Science & Mathematics

A global $\ell_1$ budget limits how much an adversary may change an image, not how that change is distributed. We ask how much of an $\ell_1$ attack's observed success remains when large coordinate changes are restricted. We add a per-coordinate cap to the budget, enforce the joint constraint by exact in-loop projection, and sweep the cap for APGD-$\ell_1$, SLIDE, and FMN-$\ell_1$ without altering their objectives. Across 18 standard-model settings, caps of at most $32/255$ retain a median $97\%$ of attack success relative to each method's native uncapped baseline, while median peak magnitude falls by $80\%$ and median effective support nearly triples. At matched realized $\ell_1$, capped perturbations remain structurally distinct; separate ImageNet comparisons also show lower LPIPS and DISTS at selected caps. Direct in-loop constrained optimization preserves up to $88$ percentage points more attack success than post-hoc clipping, highlighting the importance of constraint enforcement. Suitable caps nevertheless vary across attacks and models, and adversarially trained classifiers exhibit more heterogeneous concentration sensitivity. These results establish perturbation concentration as a complementary, model- and attack-dependent axis of $\ell_1$ attack evaluation.

3D Human Pose Estimation (3D HPE) is used in indoor environments, such as smart homes, to analyze human movement and monitor human states. Millimeter-wave (mmWave) radar is suitable for this task because it is robust to lighting variations and does not capture images of human appearance. However, radar observations often provide insufficient information about certain body parts or contain overlapping reflections from multiple body parts, limiting accurate 3D joint localization. To address these limitations, we propose the Initialization-to-Refinement Pose Framework (I2R-Pose), comprising two stages: Motion-Aware Pose Initialization and Observation-Guided Refinement. In the first stage, joint-wise features are learned by incorporating temporal relationships across consecutive radar observations and human skeletal structure to compensate for insufficient joint information and estimate an initial 3D pose sequence. In the second stage, separately extracted radar observation features predict joint-wise positional corrections that refine the initial pose and generate the final pose sequence. The framework supports individual and fused inputs of time-Doppler/time-range (TD/TR) maps and radar point clouds (PCs). To evaluate performance across indoor activities, we construct the Radar Motion and Spatial Information for 3D Human Pose Estimation (RaMoS-Pose) dataset, providing synchronized TD/TR maps, radar PCs, and 3D joint ground truth (GT). Evaluations on RaMoS-Pose and the publicly available MVDoppler-Pose dataset show that I2R-Pose reduces mean per-joint position error (MPJPE) by 39.50% on MVDoppler-Pose and by 15.13% on RaMoS-Pose with fused inputs, compared with the best-performing existing methods evaluated on each dataset.

Neural Radiance Fields (NeRF) achieve high-quality novel-view synthesis by representing a scene as a continuous volumetric radiance field. However, the geometry encoded by the density field of vanilla NeRF can be noisy and inaccurate, since different combinations of shape and radiance may explain the same input images. To reduce this ambiguity, we augment vanilla NeRF with a geometry-optimization branch based on self-supervised cross-ray consistency. Specifically, this branch samples candidate surface points from the density field, generates virtual rays that intersect at each point from different directions, and renders their appearance using the NeRF network. Based on the visibility-weighted color consistency across these rays, we constrain the implicit geometry. We further regularize the implicit geometry by penalizing opacity at candidates with inconsistent appearance predictions and encouraging local normal smoothness. These geometric constraints are derived from the NeRF predictions and require no ground-truth depth or surface-normal supervision. Experiments on eight Replica scenes show that the proposed method reduces the mean Chamfer distance by 30.8% compared with vanilla NeRF.

Tool-using LLM agents increasingly answer e-commerce queries through multi-step tool interactions. The expected output of a tool call can shift between agent development and execution, allowing a plausible but incorrect value to be returned without an explicit failure signal. These silent errors may propagate through subsequent decisions into final answers. To study agents' resilience across this end-to-end process, we formulate Silent-error Resilience Task, an interactive task that evaluates whether they complete the original query after a silent error corrupts an intermediate state. We introduce RITE, a benchmark built from human-authored industrial workflows and production-log-derived inputs. To enable controlled and reproducible evaluation, we use a semantic error taxonomy to inject errors into fixed workflow prefixes, freeze recorded tool behavior, and let agents continue freely. It contains 167 matched units across nine e-commerce workflows, yielding 501 silent-error, correct-value, and explicit-failure instances. Across 10 recent LLMs, silent errors roughly halve task success and are consistently harder than explicit failures. Recognition, use of an available verification path, and integration of the resulting evidence form successive bottlenecks. We also propose a light State Checker as an intervention to test whether resilience can be improved.

