Papers

Magic state distillation is a widely considered primitive in fault-tolerant quantum computation for the preparation of high-fidelity non-Clifford resources. The most commonly considered class of such protocols are generalised triorthogonal codes; these distil $n$ noisy input $\mathrm{T}$ states into purified third-level diagonal magic states. Extensive prior work has searched this space of protocols, often using heuristic methods that do not guarantee optimality. Existing classification work is limited to protocols distilling $k$ output $\mathrm{T}$ states with $n+k\leq 38$ (Nezami and Haah, Phys. Rev. A 106, 012437 (2022)). In this work, we significantly expand this classification to all protocols with lengths $n\leq 54$. We restrict to distance $d\geq 3$ to keep the classification to a sensible size, and because efficient searches are well understood at distance $2$ (Singh et al., arXiv:2606.28518). Moreover, our results are complementary to synthillation (Campbell and Howard, Phys. Rev. A 95, 022316 (2017)), which can distil third-level states from $\mathrm{T}$ states, but only at distance $2$. Under optimality in terms of input count, space footprint, and distance for a given output, we find $74$ optimal generalised triorthogonal protocols in our range, $65$ of which are new to the literature. To achieve our classification, we extend the classification of unital triorthogonal spaces of Nezami and Haah from length $38$ to $54$ using a directional derivative method. Using these as the stabiliser spaces, we add logical rows that satisfy the triorthogonality constraints to create full protocols.

Contingency analysis using the AC power flow (AC-PF) model is a critical tool for accurate grid security assessment, but its computational burden increases with the number of operating scenarios and outage configurations to evaluate. Recent ML-based approaches typically require outage-specific training data, leading to offline training costs that scale with the number of contingencies. This work proposes a framework that reuses a single ML model trained solely on basecase AC-PF data to estimate post-contingency operating states under arbitrary single-line outages. The proposed approach formulates post-contingency state prediction as a fixed-point iteration. If the ML model is a deep neural network (DNN), we derive sufficient conditions that guarantee convergence and develop semidefinite programming (SDP) formulations to certify these conditions for a given DNN. Numerical tests on the IEEE 118-bus system demonstrate that the proposed SDP formulations are tight, that the certified conditions hold for all tested contingencies, and that the resulting method produces accurate post-contingency state estimates within only a few iterations.

We prove the soliton resolution conjecture for the Tzitzéica equation, which is an important integrable model arising in affine differential geometry and characterizing proper affine spheres. Recently, Huang, Wang and Zhu (Math. Ann. 395, 18 (2026)) reported the long-time asymptotic result for the Tzitzéica equation with purely continuous spectrum. Inspired by their work, we carry out a full asymptotic analysis of the Tzitzéica equation. Compared with the previous work, by developing the $\bar{\partial}$-nonlinear steepest descent method, we reveal that the solutions of the Tzitzéica equation exhibits three different asymptotic regions depending on $\xi:=x/t$. For the region $\xi\in(-\infty,-1]\cup[1,\infty)$, where there are no stationary phase points, we rigorously prove that the solution satisfies $u(x,t)=O_\xi(t^{-1})$ for $\xi\ge1$ and $u(x,t)=2\pi i+O_\xi(t^{-1})$ for $\xi\le-1$. In the region $\xi\in(-1,1)$, the phase function $\theta(z)$ has two stationary phase points. The corresponding asymptotic approximations can be characterized with an $N$-soliton solution as well as an interaction term between soliton solutions and the dispersion term with a residual error of order $O(t^{-3/4+b_K})$ with $0\le b_K<1/4$. In the transition regions $\{(x,t):1-\varepsilon<|\xi|<1\}$, the corresponding asymptotic approximation can be characterized by the dispersion term with a residual error of order $O(t^{-3/4+b_\varepsilon})$ where $0\le b_\varepsilon<1/4$. Our results provide a rigorous analysis of the long-time asymptotic behavior of solutions of the Tzitzéica equation in different regions and confirm the soliton resolution conjecture for the Tzitzéica equation.

