complete graph also contains every other edge between its points, and their freely chosen colours can affect the encoded cost. To control these interactions, we construct a large comparison grid with two stable colour phases and use program points to turn this reference into colour preferences on other edges. The optimum differs from a computable baseline plus the amplified minimum number of uncut source edges by an error small enough to distinguish consecutive source values. The reduction is deterministic, with polynomially many points and polynomial coordinate bit length.

In data clustering, it is often desirable to find not just a single partition into clusters but a sequence of partitions that describes the data at different scales (or levels of coarseness). A natural problem then is to analyse and compare, as a whole, the (not necessarily hierarchical) sequences of partitions that underpin such multiscale descriptions. Using tools from topological data analysis, we introduce here the Multiscale Clustering Filtration (MCF), a filtration of abstract simplicial complexes that encodes arbitrary cluster assignments in a sequence of partitions across scales of increasing coarseness. We prove the stability of the MCF with respect to small changes in the sequence of partitions, and establish that its persistence diagram (PD) provides an interpretable representation of the multiscale clustering: at dimension 0, it captures violations of the refinement order, and at higher dimensions, it tracks the emergence and resolution of higher-order inconsistencies between cluster assignments. We also introduce an equivalent construction of the MCF via a nerve complex filtration, and show that this nerve-based MCF has computational advantages in certain settings, aiding scalability. Further, we show that in the hierarchical case, the MCF reduces to a Vietoris-Rips filtration of an ultrametric space, and we establish that pairwise measures, such as conditional entropy or zigzag persistence, are fully determined by a graph representation of the Sankey diagram and thus fail to describe the higher-order inconsistencies across scales that MCF captures. Using synthetic and real-world data from stochastic block models, single-cell transcriptomics and temporal social interactions, we illustrate how the MCF PD serves to characterise and compare multiscale clusterings beyond pairwise or hierarchical baseline methods.

Background and Objective: Our main goal is ultrasound guidance: moving a probe until its image matches a target view. Published methods for each part of a guide are evaluated in separate simulators. We asked whether these methods, combined into one guide, are good enough. Methods: We built USB Gym, a testbed whose standard engine simulates ultrasound from public patient CT, with a network drawing organ textures from CT labels. Its guide has three replaceable parts (slots), an organ segmenter, a controller and a reward, which received 20 published methods, each run on the same 60 episodes of six held-out patients. In each slot, a published method replaced ours only if detectably better on these patients. The better of this combination and our original guide on 20 new CTs is the benchmark baseline. A pose oracle, reading the exact probe pose available only in simulation, set the reference. The results below are exploratory. Results: On these CTs, the probe started on average 16.0 mm from the target view; the pose oracle brought it to 1.5 mm. Of the results known when the combination was fixed, only an image-similarity reward (SSIM) was slightly but detectably better, so the combination was our guide with that reward. This combination was the benchmark baseline, ending at 13.7 mm (16% of the oracle's progress). Conclusions: Combining untuned published methods with ours did not guide the probe near the target view; the task remains under-explored. The testbed accepts new methods; no guide was tested on hardware.

Can a regulator promise both a risk ceiling and a development deadline when safety productivity is unknown? A ceiling below the final model's unprotected hazard requires a minimum stock of safety knowledge, so both promises hold only if weak research can be ruled out. Learning first and then replaying development, with spare compute in safety, comes close to that minimum. In a calibration with only state risk, constant safety yield, and full knowledge transfer, it needs only 3.3 percent more productivity than the necessary bound. Customer services pay for the guarantee. Rules that fix compute allocation fix dates, not risk.

Millimeter-wave radar resolves the micrometer displacements of a moving body as micro-Doppler, enabling gait, gesture, activity, pose and vital-sign sensing. Each task relies on a downstream model trained on real data, yet radar datasets lag orders of magnitude behind vision and language. Brute-force collection will not close that gap, and existing synthesis methods (forward-only simulators, learned generators and static-scene inverse renderers) cannot fit a real capture of a moving body, leaving a sim2real gap. We present 4D Point Splatting (4DPS), a physics-based differentiable renderer that fits an explicit moving scene to a real capture. Using the synchronized camera to initialize the room and body, we represent the scene as oriented points carrying material, normal and velocity, and splat each point into range and Doppler bins. Our output is the raw complex return a real radar records, so we apply the same signal processing on the rendered data to obtain any representation for training any SOTA downstream model. We fit 235 sequences on HuPR and 100 on RT-Pose, two datasets with different radars. Our reconstructions set the state of the art, with median correlations of 0.940 and 0.830 on range-azimuth, 0.829 and 0.695 on range-Doppler, and 0.988 and 0.966 on micro-Doppler. Our renders also set the state of the art in downstream sim2real gap. On RT-Pose, a pose model trained on them reaches 21.9 cm MPJPE on real data, against 16.7 cm for real training and 127.6 cm for the RF-Genesis baseline, where both the rendered- and real-trained models learn nearly the same features. Our custom CUDA kernels fit a sequence in 5.7 minutes on HuPR and 12.9 on RT-Pose on one H100, making dataset-scale inverse rendering of mmWave radar possible for the first time. This work can enable synthetic data generation for any human-centered downstream task and model, and scene editing for data augmentation.

