Papers

A major challenge in wearable device activity recognition is the heterogeneity across datasets in sensor placement, channel count, sampling rate and subjects. Moreover, the motion patterns that distinguish activities often appear as subtle, scale-specific variations within otherwise similar waveforms. As a result, raw signals are often ambiguous, and models trained on them tend to rely on cues specific to a particular sensor setup or participant, which limits both cross-dataset transfer and generalization to unseen participants. Addressing this requires representations that preserve such activity-relevant, scale-specific variation. We introduce W-JEPA, a self-supervised framework that exploits the multiscale structure of inertial motion and leverages a joint-embedding predictive architecture, by organizing windows into a fixed grid of wavelet tokens indexed by scale, sensor axis, and time. These wavelet tokens help expose activity-specific detail within broadly similar raw signals, and serve as the units of masked latent prediction. W-JEPA predicts masked token representations supplied by an exponential-moving-average (EMA) target encoder. Because recovering a masked token requires relating motion at one scale and axis to motion at other scales and axes, the model learns the cross-scale, cross-axis structure of human movement. Across six wearable activity benchmarks, W-JEPA outperforms state-of-the-art methods under random-window splits and remains the strongest when subjects are held out, improving average macro-F1 by 8.7 percentage points over the strongest published baseline. These results indicate that physically grounded multiscale tokenization, combined together with latent predictive learning, provides a promising foundation for transferable representations of heterogeneous wearable inertial data.

We prove an endpoint Pucci maximum principle for domains in $\mathbb R^2$, formulated using a generalization of elementary symmetric polynomials to noninteger orders. As a result, we sharpen an earlier result of Astala, Iwaniec and Martin, and deduce a new Sobolev inequality for $q$-Hessians where $1 < q < 2$. The fractional elementary symmetric functions we use are derived through fractional differentiation. Their signed singular value extension coincides with the Burkholder functional introduced by Burkholder and gives us access to a local Burkholder inequality due to Astala, Iwaniec, Prause and Saksman, leading to a different proof of the Pucci maximum principle from the one in that used quasiconformal maps. We include an elementary real-variable proof of this local Burkholder inequality, obtained by combining a few known techniques from existing literature.

In this paper, we construct normalized periodic kernels from univariate functions satisfying suitable moment conditions and derive sufficient conditions for high-order convolution approximation. Two computable quasi-interpolation schemes are obtained by discretizing the convolution integral using either a full tensor grid or random samples. In each case, the analysis separates the continuous convolution error from the error introduced by discretization. The contribution of kernel energy to the error exposes the dimensional dependence of tensor-product kernels and motivates our weight-adapted ANOVA constructions. Under suitable summability conditions on the weights, we establish convolution error and kernel energy estimates that yield dimension-uniform error bounds for the resulting quasi-interpolants. Numerical experiments corroborate the theoretical convergence rates and illustrate the effect of coordinate weighting on dimensional dependence.

Diffusion large language models (dLLMs) have advanced rapidly, yet strengthening their reasoning through reinforcement learning (RL) remains an open challenge. Any-order generation can suffer from entropy degradation, which narrows the reasoning paths available for exploration. Recent work mitigates this issue by applying standard GRPO with left-to-right generation. However, independently generating every rollout from the input prompt can leave alternative continuations from intermediate states underexplored, while trajectory-level advantages assign uniform credit to all tokens without distinguishing the contributions of individual reasoning segments. We propose ForkGRPO, an RL framework that couples efficient exploration with fine-grained segment-level credit assignment. It samples multiple continuations from high-entropy intermediate states along a reference trajectory, prioritizing textual boundaries and reusing shared prefixes. The continuations provide complementary learning signals: within-state group-relative advantages guide which continuations to reinforce, while Monte Carlo state-value differences indicate which reasoning segments improve expected outcomes. Among the evaluated architecture-preserving RL methods, ForkGRPO achieves the best performance on three of four benchmarks, while improving over the directly comparable JustGRPO by up to 1.2 percentage points and using up to 27.8% fewer generated tokens. ForkGRPO-Fast further achieves a $2.49\times$ training wall-clock speedup with negligible accuracy loss.