Rare-earth bilayer triangular lattice (TL) antiferromagnets provide a versatile platform for realizing diverse magnetic states by combining geometric frustration, interlayer coupling, and strong single-ion anisotropy. Here, we report a family of R2O2Se (R = Sm, Eu, Tb-Lu; ROSe) single crystals featuring bilayer equilateral TLs. Magnetic susceptibility and specific-heat measurements reveal predominantly antiferromagnetic (AFM) interactions and diverse magnetic ground states across the series, including successive AFM transitions in TbOSe and single AFM transitions in SmOSe, DyOSe, HoOSe, and YbOSe. Notably, DyOSe and HoOSe exhibit pronounced 1/2 magnetization plateau-like features for fields along the c-axis, revealing field-induced magnetic states. In HoOSe, complementary thermodynamic and magnetization measurements resolve three critical fields and a rich field-temperature phase diagram containing five distinct magnetic phases. These results establish rare-earth bilayer TLs as a chemically tunable materials platform for accessing novel quantum spin states through the interplay of lattice geometry, single-ion anisotropy, and competing magnetic interactions.

Safe reinforcement learning (RL) commonly enforces expected-cost constraints, but such expectation safety may fail to control the probability of rare high-cost trajectories. Chance-constrained MDPs (CCMDPs) impose a stronger probability-level requirement, but are widely viewed as harder because the chance constraint is nonconvex and depends on the full trajectory rather than a Bellman-linear expectation. In this paper, we reveal that this computational difficulty does not necessarily imply a higher statistical price. For tabular discounted CCMDPs with fixed bounded successor support and access to a certified planning oracle, we establish a model-based upper bound, with a matching lower bound up to logarithmic terms. Technically, our key idea is the Bellman distributional certificate, which constructs a Bellman recursion for constraint violation probabilities before policy selection. The certificate can be reused across candidate policies; combined with shared row-wise reverse-KL confidence sets, it gives a policy-uniform trajectory-KL transfer without a union bound over policies or time–budget Bellman tables. For stochastic policies, we give a model-free variance-reduced policy-gradient algorithm with a finite-sample expected KKT-residual guarantee and independent validation of every accepted policy. Numerical experiments on synthetic CCMDPs and an IEEE 14-bus energy storage control benchmark illustrate the safety and mechanism behavior of the proposed algorithms.

Reliable skin lesion segmentation and classification are central to dermoscopic computer-aided diagnosis. Existing multi-task frameworks couple the two tasks architecturally without clinical knowledge, while knowledge-injecting approaches rely on the macroscopic ABCD rule, which was not designed for dermoscopy. Dermoscopic diagnosis is grounded in Pattern Analysis, a microscopic framework structured around dermoscopic features. We propose MDSkin-Net, which incorporates cue-level Pattern Analysis priors into a hybrid CNN-Transformer architecture. At its core is a Pattern Analysis-Guided Attention Module (PAGAM) comprising three priors motivated by distinct dermoscopic cues: an improved Efficient Channel Attention (iECA), a Multi-Scale Spatial Attention (MSSA), and a Biased Asymmetry Attention (BAA). We further introduce a multi-scale spatial alignment regularization (MSAR) that uses the segmentation ground-truth mask as hierarchical soft supervision, confining the classification head to lesion-localized evidence and coupling both task pathways through a shared spatial prior. Trained exclusively on the ISIC 2017 training split without external dermoscopy data, the MDSkin-Net ensemble transfers robustly under zero-shot evaluation, reaching a Dice Similarity Coefficient (DSC) of 92.38% and a melanoma AUC of 97.84%on PH2, and a DSC of 88.92% on the ISIC 2018 Task 1 test set. On the in-domain ISIC 2017 benchmark, the ensemble attains a mean Area Under the Curve (AUC) of 91.60% across the two classification tasks (melanoma and seborrheic keratosis vs. rest), and a DSC of 84.72% for segmentation. Classification remains competitive with baselines; in-domain segmentation trails single-task specialists, yet the proposed priors and alignment regularization yield representations that generalize consistently across cohorts of different scales.

Autoregressive LLM inference at long context is bounded by memory bandwidth, not arithmetic throughput, and the binding resource shifts from model weights to the Key-Value (KV) cache as sequence length grows. For Llama-3-70B in BF16, the 140 GB weight footprint exceeds the 80 GB HBM of a single accelerator, and one 128k-token sequence adds 42 GB of KV cache. Techniques that compress, evict, page, share, or offload KV state have proliferated, but reported gains use inconsistent workloads, hardware, and quality metrics, preventing cross-paper comparison. This SoK paper unifies the field analytically, with a protocol that strictly separates derived and reported claims. We derive closed-form arithmetic intensity as a decaying function of context length, parameterized by hardware topology for NVIDIA H100, NVIDIA B200, and AMD MI300X, including per-die bandwidth partitioning and the crossover lengths where KV traffic overtakes weight traffic. We classify the literature into five domains, quantization, token eviction, KV paging, prefix caching, and heterogeneous tiering, evaluating one method per domain at 128k context under a single protocol. The central finding is a three-regime structure: below a hardware-specific crossover, weight traffic dominates and KV compression yields negligible speedup; beyond it, KV traffic dominates and each domain trades quality for bandwidth savings approaching the roofline bound. Paging and prefix sharing are lossless but address capacity, not bandwidth. Quantization and eviction cut bandwidth directly, with degradation that accelerates below 4-bit precision and turns discontinuous for eviction on position-sensitive tasks. Tiering converts the bandwidth wall into an interconnect problem bounded by PCIe or NVLink rather than HBM. We close with design rules for selecting a compression domain given hardware, context length, and quality budget.

