Papers

Differentiable logic gate networks, which operate using only logic gates, have recently attracted attention as an efficient alternative to conventional neural networks. However, despite their efficiency, the scaling behavior of logic gate networks remains underexplored. By contrast, scaling model capacity is a central design principle in deep neural networks and typically leads to improved performance. This discrepancy raises a key question: Can similar scaling benefits also be achieved in logic gate networks? In this work, we focus on width as a primary scaling axis and conduct a systematic analysis of its behavior in logic gate networks. We observe that naive width scaling often introduces redundancy among logic kernels, limiting the effective use of additional kernels and leading to performance saturation. To address this limitation, we propose a dynamic logic kernel framework that reorganizes kernel utilization by promoting specialization across kernel groups. This enables the network to better utilize increased width via input-dependent kernel routing, while ensuring that both routing and computation are implemented entirely with gate-level Boolean operations at inference time. We further find that kernel redundancy is most pronounced at the first gate level, motivating an early-stage dynamic logic kernel strategy that concentrates adaptation at this level. Experimental results demonstrate that our approach improves kernel utilization and increases kernel diversity, leading to higher accuracy with improved parameter efficiency.

Quantum low-density parity-check (QLDPC) codes are promising candidates for low-overhead fault-tolerant quantum computation, but their practical use requires fast, low-complexity, and reliable decoders. Belief propagation (BP) is attractive because of its local message-passing structure, yet standard flooding BP often suffers from convergence failures on QLDPC codes due to short cycles, degeneracy, and symmetric decoding trajectories. Learned sequential BP improves convergence by using a reinforcement-learning policy to choose variable-node update orders, but a single learned trajectory can still be sensitive to unfavorable local Pauli decisions. We propose a learned parallel bit-flipping sequential BP decoder with a new quaternary score combining syndrome gain, a quantized Pauli log-likelihood penalty, and Q-table lookahead at hypothetical post-flip states. Our decoder first runs learned sequential BP for a fixed number of iterations. If the syndrome is not satisfied, it constructs quaternary bit-flipping candidates, each corresponding to changing the current Pauli decision of one qubit. The selected candidates initialize independent learned sequential BP continuations from the same decoder state. Since these continuations are independent, they can be executed in parallel, so testing several candidates mainly increases parallel hardware resources rather than sequential decoding latency. Simulations on representative QLDPC codes over the depolarizing channel show that the proposed decoder improves the reliability of learned sequential BP while having a parallel low-latency structure.

Unmanned aerial vehicles (UAVs) searching for ground targets from a high altitude face a unique challenge, particularly when the target is already within the field of view but effectively unobservable because of its small apparent scale. Standard object detectors often underperform in such scenarios because of resolution downscaling and limited context. In contrast, active object search frameworks address this challenge by directing the agent to a suitable pose to gather richer visual information. However, flight regulations in urban airspace often restrict such physical movements for UAVs. As an alternative active sensing approach, the UAV can leverage the pan-tilt-zoom (PTZ) mechanism of the onboard camera to dynamically adjust its field of view and sequentially gather enhanced visual information from specific regions of interest. Once the candidate locations of the target are identified, it can deploy more expensive object detection schemes, such as an ensemble of multiple models, to get better reasoning at a fixed scale. In this work, we formulate the sequential exploration with PTZ operation as a partially observable Markov decision process (POMDP), in which the agent maintains a belief state over the target's true location. To solve the POMDP, we deploy partially observable Monte Carlo planning (POMCP), where we condition the sensing reliability on target object scale and deploy selective ensemble detection as an additional reasoning step. We validate our methodology with experiments in a photorealistic simulator under different environmental conditions and vehicle states, showing detection of ground targets at variable scales with significantly fewer steps and minimal dependence on sensor resolution compared to baseline methods.

We connect the spurious-reward paradox to a model's reachability and propose random-reward reinforcement learning (RL) as a useful tool for the probing enterprise, addressing a decade-long debate over what probing performance actually reveals about a model. There are two prevailing explanations for the surprising finding that even random rewards can improve the performance of large language models (LLMs): one attributes the gains to particular mechanisms within RL training; the other to data contamination. Our results motivate a different view: spurious-reward RL can probe a model's reachability, or what further training can attain from its current state under specified constraints, beyond what is reflected in its current performance. Two OLMo checkpoints with the same accuracy on synthetic arithmetic (3.5%), for example, reach 8.5% and 55% in their best runs under the same correctness-rewarded RL. Examining OLMo checkpoints across pre-training and mid-training reveals three distinct regimes of training response: early on, RL produces little improvement even when correct answers are rewarded; later in pre-training, rewarding correct answers becomes effective while random rewards remain weak; and, upon entering mid-training, even random rewards can produce large gains. A similar ordering appears in a number-masked supervised fine-tuning (SFT) analysis of these checkpoints, suggesting that the pattern is not specific to a particular RL mechanism. Moreover, RL with random rewards offers a distinctive perspective on what training can make an LLM do, since its reward signal supplies no information about which answers are correct. By asking what training can attain without correctness feedback, it addresses the label-leakage side of a central problem in decodability-based probing: whether a successful probe reveals the model's capabilities or learns the task itself.

We give an explicit three-dimensional Hadamard matrix of order $22$. The example is presented by listing its twenty-two $22\times22$ two-dimensional slices.