Training LLM agents for policy-constrained tool environments, such as an airline backend with booking, refund, and cancellation policies, requires tasks that exercise the environment's rules individually and in combination. Existing synthesis pipelines let a model propose tasks and retain only those it immediately solves, biasing training data toward easier behaviors. We introduce Gradus (Graded Rungs of Ascending Difficulty from an Underlying System), which decouples what a task should teach from the concrete task and solution used to teach it. Without seed tasks, Gradus treats a fixed environment as a latent syllabus, extracting guarded behavioral operators that capture how actions depend on current conditions and how their effects reshape subsequent decisions. It composes these operators into increasingly demanding tasks while using corpus-level feedback to target missing outcome combinations. Crucially, Gradus jointly calibrates tasks and solutions against the environment: execution feedback identifies which side to refine while preserving the intended behavioral requirements. Verification thus becomes a recovery signal rather than a terminal filter, coupling coverage-guided exploration with the preservation of difficult examples. From off-the-shelf backbones, the resulting curriculum lifts a 30B model from 59% to 89% on $τ^2$-Bench and from 47% to 83% on CAR-Bench with an Opus 4.5 proposer, and to 76% and 62%, respectively, with an open-weight Qwen3-235B proposer. These results show the value of jointly controlling which behaviors are explored and which survive into verified training data.

Video generation models are promising foundations for embodied world models, but only if their generated rollouts are physically valid and faithful to task instructions. Multimodal large language models (MLLMs) are increasingly used to judge these properties, yet their agreement with human judgments on embodied videos has not been systematically evaluated. To address this gap, we introduce RoboJudge-Bench, a benchmark for evaluating MLLMs as judges of embodied video generation. The benchmark measures two fundamental dimensions, Physical Adherence and Instruction Alignment, across 800 human-annotated videos from 8 source corpora and 10 video generators, spanning grippers, human hands, and dexterous hands. Experiments reveal uneven alignment between existing judges and human ratings: general-purpose MLLMs show lower agreement on physical adherence, while specialized video judges designed for general-domain physical reasoning do not consistently transfer to embodied scenarios. Motivated by these findings, we construct RoboJudge-Train, comprising 12K videos annotated with 25K human-verified scores and model-assisted rationales for Physical Adherence and Instruction Alignment. We fine-tune a 9B MLLM on this dataset with supervised learning followed by reinforcement learning, yielding RoboJudge-9B. Among the model judges evaluated on RoboJudge-Bench, RoboJudge-9B achieves the highest overall agreement, reaching a Pearson correlation of 0.719 with a statistically significant 0.089 gain over Gemini-3.7-Flash, the best-performing proprietary baseline in our evaluation. Training on RoboJudge-Train also yields gains on external benchmarks, including VideoPhy2 and Physion-Eval. We further use RoboJudge-9B as a downstream evaluator to benchmark video generation models according to Physical Adherence and Instruction Alignment.

Empirical scaling laws relate loss to parameters $N$, training tokens $D$ and compute $C$. This technical note explains three mechanisms behind their exponents. Local approximation bounds function error at rate $N_e^{-γs/d}$, with effective parameters $N_e$, smoothness $s$, reuse exponent $γ$ and dimension $d$; the loss function determines the risk bound. Independent feature coverage gives $α_D=(β-1)/ζ$ from predictive-weight decay $β$ and occurrence rarity $ζ$. Separable-loss optimization determines training allocations and deployment crossovers. Nonlinear scoring can produce apparent emergent abilities, while grokking concerns delayed generalization and verification constrains test-time scaling. For regular molecular configurations, the $\mathrm{SE}(3)$ quotient gives $d_{\mathrm{int}}=d-6$ and relative approximation-exponent gain $6/d_{\mathrm{int}}$. Optimized dense-to-equivariant excess-loss ratios scale as $C^{p_{\mathrm{eq}}-p_{\mathrm{dense}}}$, where $p_j$ are compute-loss exponents and the prefactor depends on loss amplitudes and FLOP costs. Domain examples also cover visual resolution, conformational coverage and correlated-sample precision.