Lee and Sidford [LS19] showed that the linear program $\min\{c^\top x\ :\ A^\top x=b,\ l\leq x\leq u\}$ over $x\in\mathbb{R}^m$, where $A\in\mathbb{R}^{m\times n}$, can be solved to accuracy $\epsilon$ using $O(\sqrt n\log^{13}m\cdot\log(mU/\epsilon))$ solves of linear systems in $A^\top\mathbf DA$ for positive diagonal matrices $\mathbf D$, where $U$ bounds the magnitudes of the input and of the initial point. Theirs is the first such bound governed by $\mathrm{rank}(A)=n$ rather than by the number of constraints $m$. Song [Son19] mentioned that reducing the $\log^{13}m$ factor is an interesting future direction. Given an interior point and a finite certified magnitude bound $U$, we give an algorithm that uses $O(\sqrt n\log^{4}m\cdot\log(mU/\epsilon))$ solves of such linear systems, improving the Lee–Sidford bound by a factor of $\log^{9}m$. We conjecture that $O(\sqrt n\log(mU/\epsilon))$ such linear-system solves suffice.

The emergence of autonomous laboratories motivates rapid conversion of experimental data into reliable atomistic models on time-scales compatible with closed-loop optimization. Here we develop an energetically driven structure matching framework for analysis during ongoing total X-ray scattering experiments. Using data from gold nanoparticles, we match against idealized spherical, octahedral, decahedral, and icosahedral geometries, their machine-learned interatomic potential (MLIP)-relaxed structures, and molecular dynamics (MD) ensembles. Idealized models are fast to generate but can misassign morphology and systematically underestimate size by neglecting surface relaxation, strain, and thermal disorder. MLIP relaxation markedly improves both, while MD ensemble averaging agrees best with experiment. We therefore introduce a hierarchical workflow combining rapid idealized screening with targeted MLIP and MD refinement of top candidates, delivering improved structural feedback without interrupting autonomous operation. This framework provides a route towards autonomous campaigns in which the target structure itself can be updated in response to the evolving energy landscape of structures compatible with the experimental data.

Let $(A,I)$ be a henselian pair such that $A$ is a domain of characteristic zero and $I\ne0$. We prove that any flat affine henselian scheme of finite presentation over $(A,I)$ has nonzero first structure-sheaf cohomology whenever one of its fibers over a point in $V(I)$ has positive dimension. This in particular gives a negative answer to the question by Devadas in the mixed-characteristic case. As an application, we also prove that the first structure-sheaf cohomology does not vanish for henselian affinoids of positive dimensions.

The GlueX experiment at Jefferson Lab is well suited for exclusive studies of strangeness production, combining a linearly polarized photon beam at $8.0 < E_\gamma < 8.8$ GeV with a large-acceptance detector that reconstructs weakly decaying hyperons. We outline a research program built on three aspects: (i) spectroscopy of strange mesons, where many excited $K^*$ states remain unestablished and crypto-exotic strange hybrids must be identified from combined spectrum and decay information, anchored by the $K^*(892)$ as a standard candle; (ii) spin dynamics, where the first precise measurements of the complete spin-density matrix elements in the photoproduced $K^*(892)\Lambda$ system beyond the Schilling formalism constrain the $t$-channel strange production mechanism; and (iii) hadronization, probed through the newly published $p\bar{p}$, $\Lambda\bar{\Lambda}$, and $p\bar{\Lambda}$ photoproduction cross sections and emerging $\Lambda\bar{\Lambda}$ spin-correlation measurements. We place these efforts in the context of the broader GlueX strangeness program and give an outlook toward the high-statistics GlueX-II era.