We introduce GPWM, a Generative Particle World Model that predicts how 3D scenes evolve by generating the motion of particles and rigid bodies over time. Unlike video world models that generate scene appearance, GPWM operates directly in 3D physical space. A single representation and architecture cover rigid, deformable, cloth, and granular materials, in both passive physics and robotic interaction. GPWM is an object-aware multi-resolution transformer that combines flow matching with diffusion forcing to generate coherent physical trajectories far beyond its training horizon. We train and evaluate GPWM on ParticleWorlds, a 25.5-million-frame corpus spanning diverse geometries, materials, interactions, and robot embodiments. GPWM outperforms the best prior learned-dynamics baseline in each of six physical domains, reducing prediction error by 31% on average. Breaking from the domain-specific setting of prior methods, a single Unified GPWM models these heterogeneous regimes with one set of weights, largely outperforming per-domain baselines and even surpassing our own specialist models in two domains. GPWM further generalizes to unseen geometries, physical properties, object counts, and rollout horizons. Beyond prediction, GPWM enables contact-rich planning by learning a generative prior over joint actuator-object motion. Instead of searching over unconstrained actions, planning searches the model's latent noise for interactions that achieve a desired object-space goal. With the same world model, noise-space planning substantially outperforms action-space search on pushing and picking, where complex contact sequences are hard to discover. Executed open-loop in simulation, the resulting plans achieve goal-directed manipulation without reference action trajectories or goal-specific policies.

We study learning rate (LR) scaling for hybrid architectures combining Transformer and State-Space Model (SSM) blocks, a class adopted by several recent production language models. In particular, we focus on the gap between the theoretical scaling rules derived for SSMs under zero-order-hold (ZOH) discretization at infinite width with growing state size, and the field-standard practical implementations using simplified-ZOH Mamba at fixed state size. Surprisingly, in this practical regime hybrid architectures achieve a near-zero LR transfer gap across widths 256-2048 and depths 4-32 up to billion-parameter scale using only the original $μ$P prescription, even though SSM operations fall outside its Tensor Programs representability conditions and every parameterization we test fails the standard coordinate-check diagnostic of $μ$P correctness. We attribute this to a two-condition decomposition of LR transfer in hybrid architectures: a global update-to-weight invariance, enforced by $μ$P's initialization and LR scaling; and a local per-component balance, provided by AdamW's per-parameter normalization. Our observations show that the optimal LR is invariant to width up to 8$\times$, that this width invariance holds across depth, sequence length, batch size, and Transformer-to-SSM ratio, and that it transfers to Nemotron-H, a production hybrid outside our custom architecture set. We hope these findings fill the gap between theoretical scaling rules and practical hybrid implementations, and stimulate further research toward bridging it.

Much of human culture is tacit and is typically learned through situated social interaction. As LLMs are deployed as interactive agents across diverse communities, assessing their cultural competence demands moving beyond existing question-answering approaches, shifting from what they know to how they adapt. In this work, we broaden the scope of cultural adaptation in NLP by investigating whether LLMs can adapt through repeated interactions via a sandbox environment. By reframing cultural adaptation as a continuous process of norm assimilation through social interaction, our work demonstrates that memory mechanisms foster longitudinal learning across repeated interactions, facilitate transfer to different social settings, and enable multilingual norm assimilation across five languages. Together, our findings lay the groundwork for continuous, interaction-driven cultural adaptation in LLMs.

Retrieval-based factuality evaluation, where LLM-generated claims are verified against evidence from authoritative medical corpora, has become the dominant paradigm for scalable hallucination detection in high-stakes clinical settings. Despite the urgency of reliable and transparent medical fact verification, most systems measure performance with aggregate metrics like F1, which obscure where and why failures occur. Existing RAG diagnostics require gold answers or annotated gold evidence, neither of which exists in this regime. We introduce two comprehensive taxonomies, grounded in a case study on the open-ended MedExpert dataset and 3 closed-ended datasets, decomposing failures into retrieval-stage errors along five quality dimensions, and verifier-reasoning errors into six consecutive steps. We adapt an automatic pattern induction pipeline using LLM-as-Judge to label evidence quality and classify verifier reasoning errors at scale, and then stress-test our findings across 4 retrieval methods and 6 frontier verifier models. Our analysis reveals that scaling model size, adding reasoning effort, expanding to authoritative web sources, and applying medical fine-tuning do not resolve these failure modes, demonstrating that they represent fundamental limitations of the retrieve-then-verify paradigm in open-ended medical settings rather than artifacts of outdated systems. We release our code and data at https://anonymous.4open.science/r/Medical_RAG_eval-4AB5 for the full reproducibility of our results.