Large language models (LLMs) exhibit both knowledge of language, rendered by, for example, grammatical judgment ability, and abilities to perform tasks through language, such as math-word problem solving. Motivated by neurolinguistic studies of their neural dissociation in human brains, in this study, we investigate the overlap of their circuits in large language models (LLMs). Specifically, we examine the overlap and causal effects between the circuits extracted for linguistic and arithmetic tasks, using controlled materials and circuit discovery methods from the mechanistic interpretability field. Our experiments with five LLMs show that circuit similarity is mostly explained by task domains (linguistic vs. arithmetic), rather than by their input and output vocabularies or by the syntactic similarities between natural-language sentences and arithmetic expressions. This finding suggests that LLMs may process linguistic and arithmetic information using largely separate neural circuits, highlighting an ability-based modular organization of circuits within these models.

Uncertainty in the terrestrial carbon cycle remains a major constraint in climate projections, partly due to imperfect land surface representations in Earth system models. We propose a data-driven framework for high-resolution land-use reconstruction, treated as a semantic-segmentation task, that combines coarse-resolution scenario data and climate classification with static geophysical features to produce annual land-use maps at 1 km resolution. We introduce TerraNostra-Bench, a change-aware evaluation protocol that separates spatial downscaling from autoregressive temporal rollout and scores models on the rare land-use-change pixels rather than the dominant static background. Using it, we compare convolutional (U-Net) and transformer (Swin) backbones in autoregressive and non-autoregressive schemes against persistence, linear, and shallow convolutional baselines. The autoregressive prior is the most informative input yet accumulates error over long rollouts: autoregressive models lose 5.7-7.2 accuracy points over 13 years, against 2.2-2.8 for non-autoregressive models. This trade-off is governed by the autoregressive choice rather than the backbone. It can be tuned by an inference-time prior-trust threshold that requires no retraining and by the training-time weighting of reconstruction and change objectives; the two levers trace complementary frontiers, with calibration recovering up to 10 accuracy points while keeping change sensitivity above the persistence floor. Extrapolating beyond the training window, backward to 1955-1959 and forward to 2015-2019 under scenario forcing, the models retain most of their in-window skill and show the same trade-off, indicating it is intrinsic to the design. This work provides the methodological foundation and evaluation surface for a forthcoming temporally continuous land-use product.

Action-free videos are an appealing substitute for costly teleoperated data. Latent action models (LAMs) turn videos into a supervision signal by inferring abstract actions between frames. Existing LAMs observe a single camera view, which resolves 6D robot actions only ambiguously missing out-of-plane movements and potential occlusions. We propose a LAM that ingests synchronized multi-camera views using cross-attention to resolve depth and occlusions before compressing them into latent actions. The latent actions are fed into a policy in which a latent expert and an action expert are coupled by sequential flow matching around a frozen VLM backbone. We further evaluate in strict cross-embodiment settings where task-specific demonstrations come from other embodiments. We show that dual-view LAM pretraining improves the decodability of ground-truth 6D actions by 17% compared to single-view, while scaling single-view data by two orders of magnitude is not sufficient to fully recover out-of-plane motions. The downstream policy improves success rate by +13.5% compared to +6.2% for single-view LAM. A single cross-embodiment is sufficient for skill transfer, and adding a second viewpoint outperforms adding up to four further embodiments. The results transfer to hardware where demonstrations come from a human hand improving success rate by +20% over action-only policies in 850 rollouts. We show that video and robot data are complementary, and that benefits of learning from video are task-dependent. Website: https://learnsyslab.github.io/multiview_latent_action_pretraining/

6G will be designed faster than it can be built: AI-driven search already proposes physical-layer algorithms at machine speed, while carrying a single one onto embedded baseband silicon remains specialist work. That silicon imposes hard memory and numeric limits, but also exposes cost structures a GPU lacks, and a design flow that never runs on the target sees neither. We present an agentic workflow in which an LLM coding agent implements algorithms on a commercial baseband SoC, optimizes them down to agent-written vector assembly, and validates them on silicon. The agent is guided by a knowledge base of measured platform facts, checked at every step against evaluators it is instructed never to edit, and metered per phase. We showcase the workflow on an AI-discovered orthogonal time frequency space (OTFS) detector portfolio on the open 6G4All platform. It yields twelve kernels that run cycle-exactly on silicon, their optimized versions at 18k-405k cycles (30-660 $μ$s) per frame. It also yields a new detector shaped by the chip's measured costs that needs 13.9% fewer cycles than the port of the discovered detector while matching its error rate, with a measured edge our evaluation cannot statistically resolve, a structure that a hardware-blind search objective had no reason to find. What is not specific to OTFS is the workflow: an LLM agent exploring an implementation space under real hardware limits, with the cost of every choice measured rather than assumed.