Prior work has examined the self-consistency of LLM-generated explanations using surface-level perturbation methods. However, the strength of these perturbations is not explicitly measured and controlled. In this work, we propose an LLM-as-a-judge approach to measure perturbation strength in a unified manner across input and CoT perturbations. We then evaluate the self-consistency in explanations generated from various LLMs under controlled strength conditions, ensuring a fair comparison across perturbation types. Experiments show that our proposed LLM-based perturbation strength measure outperforms other embedding- and probability-based approaches and that input perturbations generally affect LLMs more strongly than CoT perturbations. Our work suggests that judgments about a model's self-consistency is fair only within the same perturbation type.

Molecular dynamics is widely used to predict ionic conductivity in solid electrolytes, but the reliability of these predictions is often difficult to assess. Our analysis identifies finite trajectory length, limited cell size, and insufficient sampling as intrinsic limitations of finite atomistic ion-transport calculations. Cubic Li7La3Zr2O12 is used as a representative solid electrolyte to quantify their consequences. These limitations can remain hidden behind apparently linear mean-squared displacements and well-behaved Arrhenius relations, leading to inaccurate diffusivities, activation energies, and extrapolated ionic conductivities. Such inaccuracies can misrank candidate solid electrolytes and consequently misdirect computational screening and experimental validation. Here, we establish the local mean-squared-displacement exponent, {\alpha}(t), as a quantitative reliability criterion that links dynamical convergence to errors in diffusivity and Nernst-Einstein ionic conductivity. The criterion further determines the minimum trajectory length required to achieve a prescribed accuracy as a function of temperature. Independent replicas reduce statistical uncertainty, while selective single-axis expansion mitigates finite-size effects. By establishing when simulated ionic conductivity is quantitatively trustworthy, this approach enables more reliable materials ranking and more efficient use of computational and experimental resources, thereby accelerating the development of high-performance solid electrolytes for solid-state batteries.

High photovoltaic (PV) penetration planning requires repeated nonlinear PV and power-flow evaluations under uncertain irradiance and demand, particularly when two competing objectives, penetration and total electrical loss, are optimized simultaneously subject to reliability constraints. This paper develops a decision-focused, surrogate-assisted stochastic multiobjective framework in which a calibrated PV model coupled with a distribution-network simulator provides the high-fidelity reference, while matched data-only and physics-informed surrogates accelerate optimization and are evaluated empirically for decision reliability across the loss-penetration tradeoff. Although the two surrogates exhibit nearly identical held-out errors, optimizer-realistic screening yields markedly different operational validity rates of 44.4% and 93.4%, respectively, motivating selection of the physics-informed model on decision performance rather than test accuracy. Surrogate inference cuts the cost of a 60-scenario design evaluation from 286.0 to 0.46 ms, an approximately 620 time speedup. A decision-gated enrichment strategy further targets Pareto-critical, constraint boundary, and surrogate-disagreement regions, reducing prediction error by 39.1%. Independent 500 scenario certification verifies all 80 screened candidates and identifies 46 non-dominated designs; the selected compromise attains 70.74% expected penetration with 297.9 kW expected total electrical loss. Cross-feeder transfer eliminates baseline voltage violations on the 85- and 69-bus systems, but produces overvoltage and a 62.7% increase in feeder-network loss on the 33-bus system, demonstrating that certified penetration is topology-specific and demands decision-level auditing, independent certification, and feeder-specific validation.

In this paper, we propose I-Parakeet, an integer-only implementation of NVIDIA's Parakeet-CTC (0.6B parameters) that runs on a smartphone NPU without any floating-point operator or CPU fallback. Modern Conformer ASR models are hard to deploy on edge devices because of their size, and quantized models still fall back to floating point for numerically sensitive operations. This prevents them from fully exploiting integer accelerators such as mobile NPUs. To achieve this, our contributions are threefold. First, we derive an integer formulation of the relative-positional self-attention at the core of the Conformer. We fuse its two score branches with different quantization scales and the relative shift into integer-only operations. Second, we introduce a minimax-optimized Swish approximation that minimizes the maximum error of the Swish output. Third, a layer-wise range analysis of activations yields two targeted remedies: an INT16 grid for the BatchNorm output and percentile calibration for the heavy-tailed pre-encoder activations. I-Parakeet achieves 4.97% WER on LibriSpeech test-other, running on a Qualcomm NPU at a real-time factor of 0.048, 7.5x faster than a CPU baseline.