Vision-language models must reconcile conflicting text and images. Prior work has characterized the resulting behavior in terms of modality biases–e.g., a "blind faith in text"–but why preferences vary across models, domains, and input conditions remains unclear. We develop a predictive account from a Bayesian perspective, positing that VLMs combine conditionally independent evidence from each modality with prior beliefs, subject to modality-specific biases. The resulting two-parameter cue combination model accurately predicts VLM preferences under unseen combinations of input degradation and transfers to new forms of unreliability without refitting. Separating evidence strength from integration bias explains how VLMs can follow text even while weighting visual evidence more heavily, reconciling reports of text and vision dominance under a single combination rule.

The existence of multiple superconducting phases is a striking feature of UTe2, arising from the spin and orbital degrees of freedom and supporting a spin-triplet superconducting state. We present field-temperature phase diagrams under pressure for fields applied along the three principal crystallographic axes, as determined by AC calorimetry measurements. The pressure evolution of the multiple superconducting phases is highly anisotropic with respect to magnetic field direction, revealing rich superconducting phenomena arising from complex fluctuations.

Large Language Model (LLM) serving systems must balance latency and throughput while efficiently utilizing distributed GPU resources. This challenge arises because inference consists of two fundamentally different phases: prefill, which determines time-to-first-token (TTFT), and decode, which determines inter-token latency (ITL). The use of tensor parallelism (TP), pipeline parallelism (PP), and hybrid configurations exposes different trade-offs between these latency dimensions, making the optimal parallelism configuration highly dependent on workload characteristics, latency requirements, and context length. Moreover, despite the widespread adoption of distributed inference, there is still limited understanding of how parallelism configurations should be selected under varying workload conditions and service-level objectives (SLO). In this paper, we develop a systematic framework for characterizing LLM inference parallelism via throughput-latency trade-offs, SLO-aware serving, and long-context scalability. We introduce the notion of parallelism operating regions by constructing TTFT-ITL feasible sets and Pareto frontiers across different workload compositions, thereby revealing a systematic transition from TP-dominant to PP-dominant configurations as workloads become increasingly prefill-heavy. We further formulate SLO-aware parallelism selection using goodput, showing that stringent TTFT requirements favor PP-dominated configurations, whereas relaxed latency constraints progressively shift the optimality toward TP-dominated configurations with higher serving capacity. Finally, we demonstrate that increasing context length fundamentally changes the relative advantages of parallelism configurations, with PP-dominant configurations exhibiting more favorable scalability and SLO attainment in the long-context regime.

Reticular chemistry allows metal-organic frameworks (MOFs) to be designed from their building units, yet their synthesis remains constrained by sparse experimental data, noisy crystallization outcomes, and the absence of benchmarks for comparing active-learning strategies. While data-driven approaches such as Bayesian optimization (BO) can improve synthesis success rates, selecting appropriate optimizers beforehand for mapping MOF crystallization landscapes remains largely empirical. Here we present ActMOF, an experiment-grounded benchmark for comparing search strategies in aluminum-carboxylate MOF synthesis. ActMOF is emulated from experimentally measured microwave syntheses with each encoded condition assigned a crystallinity index calculated based on diffraction peak intensity and full width at half-maximum. By benchmarking policies guided by Gaussian processes (GP), tree ensembles and large language models (LLM), ActMOF placed the GP family among the leading policies for active learning in sparse synthesis landscapes. It then demonstrated transferability by guiding the microwave synthesis of six reticularly related aluminum MOFs. All six campaigns gave crystalline products within a fixed budget of 12 experiments each, and established a reusable testbed for AI-driven MOF synthesis.

Discrete multi-group pipe routers place channels on a grid or coarse graph. The resulting centerlines are topologically feasible but typically contain many axis-aligned corners. For permanently occupied fluid tubes—as in additively manufactured hydraulic manifolds—those corners are undesirable: they degrade geometric fairness and, in practice, flow quality. This technical report studies the downstream geometric stage that starts from a fixed routing topology and a feasible coarse layout, and refines tubular centerlines under permanent clearance. Unlike classical robotic path smoothing, which usually fair one temporary open trajectory, our setting is a branched manifold with declared terminal usages, finite tube radii, and static cross-group occupancy. We formalize the refinement problem, highlight the multi-usage junction conflict that a single shared branch point cannot resolve, and outline a hybrid trunk-share / junction-split parameterization with soft geometric costs and hard clearance acceptance. On a multi-group suite ($8$ groups $\times$ $8$ terminals, $r{=}2$) with capsule clearance, the refine reaches full hard-accept success, $\kappa r\lesssim 1$, and about a $95\%$ relative drop in the soft objective. Physics-based pressure-drop optimization is intentionally deferred to a later PDE stage; here the objective remains geometric.