Vision-language models must reconcile visual evidence with memorized world knowledge when the two conflict. How they resolve this conflict shapes the reliability of multimodal systems, yet prior work characterizes it behaviorally without a component-level causal account. We combine activation patching across three granularities (residual stream, attention heads, and MLP sublayers) with model-component ablation studies and mechanistic analysis. Across three VLM families, we find that visual grounding emerges by default, whereas prior grounding depends on a small set of causally necessary attention heads (2.5-4.8%) concentrated in the second half of the network. These heads enable answers from stored world knowledge (e.g., "red" for a strawberry) despite conflicting visual input. Ablating them flips predictions from knowledge-grounded to visually grounded answers in 68-96% of cases under prior-knowledge prompts, but changes only 0.8-7.5% of visually grounded predictions, establishing an asymmetric causal structure. The identified heads decompose into routing heads, which modulate information flow, and writing heads, which directly project answer tokens into the residual stream. This structure is consistent across model families and scales, revealing a sparse causal circuit underlying perception-knowledge conflict in VLMs.

Hand avatar reconstruction typically relies on multi-view capture or per-subject optimization, limiting its scalability. We introduce InstantHand, a feed-forward framework that constructs an animatable 3D Gaussian hand avatar from a single image using off-the-shelf MANO, camera, and mask estimates, without test-time optimization. Image-averaged losses provide insufficient supervision for alpha leakage, which thickens thin finger contours. To address this issue, we propose a Soft Alpha Band regularizer that concentrates supervision on silhouette boundaries. Furthermore, colors regressed from image features can drift toward a learned prior even in observed regions. Our Reliability-gated Appearance Transfer preserves observed appearance by transferring source colors according to each Gaussian's rendering contribution, while completing unobserved regions using the learned prior conditioned on global source identity. InstantHand constructs an avatar in 2.84 seconds and renders at 161 FPS, matching per-subject-optimized baselines on InterHand2.6M and remaining competitive in zero-shot evaluation on HanCo.

Instructional videos are the dominant medium for completing physical tasks, yet rarely match the user's visual context. Motor simulation and cognitive load theories predict this mismatch matters, but how much, which attributes are responsible, and how users experience it remain unknown. We conduct two studies (56 participants, two first-aid and two cooking tasks) using researcher-prepared recordings to control visual alignment. Study 1 (N=16) compares fully aligned In-Context instructional videos (ICON) against business-as-usual Internet videos. ICON yields statistically significant gains: 11.1% higher completion quality and 15.5% faster completion. Qualitative analysis decomposes the effect into four visual context attributes: Task Object Intrinsic, Task Object State, Environmental Context, and Observational Context. Study 2 (N=40) ablates attributes one at a time. All four degrade objective performance yet users perceive no performance loss for any of them. Visual context misalignment is substantial, decomposable, and invisible to the user for any single attribute.

Sentiment Analysis (SA) is a fundamental NLP task concerned with the identification and categorization of subjective information in text. While lexicon-based methods offer transparency, they often struggle with linguistic phenomena such as negation and intensification. Syntactic parsing has been integrated into SA pipelines to address this, but conventional parsers introduce significant computational overhead. Recent work shows that reformulating dependency parsing as a Sequence Labeling (SEL) task can improve efficiency. We propose and compare two strategies for integrating syntactic knowledge into a neural sentiment analysis model under unified Multi-Task Learning (MTL). The first, Standard-MTL, jointly learns dependency parsing (via sequence labeling) and sentiment polarity classification using a shared encoder with two task-specific heads. The second, Contrastive-MTL, provides implicit syntactic guidance through a contrastive auxiliary objective and a syntax-biased attention pooling mechanism, without requiring explicit label prediction. Both rely on syntactic annotations during training and eliminate need for hand-crafted compositional rules or external parsing at inference. We evaluate eleven datasets spanning seven languages---the MARC corpus (English, Spanish, French, German) and seven datasets from SemEval-2022 Task 10 (English, Spanish, Norwegian, Basque, Catalan)---using four established SEL encodings. With the relative encoding, Contrastive-MTL outperforms a syntax-naive baseline on 10 of 11 datasets and surpasses the external rule-based pipeline (SELSP) that parses at inference on all six datasets where the latter applies, while being 1.5--9$\times$ faster at evaluation. Standard-MTL is more dataset- and encoding-dependent. These results show implicit syntactic supervision during training can match or exceed explicit parsing at inference without computational overhead.