Representation learning, or featurization, of mean-field Bloch states provides an important route for incorporating electronic information into machine-learning models of first-principles condensed-phase systems. A challenge in representing Bloch states is the gauge redundancy of each state. In this work, we propose to address this issue by explicitly incorporating gauge equivariance in the machine-learning framework. We develop a $U(1)$-equivariant autoencoder that compresses the plane-wave-basis Bloch states obtained from density functional theory into a low-dimensional latent representation. We achieve an average overlap of 0.970 between original wavefunctions and the reconstructions from a 64-dimensional latent space using a dataset of two-dimensional insulators. We further show that $U(1)$ equivariance endows the latent space with physically meaningful structures, including smoothness and topological information. Finally, we present a proof-of-principle application in which the latent representations are used to predict the off-diagonal elements of the $GW$ self energy matrix, and demonstrate the importance of enforcing gauge equivariance in the prediction.

We consider some new velocity estimates for general steady Navier-Stokes solutions in plane domains. Namely, we obtain some optimal estimates of the difference between mean values of the velocity over two concentric circles in terms of the $L^2$-norm of the vorticity in the annulus between the circles. The sharpness of the estimates is demonstrated by some unexpected counterexamples.

We propose a Quantization-Constrained Parameter Method (QCPM) for determining the Woods-Saxon (WS) potential depth $V_0$ and diffuseness $a$ in studies of $\alpha$ and cluster decay half-lives. In this approach, the parameters are derived analytically by imposing the Bohr–Sommerfeld (BS) quantization condition, thereby eliminating the need for parameter fitting. The calculated diffuseness values exhibit a strong dependence on nuclear shell structure. The reliability of the QCPM-derived WS potential is validated through optical model analyses of elastic $\alpha$ scattering data. The resulting $\alpha$ and cluster decay half-lives show good agreement with experimental values. Our results highlight the significant role of daughter-nucleus deformation in improving the consistency between theoretical predictions and experimental data. In addition, we introduce a modified Woods–Saxon (mWS) potential that effectively approximates the surface behavior of folding potentials incorporating the nuclear medium effect. This modification leads to closer agreement with the measured half-lives. Free of adjustable parameters such as $V_0$ and $a$, the QCPM enhances the predictive power of phenomenological potentials for $\alpha$ and cluster decay studies.

Contact-rich manipulation requires robots to regulate force against surfaces whose geometry deviates unpredictably from training conditions. Trajectory-based imitation learning, which reproduces observable outputs, breaks down under such shifts. We propose Impedance Cloning, which instead imitates the biomechanical priors that generate motion – the stiffness and equilibrium point – and thereby passively absorbs contact uncertainty. Because these parameters encode intent rather than outcome, they generalize across surface geometries where trajectory reproduction does not. We extract them from bilateral teleoperation demonstrations via a particle filter without force/torque sensors and evaluate the framework on two CRANE-X7 manipulators. In a wiping task with joint-space actions, the trajectory-based baseline loses contact below -6 cm, whereas the proposed method maintains a consistent 4-5 N contact force above -6 cm, with a gradual decrease below; with Cartesian-space actions, its force-height slope over 0 to +8 cm is 0.13 +/- 0.03 N/cm, versus 0.34-0.83 N/cm for fixed-impedance baselines. In a pick-and-place task with 10 diverse cups (100 trials), the proposed method succeeds in 84 trials, outperforming the fixed-impedance baseline (74/100) and performing comparably to a variable impedance control baseline (82/100) with one demonstration instead of ten. In a grasping task, the representation reduces torque tracking error with both ILBiT and Mamba backbones, confirming its generality across architectures.

Existing attacks and defenses for diffusion-based large language models (dLLMs) target specific vulnerabilities but lack a shared framework explaining why attacks succeed. We propose one by interpreting safety alignment as shaping the denoising energy landscape: a well-aligned model routes harmful queries toward safe outputs through an energy barrier that separates the two regions. Current jailbreak attacks reduce to two strategies for circumventing this barrier: obscuring the query's safety disposition at initialisation, or intervening mid-trajectory to force the denoising path across the energy barrier. From this perspective and the result that masked diffusion models minimise kinetic energy during denoising, we derive three complementary, training-free detection signals: a step-0 ratio that reads the initial safety disposition from the logit distribution before generation begins, and two trajectory-velocity signals that track kinetic energy in complementary subspaces of the logit space. An attack must either reveal its intent at initialisation or expend kinetic energy to cross the barrier in at least one monitored subspace, so the three signals cover each other's blind spots in the energy budget by construction. Evaluation across three dense dLLMs (LLaDA-8B, LLaDA-1.5, Dream-7B) and a sparse mixture-of-experts dLLM (LLaDA-MoE-7B) confirms this complementarity. In stress tests of known attacks, every configuration that evades detection also fails to produce harmful content, suggesting that the detection and barrier-crossing thresholds are hard to separate.