The iodine $4d$ giant resonance provides an unusually strong and predominantly iodine-centered photoionization channel in iodoalkanes. Recent attosecond measurements by Schröder et al. [Phys. Rev. A 112, L051104 (2025)] reported pronounced, nonmonotonic, and isomer-dependent variations of the streaking-derived $4d$ photoemission delay across the molecular series. We calculate field-free molecular Eisenbud–Wigner–Smith (EWS) delays from photoionization amplitudes represented in a multicenter B-spline basis of the time-dependent density-functional theory (TDDFT). A tensor formulation is developed in which the delay for randomly oriented molecules is evaluated from the same rank-0 and rank-2 contractions that determine the cross section and anisotropy parameter. The resulting TDDFT delays of CH$_3$I, C$_2$H$_5$I, 2-iodopropane, and both calculated conformers of 1-iodopropane form a narrow bundle throughout the 90–118~eV region and closely follow atomic-I and I$^-$ reference calculations. This robustness is consistent with the similarly weak molecular dependence of the giant resonance cross section and angular distribution, but it is in marked contrast with the much larger molecular variation found by streaking experiments. The asymptotic Coulomb–laser-coupling/continuum–continuum correction is controlled mainly by the photoelectron energy and residual charge and is therefore expected to be nearly common to the iodoalkane series. This does not exclude target-specific molecular IR effects beyond the universal Coulomb term.

As large, powerful pre-trained models become more prevalent, there is growing interest in developing methods that allow different neural networks to be seamlessly combined with little or no direct communication between them. Take autoencoder-based data compression as an example: the ability to independently choose architectures and train encoder and decoder models, while still ensuring they work together, could fundamentally transform how such models are built, deployed, and updated. In this work, we introduce a novel approach that leverages topological regularization to align the latent spaces of two independently trained autoencoder models, entirely eliminating the need for coordination during training. We provide a conditional theoretical analysis showing how the structural agreement encouraged by our regularizer can translate into approximate geometric compatibility between independently learned representations. Reconstruction experiments validate this approach empirically, substantially narrowing the interoperability gap across multiple datasets.

We study symmetric 2-(v, k, {\lambda}) designs admitting a flag-transitive point-primitive subgroup G of automorphisms, where gcd(k, {\lambda}) = p and p is a prime. In this paper, we show that G is of affine or almost simple type.

Korean neologisms arise through diverse mechanisms, a demanding test of language models beyond lexical memorization. To enable systematic evaluation, we introduce NeoTrace, an automated, rerunnable framework that collects and dates neologisms across genres. Its 4,271 entries support KoAGE, a cutoff-aware benchmark with three acquisition and two generation tasks—the latter, to our knowledge, the first to compare model coinages with attested neologisms under their original materials and community context. Across 21 models, lexical recall diverges from contextual inference: in a five-model temporal analysis, models define only 13-30% of cutoff-year words, yet can infer meaning from one usage. However, structural understanding remains limited: even correct definitions barely improve formation analysis of pre-cutoff neologisms. Given word-building materials, single models often follow human formation mechanisms, whereas multi-agent coinage favors abbreviation and depends on interaction framing. KoAGE thus separates what models know, learn from use, and coin.

Iterative retrieval-augmented generation (RAG) accumulates knowledge from retrieved evidence to answer complex questions, but existing methods struggle to reconstruct entity correspondences and inferential dependencies across retrieved passages as new information emerges. We propose RCER, an iterative RAG method that reconstructs these cross-evidence relations directly from accumulated source passages. Inter-round relational evidence acquisition identifies entity correspondences and relational dependencies to drive dynamic problem decomposition and iterative retrieval. Reasoning context reconstruction independently constructs a coherent reasoning context through joint selection of complementary passages and derivation of cross-evidence inferences. Cross-evidence compositional reasoning leverages these inferences to establish entity correspondences and compose relations across selected passages for answer derivation. Experiments show that RCER outperforms the best baselines by 2.83 EM and 3.06 F1 points on average across three multi-hop QA benchmarks and achieves superior performance on narrative and implicit reasoning tasks.