Real-time hybrid-functional calculations require repeated reconstruction of nonlocal exchange, whose cost grows rapidly with Brillouin-zone sampling. We develop two methods to improve the efficiency of time evolution and spatial processing. The first learns the time evolution of adaptively compressed exchange (ACE) factors and predicts their updates, reducing the frequency of explicit reconstruction. The second uses crystal symmetry to reuse exchange fast Fourier transforms (FFTs) among equivalent real-space points. This symmetry-based FFT reuse (SBFR) reduces the number of FFTs while retaining the full $\bm k$-point sum. The methods can be used with a variety of hybrid functionals, and we use HSE06 in this work. Relative to an ACE reference using full-row mesh FFTs, learned prediction provides more than fivefold acceleration, SBFR more than fourteenfold, and their combination more than sixty-threefold. The combined method shows efficient parallel scaling. After learning, propagation takes less than 0.5 s per step. The reduced computational cost allows us to study nonlinear and nonequilibrium electron dynamics with hybrid functionals and to examine the role of excitonic effects in semiconductors and insulators.

EEG visual decoding remains difficult because scalp recordings are noisy, spatially coarse, and variable across trials and subjects. Most systems align each trial with a dense image embedding, although that embedding mixes semantic content with instance-specific texture, pose, and background. This target can demand details absent from EEG and provides no explicit account of the visual factors supporting a prediction. We propose \textbf{ProtoEEG}, a prototype-mediated framework that reformulates EEG--image alignment as soft composition over a shared learnable visual codebook. A trainable EEG encoder and a frozen visual teacher produce assignment distributions over the same $K=64$ prototypes. Their weighted reconstructions form a structured representation for retrieval and image generation. Prototype contrast supplies the main cross-modal training signal, distribution alignment coordinates codebook use across modalities, an instance anchor preserves fine-grained discrimination, and diversity regularization prevents redundant prototypes. The shared assignment profile also makes the learned bottleneck directly testable through prototype retention, random selection, and identity shuffling. Across five seeds, ProtoEEG reaches 64.0\% Top-1 and 91.5\% Top-5 on THINGS-EEG2 200-way retrieval. After retraining on EEG-ImageNet, it obtains 44.8\% Top-1 and 73.8\% Top-5 in 80-way retrieval. ProtoEEG also ranks first on all seven reconstruction metrics. Removing prototype contrast lowers Top-1 by 56.9 percentage points, while shuffling prototype identities reduces it to 0.5\%. Prototype analyses further show distributed assignments and broad codebook use rather than one-prototype routing. These results support structured alignment targets as a practical alternative to increasing encoder or teacher capacity alone.

Face presentation attack detection is typically studied under fixed camera or handheld acquisition, whereas smart glass capture introduces a more challenging setting with ego motion, changing viewpoint, background clutter, distance variation, and less controlled illumination. To address this challenge, we present \emph{SmaGla-FPAD}, a smart glass based face presentation attack database acquired using Ray-Ban Meta Wayfarer Smart Glasses across indoor, outdoor, and in-car settings. The database contains 297,000 sample images with both bona fide and silicone mask based attacks. We further propose \emph{DeCoupPAD-FAS++}, a decoupled representation learning framework that separates spoof related evidence from nuisance sensitive capture variation through complementary attack and style branches, pseudo-domain supervision, adversarial suppression, and prompt guided semantic alignment. Experiments on SmaGla-FPAD show that the proposed method achieves the best balance between attack rejection and bona fide acceptance in both in domain and cross scenario evaluation. Ablation results further confirm the contribution of disentanglement, pseudo-domain supervision, and prompt based regularisation under smart glass capture.

Agentic AI systems are deployed to execute complex tasks across domains such as writing, coding, and data analysis. As execution is offloaded to AI agents, the human role shifts toward oversight. Effective oversight requires recognizing when closer scrutiny is needed, even though flawed outputs may appear convincing. Our formative study (N=12) revealed that users accepted plausible outputs without thorough evaluation, and only recognized these lapses in retrospect. To examine this systematically, we conducted a controlled within-subject study (N=36) in which participants completed programming tasks with an agent that produced correct outputs, silent defects, or visible failures. Combining task outcomes, interaction logs, gaze, physiological signals, and self-reports, we examined perceived, behavioral, and effective oversight. Participants were often highly confident while missing silent defects: they performed generic checks, failed to generate diagnostic evidence, or even misinterpreted defect evidence. We discuss how studying oversight across multiple dimensions can support future work on timely, task-calibrated re-engagement.