One-step generators enable high-quality visual generation with a single network evaluation, but their post-training is difficult: general implicit generators provide neither tractable likelihoods nor denoising trajectories, and many rewards are non-differentiable. We introduce Reward-Weighted Transport Distillation (RWTD), a post-training method that requires only generated samples and scalar reward evaluations. Rather than aligning solely to the conventional reward-tilted reference distribution, RWTD constructs an adaptive target that mixes separately tilted current and reference distributions. The current component incorporates improvements discovered during training, while the reference component anchors the target to the pretrained generator. RWTD realizes this target through feature-space optimal transport and fixed-point regression. Theoretical analysis shows that the fixed-point distributions of RWTD interpolate between off-policy reward tilting of the reference and on-policy tilting of the current model, providing a principled approach to balancing reward adaptation with retention of prior knowledge. Empirically, RWTD substantially improves the GenEval score of the one-step SANA Sprint 1.6B backbone from 0.73 to 0.80, while separate preference alignment experiments demonstrate strong cross-reward generalization that yields balanced improvements and preservation of compositional capabilities.

Simultaneous speech-to-text translation (Simul-S2TT) consists of generating partial translations while the incoming audio frames are processed by the system. However, the streaming nature of this setup creates the challenge of deciding the best moment to perform an accurate translation while minimizing the delay. To address this challenge, we utilize the cross-attention mechanism of the encoder-decoder architecture to find the right alignment between the input speech frames and the target text tokens. In this paper, we propose the Recent Frame Attention Policy (RFAP) and the Dual-Condition Attention Policy (DCAP) that allow offline trained speech-to-text translation models to be used in streaming scenarios without requiring additional training. Results on three different language translation pairs over the CVSS-C corpus show that the RFAP is able to surpass other policies with gains of up to 4.0 BLEU while reducing the translation delay by almost 1 second. Moreover, the DCAP is able to preserve a high translation quality when the latency is very low.

Effective behavioral intervention in chronic disease management requires a sequence of incremental steps, each grounded in what similar individuals have achieved. Multimodal records of daily behavior make such evidence available outside the clinic. Counterfactual explanation, an explainable AI approach, can turn these records into guidance, yet endpoint-only methods return a target state without a route to it. We propose POROS (Peer-Grounded Optimal Routes Over States), a domain-agnostic framework rooted in Bandura's self-efficacy theory and Festinger's social comparison theory. POROS constructs a Behavioral Progression Graph, a directed acyclic graph over observed patient states in which every edge requires both peer-grounded behavioral proximity and strict improvement in the observed health outcome. Every edge is bounded by a feasibility threshold derived from the cohort's day-to-day behavioral changes. Minimum-cost paths through this graph decompose behavioral gaps into incremental, peer-grounded steps. We evaluate POROS on two independent longitudinal cohorts of patients with type 1 diabetes, with a per-patient predictor estimating individual time in range (TIR, blood glucose within 70-180 mg/dL) along each path. Compared with an endpoint-only baseline, POROS returns multi-hop paths to the same target for 14.4% and 19.8% of days below the 70% TIR threshold, reducing the per-step change in carbohydrate intake, insulin dose, and bolus timing by 38-42% on the larger cohort. POROS also supports improvement within the same outcome class, connecting 78.1% and 83.7% of outcome-ordered pairs of well-controlled days. Together, these findings demonstrate how POROS translates multimodal health records into incremental, peer-grounded guidance for chronic disease management through observational path planning.