Long-horizon agentic RL is constrained by the futures visited by the current policy. When rollouts fail to reach useful behaviors, improved credit assignment cannot provide supervision for states and actions that were never encountered. On-policy distillation (OPD) introduces teacher guidance on policy-induced states, and recent multi-turn variants improve how this guidance is scheduled, replayed, refined, or validated. However, they do not directly use the oracle-controlled future itself as the training trajectory. We introduce State-Rooted Executable Rollout Optimization (SERO), which augments on-policy agentic RL with executable oracle continuations. SERO restores a state actually reached by the current policy and lets a training-only oracle continue interacting with the real environment, making the downstream observations and oracle actions themselves available for learning. It then applies policy-reachability weighting, which weights each oracle step by the frozen behavior policy's compatibility with the preceding oracle actions. This allows deeper supervision to become effective as the policy becomes more compatible with the oracle branch. The oracle loss is optimized separately from the base RL estimator and no oracle is used at inference time. We evaluate SERO on two challenging interactive-agent benchmarks, ALFWorld and WebShop, using Qwen2.5-1.5B-Instruct and Qwen2.5-7B-Instruct. On ALFWorld, learning the full executable continuation is substantially more effective than supervising only the root oracle action, while policy-reachability weighting further improves out-of-distribution performance.

Infectious disease outbreaks often spread rapidly before signs of widespread transmission become apparent, leaving a big challenge for rapid response and effective intervention. However, current infectious-disease early-warning systems face three major challenges: fragmented epidemiological evidence across diverse surveillance sources, highly uneven data availability across diseases and geographic regions, and delayed warning procedure caused by manual data collection, analysis, and reporting. To address these challenges, we present EpiGuard, an autonomous, continually self-improving agentic platform for timely and reliable early warning of more than 50 infectious diseases across different regions. EpiGuard proactively monitors and captures newly released official surveillance records, using the latest available evidence for risk assessments. It dynamically adjusts its analytical strategy across three evidence availability regimes, enabling reliable risk assessment under diverse data conditions, ranging from the absence of historical observations to sparse and abundant evidence. Moreover, EpiGuard learns from the successful execution process by preserving effective data-acquisition and analytical procedures as reusable standard operating procedures, enabling it to progressively generalize to previously unseen diseases and regions. We use officially released infectious-disease surveillance data from the China CDC and the United States CDC to evaluate EpiGuard's effectiveness. Experimental results demonstrate that EpiGuard outperforms state-of-the-art forecasting systems, improving forecasting performance by 9.68% and 14.04% on the China CDC and U.S. CDC datasets, respectively.

Robotic vascular intervention can reduce radiation exposure to physicians and improve manipulation stability, but effective guidance remains challenging because conventional fluoroscopic images provide limited spatial information. This paper developed a vascular intervention interactive system that connects a real robotic master controller with the SOFA(Simulation Open Framework Architecture) physical simulation platform and reused successful operation trajectories as visual guidance in physical phantom experiments. Experiments were conducted to validate the master-to-SOFA control loop and to compare three guidance conditions: no guidance, centerline guidance, and successful-trajectory guidance. The control validation results showed reliable command delivery under velocity-mode operation. In the physical phantom experiment, centerline guidance showed no obvious improvement over no guidance. In contrast, successful trajectory guidance provided a clear performance benefit. Compared with the no-guidance condition, it reduced task completion time, collision number, rollback number, and rotation number by approximately 66.8%, 34.3%, 55.6%, and 66.7%, respectively. These results indicate that successful trajectories obtained under instrument-vessel interaction can provide practical guidance cues, offering a simple and effective visual assistance strategy for robotic vascular intervention experiments.