Multi-teacher on-policy distillation (MOPD) integrates specialized capabilities into a single student, but existing practice typically hard-routes each prompt to a domain-matched teacher for the entire rollout. This dependence on prompt-level domain labels restricts using unlabeled training mixtures and leaves complementary signals from other teachers unused. We introduce MOPD-Router, a framework that routes supervision over the full teacher pool at each token, without domain labels or training a separate routing model. Its plug-in interface supports different metrics for selecting and weighting teacher-specific OPD signals. Within this interface, we propose ExpertAlign, which scores each teacher by whether its correction to the student at the current token expresses the specialization that teacher acquired during post-training, and compare it against two reference metrics built on teacher confidence (Entropy) and teacher-student discrepancy (Novelty). Experiments on unlabeled and domain-labeled training mixtures under strong-to-weak and same-size distillation scenarios show that ExpertAlign achieves the strongest overall performance in all four settings. On unlabeled data, it improves the overall score by 5.88 (+12.3%) points over Mean aggregation; on domain-labeled data, it outperforms standard MOPD by 3.95 (+7.8%) points without using available domain labels. These results demonstrate token-level routing can exploit cross-domain complementary supervision, and reduce exclusive reliance on prompt-level domain assignment. Code is available at: https://github.com/TURLEing/MOPD-Router.

Deploying small language models (SLMs) on offline, resource-constrained edge devices such as remote sensing satellites presents a fundamental challenge: their limited reasoning capacity hinders reliable execution of multi-step agent tasks requiring complex tool orchestration. Existing plan-solve paradigms rely on prompt-based enforcement, which our experiments show SLMs almost entirely disregard: weak models fail to invoke the plan. We propose PTC-Decoder (Plan-Tool Constrained Decoder), a training-free, plug-and-play decoder framework that combines (1) a Plan-to-Act paradigm, which elevates planning to an atomic tool and forces its invocation at the first inference step, and (2) TC-Decoder, a deterministic finite automaton that imposes token-level hard constraints on tool names while preserving freedom over parameter generation, thereby retaining SLM reasoning capability. Evaluated on 200 real remote-sensing satellite tasks across 7 SLMs, PTC-Decoder yields a statistically significant mean overall score gain of +1.21 (p<0.01), 95% CI [+1.13, +1.29]), with consistent improvements across models and other datasets. An ablation study that removes TC-Decoder causes substantial performance degradation across all quality metrics without reducing computational cost, confirming TC-Decoder as the primary driver. PTC-Decoder thus offers a lightweight yet effective solution for improving step-level reliability, with final-answer accuracy remaining an open challenge. In essence, we enforce plan adherence by constraining the permissible output vocabulary during inference, without requiring retraining.

Kilonova spectra provide information on rapid neutron-capture nucleosynthesis in neutron star mergers. However, it remains challenging to identify individual elements in kilonova spectra, primarily due to incomplete atomic data for heavy elements at near-infrared wavelengths. Recent work demonstrated that spectra of chemically peculiar stars offer excellent laboratory for studying kilonova spectra. In this paper, to expand our understanding of strong transitions of heavy elements at near-infrared wavelengths, we present high-resolution near-infrared spectra of 14 chemically peculiar stars covering a range of effective temperatures, surface gravities, and abundance patterns, obtained with Subaru/IRD. The Sr II, Ce III, and Gd III lines, previously suggested to be among the strongest transitions, are detected in the spectra of most of our targets, confirming their intrinsic importance. In addition, Eu II and Nd III lines are detected in several stars. By comparing observed equivalent widths and abundance ratios, we show that under a highly heavy-element-enriched abundance pattern, Eu II and Nd III lines may contribute to kilonova spectral features comparably to Sr II. Importantly, we find no unidentified lines systematically stronger than these transitions despite the larger stellar sample. These results further support the primary importance of the Sr II and Ce III transitions and provide new insights into the potential contribution of Eu II and Nd III.

We present results from an extension to the Keck Baryonic Structure Survey (KBSS) that focuses on the Inner Circumgalactic Medium (CGM) of QSO Line Of Sight Emitting galaxies at z$\sim$2.3 (InCLOSE). We analyze two low-mass galaxies $\log{(\rm{M_*} / \rm{M_\odot})} \leq 9$ that are within small a projected distance of a QSO $D_{\rm tran} \leq 50~{\rm kpc}$ ($D_{\rm tran} /R_{\rm vir} \leq 0.75$). One galaxy is detected as a bright Lyman- Emitter with Keck/KCWI (confirmed with follow-up Keck/MOSFIRE spectra), and the other as a serendipitous line emitter with Keck/MOSFIRE. Both galaxies have nebular and morphological properties consistent with those of typical low-mass z$\sim$2.3 star-forming galaxies. Analysis of their CGM absorption as seen with Keck/HIRES spectra of the background QSOs shows no detections of low-ionization metal absorption (low-ions; e.g., Si II), ubiquitous detections of intermediate- (e.g., C IV) and high-ions (O VI), kinematically complex absorption ($\geq$7 components per galaxy halo) spread over $|\Delta v|\pm 150~{\rm km~s^{-1}}$, and no unambiguously unbound gas is detected in either galaxy halo. We analyze the thermal properties of a subset of CGM components, finding that the majority (6/10) have temperatures consistent with heating from the metagalactic UV background ($\log{(T/\rm{K})_{\mathrm{med}}}=4.0$) while the remainder possess short-lived, intermediate temperature gas that would require additional heating or rapid replenishment ($4.6 \leq \log{(T/\rm{K})} \leq 5.1$), the internal energy of these absorbers are dominated by thermal broadening, their internal (turbulent) velocities are all subsonic, while their motions through the halo are likely supersonic. These results hint that the z2.3 CGM changes with stellar mass in terms of kinematic complexity and unbound gas fraction, while thermal properties remain similar.

Hierarchical vision-language-action (VLA) systems consist of a high-level vision-language planner and a low-level action expert that generates continuous actions. This hierarchical design has practical value only if the planner can generate plans fast enough to meet real-time control requirements, and the resulting plans actually contribute to the generation of action. We study one such system, a waypoint hierarchy pipeline adapted from $\pi_{0.5}$, and find that neither requirement is satisfied. This baseline relies on token-level autoregressive decoding (Token-AR) to generate a waypoint plan, requiring 57 very expensive vision-language model (VLM) forward passes. However, we find that erasing the waypoint endpoints has little effect on task success. Two findings reveal the misalignment of planner-executor: the planner generates outputs at an excessively fine granularity, and the executor underuses plans as a control condition. We address the latency issue with waypoint-aligned block-autoregressive decoding (Block-AR), and plan underuse issue with normalized goal modulation (NGM), a layer-wise goal path constrained by phase gating and anti-shortcut training so that the waypoint influences action generation maintaining other signals. Our method reduces the maximum number of VLM forward passes from 57 to 8 on LIBERO, including one prefix prefill, and achieves an $8.7\times$ reduction in planning latency on a Rokae dual-arm robot. With normalized goal modulation and anti-shortcut training, Block-AR's success rate on LIBERO-Long increases from 91.0% to 96.2%, while its average success rate across the four suites increases from 95.85% to 98.45%. On three bimanual tasks with this robot, success rates remain comparable across methods.

Perceived speech decoding based on non-invasive brain-computer interface (BCI) signals has been extensively studied in recent years. Research in this field primarily faces two challenges: extracting neural representations with rich spatiotemporal information and achieving cross-subject generalization. Although separate studies have proposed methods to cope with these issues, a unified approach that simultaneously tackles both challenges remains lacking. To fill this gap, we propose the Subject-Invariant Cross-Modal Perceived Speech Decoding (SICMD) method, which integrates functional magnetic resonance imaging (fMRI) and magnetoencephalography (MEG). We conduct comprehensive analyses of the fusion method, fusion position, encoder architecture, and model inputs. Our results demonstrate that the proposed method improves Top-1, Top-10, and Rankacc by more than 10.6%, 10.1%, and 1.7%, respectively, compared to baseline methods in cross-subject perceived speech decoding tasks, while reducing training costs by 88.8% and 60.5% compared to multi-subject and intra-subject decoding settings. Further visualization experiments also confirm the effectiveness of our approach.

Generalizing Electroencephalography (EEG)-based mental workload recognition to unseen subjects remains a formidable challenge due to severe inter-subject variability. While functional brain graphs effectively model distributed cognitive dynamics, their inherent subject-specificity induces two coupled distribution shifts: a class-conditional topological shift in the underlying functional connectivity, and a predictive mechanism shift in the learned representation-to-label mapping. Motivated by the subject-induced distribution shifts, we propose CDBG, a Causally motivated Dual-invariance learning framework for Brain Graphs. CDBG disentangles and mitigates these shifts via a two-stage rationale learning pipeline. First, it employs stochastic edge masking to extract sparse, workload-predictive graph rationales, regularized by workload-conditional Laplacian spectral alignment to enforce topological invariance across subjects. Second, it applies subject-wise Invariant Risk Minimization (IRM) to the graph representations, ensuring environment-wise risk stationarity. Extensive experiments on a self-built air traffic controller EEG cognitive workload dataset and multiple public datasets under a strict leave-one-subject-out protocol demonstrate that CDBG significantly outperforms state-of-the-art cross-subject and graph-based baselines, improving the Macro-F1 score by up to 4.23%, while simultaneously providing neurophysiologically interpretable functional rationales.