Next-location prediction can support proactive mobile-network decisions, but acting on unreliable predictions can waste resources. We apply MetaModelSelect, a lightweight post-hoc confidence metamodel, to selective cellular next-location prediction. The mobility-specific model combines a learned embedding of the predicted location with the four largest softmax probabilities and Softmax Concentration. It learns whether a frozen recurrent neural network's top-1 prediction is correct, using only its outputs and without retraining the backbone. On path-balanced Changchun cellular mobility trajectories, we evaluate backbones trained with cross-entropy (CE) and self-adaptive training plus entropy minimization (SAT+EM), comparing against eight scalar confidence scores, two location-aware lookup methods, and a raw-logit multilayer perceptron. Relative to maximum softmax probability, MetaModelSelect achieves mean per-run reductions in area under the risk–coverage curve (AURC), excess AURC (E-AURC), and area under the generalized risk–coverage curve (AUGRC) of 2.6%, 4.2%, and 3.0% for CE, and 2.0%, 3.3%, and 1.9% for SAT+EM. It also achieves the lowest reported mean risk at every evaluated partial coverage. The results show that MetaModelSelect improves confidence ranking in this mobility setting while leaving the underlying location predictor unchanged.

Whole-body control plays a crucial role in enabling humanoid robots to execute expressive motions. Faithful execution requires preserving both the intended action and its expressive characteristics. Existing whole-body trackers typically balance joint and keypoint tracking with physical stability across diverse motions, but these objectives do not necessarily prioritize the expressive characteristics of each action. Our analysis shows that lower average positional tracking error does not guarantee better preservation of dynamic and postural motion characteristics. To address this gap, we propose a residual learning framework with compensation objectives targeting amplitude, energy, body expansion, body lean, and pelvis height. To concentrate compensation while limiting interference with the underlying tracking behavior, motion-aware gating allocates correction capacity according to body-part activity, preserving larger correction ranges in more active parts and restricting them elsewhere. For each frozen tracker, we train motion-specific residual policies and distill them into a unified policy shared across motion categories. Simulation and physical experiments on the Unitree G1 across eight motion categories demonstrate superior preservation of dynamic and postural motion characteristics compared with the corresponding baselines.

We identify a strange-quark-number-dependent hierarchy in the response to explicit partonic dynamics by comparing PHSD5.2 with its hadron–string limit in baryon-rich Au+Au collisions at $\snn=3.0$–$7.2$~GeV. Crossing the local conversion threshold alone does not ensure a visible final-state signal: abundant-hadron flow changes little at 3–4.5~GeV, whereas the fractional yield response becomes ordered at 6.2–7.2~GeV, reaching $S_{|s|=1}\simeq0.2$–$0.3$, $S_{|s|=2}\simeq0.65$, and $S_{|s|=3}\simeq0.93$ at 7.2~GeV. The hierarchy is consistent with an enlarged recombination phase space for strange constituents released from different pre-hadronic sources, but does not by itself establish an increase in the total $s+\bar s$ abundance or stronger strange-quark clustering. Since $S_X$ is a model-defined response rather than a direct observable, matched excitation functions of $\Xi^-/\Lambda$ and $\Omega^-/\Lambda$, together with high-$m_T$ proton elliptic flow, provide experimentally accessible discriminants between the predicted PHSD and HSD dynamics.

We study continuous global optimization problems in which the objective function cannot be directly evaluated; rather, we observe only which of two selected design points appears to perform better, subject to noise. Such comparisons complicate the use of Gaussian process regression because they yield non-Gaussian posteriors. We consider a fully sequential approximation, based on moment-matching, that yields a fast closed-form update of the metamodel's mean and covariance. We then propose DOVE, a fully sequential sampling policy that combines an information gain criterion with randomized sampling designed to control error accumulation. Under fairly standard regularity conditions, we prove convergence rates for both statistical error and expected simple regret that depend on kernel smoothness. To our knowledge, these are the first such rates for fully sequential approximate Bayesian inference in continuous global optimization.

In this paper, we present a novel multiscale formulation for the Optimal Transport (OT) of large swarms. The scheme is divided in two levels, a top-level OT formulation with a terminal penalty cost to handle time-varying densities, and a low-level, Model Predictive Controller (MPC) that allows agents to track the nominal velocity under control affine dynamics and rejects external disturbances. We show that the terminal-penalty dynamic formulation reduces to a regularized static dual, which agents solve in a distributed way to compute the nominal velocity.We also prove distributional input-to-state stability in the $1$-Wasserstein metric of the density flow with respect to the mismatch that the low-level and the target motion introduce, so that the tracking error only perturbs the macroscopic flow by a bounded, quantifiable amount. Finally, we demonstrate via simulations the effectiveness of the proposed method handling time-varying targets and state perturbations.

Under anthropogenic greenhouse forcing, climate models project a decrease in year-to-year temperature variability in the mid-to-high latitudes and an increase in the tropics and subtropics. Here, we examine year-to-year temperature variability in the CESM Large Ensemble in concert with an idealized stochastic moist energy balance model (SMEBM) with effective surface heat capacity and local feedback parameter specified based on CESM. We find that the zonal-mean CESM variability change pattern largely persists after filtering out the impact of El Niño. The SMEBM largely captures this change pattern. In the SMEBM, changes in the mid-to-high latitudes are driven by changes in the effective heat capacity due to sea ice retreat, combined with changes in anomalous moist static energy convergence modulated by the polar-amplified time-mean temperature change. The CESM local feedback parameter evolves to be less stabilizing in the low latitudes, which causes tropical and subtropical variability to increase in the SMEBM.

Massively multilingual speech modeling has made remarkable progress in recent years, yet existing training resources and benchmarks still underrepresent much of the linguistic long tail. To address this gap, we introduce ELARSpeech, a large-scale multilingual speech resource comprising 30,202 hours of speech across 675 underrepresented languages, curated from the Endangered Languages Archive. Combining human-annotated and unlabeled recordings, the corpus spans diverse recording conditions, domains, and language families. Beyond corpus construction, our experiments show how ELARSpeech supports extending ASR and TTS capabilities to diverse underrepresented languages, with further benefits from pseudo-labeling unlabeled data. At the same time, our speech representation analysis reveals remaining limitations in transferring existing multilingual models to the linguistic long tail. Together, ELARSpeech provides both a resource and a testbed for extending multilingual speech technology further into the linguistic long tail. We publicly release the dataset, all trained ASR and TTS models, and the complete data construction pipeline to facilitate research on long-tail multilingual speech technologies.

Identification and ranking of salient propositions in documents is indirectly involved in many AI tasks, but few papers define proposition salience scores clearly, make diverse data available, or evaluate models directly on the task. This paper proposes PROPALIGN, a multiple-summary-based approach for propositional salience score estimation, and releases PROPALIGNBANK, an accompanying dataset for the method with 138K alignments of over 35K propositions. We compare several approaches to salient proposition extraction in English across a broad range of spoken and written genres, and demonstrate that naively asking models for judgments leads to inconsistencies. By contrast, graded, summary-based metrics are more stable. We further observe that while single models perform worse than humans, an ensemble of models incorporating linguistic features approaches manual annotation quality ceilings. In addition, analyses of the model outputs suggest insights into how models can extract salient information with and without the use of summaries.

Large language model (LLM) agents can self-evolve at test time by accumulating experience: retrieving relevant memories before acting and writing new lessons back afterwards. Yet most are single-level: trajectories, distilled workflows and reasoning items all land in one undifferentiated pool indexed by embedding similarity, with no typed abstraction hierarchy. They never forget or deduplicate, and are prone to negative transfer, where a superficially similar but directionally opposite experience is recalled and its concrete steps are copied verbatim, actively harming the agent. We introduce STRATA, a structured, outcome-grounded, self-maintaining memory that organizes experience into a three-layer graph of atomic skills and code snippets, task workflows and generalized meta-frameworks, connected by typed edges. STRATA maintains itself through five mechanisms sharing one similarity geometry: a multi-dimensional similarity that scores retrieval by operation and target; a typed reflector crystallizing procedures from successes and pitfalls from failures; community detection that induces meta-frameworks; pressure-triggered garbage collection with articulation-point protection; and contrast edges that flag confusable, direction-opposite experiences. Across four interactive benchmarks (AppWorld, WebArena, ALFWorld, ScienceWorld), two frozen backbones and seven baselines, STRATA attains the best average score while simultaneously using the lowest latency and fewest environment steps of any memory-augmented method, and 44% fewer lines of injected memory than the most context-hungry baseline. We isolate two failure modes of prior memories: single-level memory degrades sharply as the task set grows, and single-level concrete memories suffer catastrophic negative transfer, which contrast edges repair, lifting ScienceWorld from 78.7 to 96.7 within the same system.

Large Language Model (LLM) unlearning is often evaluated primarily by target suppression and clean retain utility. These metrics can miss a local failure mode: a model may perform well on standard retain prompts, yet degrade on benign reasoning when prompts lie near the forgotten domain. This phenomenon is termed forget-adjacent utility loss. A fundamental question is: where should forget representations go after they leave the removed behavior? Successful unlearning depends not only on whether forget-conditioned representations depart from the harmful computation, but also on the direction in which they are redirected afterward. We propose RA$^2$ (Retain-Anchored Attraction), a latent-space unlearning method that redirects forget-token states toward nearby retain-supported representations while preserving higher-layer behavior on retain data. On WMDP and MUSE, RA$^2$ yields higher forget-adjacent utility than strong unlearning baselines at similar broad-utility levels. These results show that the migration direction of forget representations materially affects benign behavior near the forget boundary. Our code is available at https://github.com/ZacLing/retain_anchored_attraction.

Multimodal agents increasingly rely on visual interfaces to perceive state and select actions, creating an attack surface not captured by text-only safety evaluations. We study Visual Semantic Hijacking, in which deceptive but semantically meaningful interface elements, such as forged system alerts, cause an agent to depart from its authorized task. We propose a trajectory-guided procedure that identifies candidate injection points and inserts reproducible visual payloads into the observation stream. Using this procedure, we construct MMA-SafetyBench, which contains 1,083 adversarial trajectories across web automation, desktop GUI control, mobile navigation, document understanding, and video processing. Across nine evaluated models, Attack Success Rate (ASR) reaches 87.96%, while Semantic Compromise Rate (SCR) reaches 98.09%. A 1–3–5 scoring scheme distinguishes target-matching outputs, semantic disruption without a target match, and successful preservation of the benign task. Controls covering all nine models and five domains show that attacks persist under neutral prompts. A safety-priority prompt lowers ASR without eliminating the failure mode.

In this work, we investigate the phase stability, electronic structure, and superconducting properties of the two-dimensional ternary borocarbides MB$_4$C$_2$ (M = Ca, Mg) using first-principles calculations and fully anisotropic Migdal-Eliashberg theory. Ab initio molecular dynamics, phonon dispersion, and elastic constant calculations confirm that both monolayers are dynamically, thermally, and mechanically stable. Electronic structure analysis reveals an intrinsic metallic nature with a multiband Fermi surface, where states near the $\Gamma$ point are dominated by in-plane B-$p_x$, $p_y$ orbitals, while zone-boundary states are primarily derived from out-of-plane B-$p_z$ orbitals. The electron-phonon coupling is determined to be $\lambda = 0.64$ for both materials, driven by the rigid covalent B-B and B-C honeycomb networks. By solving the anisotropic Eliashberg equations, we predict that CaB$_4$C$_2$ and MgB$_4$C$_2$ are nodeless two-gap superconductors with critical temperatures ($T_c$) of 33.4 K and 33.8 K, respectively, at a Coulomb pseudopotential of $\mu = 0.10$. Furthermore, the superconducting state remains robust under increased Coulomb repulsion ($\mu = 0.13$), retaining elevated $T_c$ values of 31.2 K and 30.9 K, respectively. These findings establish MB$_4$C$_2$ monolayers as promising candidates for nanoscale superconducting applications and enrich the family of two-dimensional multigap superconductors.

Running an AI call agent on-device makes it private, network-independent, and free of per-query cloud cost, but shipping one as a real-world service imposes constraints a cloud deployment never faces: ~1GiB memory budget, in-call latency and thermal limits, static-integer NPU execution, Korean STT input far from sLM pretraining, and cloud-level quality expectations. We bring ixi-sLM, a 1.2B Korean sLM, to mobile NPUs through ON-DEVICE MODELING: domain-adaptive modeling trims the vocabulary by call frequency to shrink the embedding and LM head that take nearly half the model under INT4 transformer weights and refits the tokenizer to call speech (18% fewer tokens on call traffic), while task-aware quantization calibrates on real calls and assigns per-module mixed precision by activation sensitivity. A single domain-adapted model runs ixi-O's two production services, AI Call Summary and AI Call Answering, at 94% of a state-of-the-art cloud LLM's task-weighted quality; our on-device modeling carries that quality onto the NPU with only a marginal drop in overall score, in an 803MiB model (28% of FP16) at 23% of the CPU's per-call energy. The same recipe deploys across Qualcomm and Apple NPUs and serves several million ixi-O subscribers in production.