Retrieving the right tables is a prerequisite for Text-to-SQL over realistic databases. Dense table retrievers rank schema elements independently, but this ignores a key source of evidence: some required tables are not mentioned in the question and become identifiable only through their join relationships to already relevant tables. We introduce JOINGR, a join-aware table retrieval method that treats the database join graph as the retrieval space. Columns are represented as graph nodes, while intra-table and foreign-key relationships are represented as typed edges. Given a question, JOINGR selects semantically similar anchor tables, traverses join edges with a query-conditioned scorer, and aggregates the resulting edge deposits into table scores. The scorer is a lightweight MLP on top of frozen query, node, and edge embeddings, trained with a pairwise margin loss over gold tables. On BIRD and Spider datasets, JOINGR is competitive with the strongest retrieval baselines. On BEAVER, a challenging enterprise benchmark with multi-hop table requirements, JOINGR substantially improves recall over dense retrieval and re-ranking baselines. Cross-domain experiments show that the learned scorer transfers across benchmarks, indicating that the method captures reusable joingraph traversal behavior.

The scalar auxiliary variable (SAV) method replaces the nonlinear part of the free energy by a positive scalar $r(t)=\sqrt{\mathcal{E}_{_\mathcal{N}}[\phi]+C}>0$, thereby yielding linear, unconditionally energy-stable schemes for gradient flows. At the discrete level, however, the standard backward Euler and Crank–Nicolson discretizations provide no guarantee that the computed $r^{n+1}$ remains positive, an inconsistency with the continuous definition that contradicts the square-root ansatz and may compromise long-time robustness. Although the SAV method has been widely applied, this subtle but consequential issue has received little attention. We first characterize this failure quantitatively by deriving a sharp criterion and a sufficient condition on the time step size, and construct an explicit counterexample showing that sign loss occurs for parameters of practical relevance. Rather than modifying the definition of $r$ as in existing positivity-preserving variants, we retain the square-root form and reformulate the discrete evolution from $r_t$ to $(r^{2})_t$, which converts the scalar equation into a convex quadratic with a strictly negative constant term, always yielding a unique positive root. For the Crank–Nicolson scheme, the product-form discretization $r^{n+1}r^{n}$ preserves this quadratic structure, while conventional alternatives do not. The resulting schemes incur the same computational cost as the original SAV method and are proved unconditionally energy-stable. Numerical experiments for the Cahn–Hilliard equation confirm the predicted positivity, energy stability, and convergence rates.

Activation steering provides a simple, training-free mechanism for controlling attributes of generative models such as sentiment, style, and helpfulness. However, standard approaches such as Difference-in-Means apply a single input-independent steering vector across all activations, limiting expressivity and ignoring the local geometry of the activation space. We propose Kernelized Activation Steering (KAS), a unifying framework that lifts activation steering into a reproducing kernel Hilbert space. KAS formulates steering as an optimization problem expressed purely via kernel evaluations, yielding an implicit, activation-dependent steering score without constructing explicit feature maps. Unlike DiM, KAS induces locally adaptive steering: each activation is modified according to its relative position with respect to source and target reference sets, producing a nonlinear steering field over the representation space. Importantly, DiM is recovered as a special case under a linear kernel, while richer kernels enable geometry-aware interventions. Across standard activation steering tasks, including jailbreaking LLMs and image style control, KAS outperforms or is on par with the existing methods.

The composition of gamma-ray burst (GRB) jets remains uncertain. We analyze eight Fermi Gamma-ray Burst Monitor (GBM) GRBs with reported redshift information, selected from 126 events by prespecified fluence and spectral criteria. After excluding compound-model DIC evaluations classified as unstable by the posterior-stability tests, a stable added-blackbody (BB) comparison gives $\Delta\mathrm{DIC}>10$ in 203 of 215 accepted bins (60–100% per burst); independently, a multicolor-blackbody plus power-law (mBB+PL) decomposition gives $F_{\rm mBB}/F_{\rm tot}>0.5$ in 214 of 215 bins. The cutoff-power-law (CPL)–mBB correlations ($r=0.83$ and 0.78) exceed the Band–mBB correlations ($r=0.41$ and 0.48), demonstrating model dependence. Time-integrated BB fractions span 0.005–0.311, with six of eight below 0.12. We then apply the non-dissipative photosphere prescriptions of \citet{2013A&A...551A.124H} (H2013) and \citet{2015ApJ...801..103G} (G2015). After excluding boundary or physically inadmissible inversions, the H2013 numbers with $\epsilon_{\rm Th}>0.5$ are 1/2, 1/4, and 2/8 among valid solutions at $R_0=21$, 100, and 1000~km, respectively; the corresponding G2015 numbers with $\sigma_0<1$ are 0/1, 1/4, and 1/8. The large number of invalid solutions at small $R_0$ (six and seven at 21~km for H2013 and G2015) is itself evidence that those assumed radii are incompatible with the adopted inversion for most events. Among bursts having valid solutions at more than one radius, increasing $R_0$ generally lowers $\epsilon_{\rm Th}$ and raises $\sigma_0$. Thus launch radius, efficiency, and spectral decomposition dominate the conditional jet constraints; no universal jet class or unique non-thermal mechanism is inferred.

Accurate glioma segmentation across pre- and post-treatment MRI remains challenging because treatment alters both image appearance and the target regions. In particular, the resection cavity (RC) is defined only for post-treatment cases, complicating joint learning from pooled data. We present RC-aware nnU-Netv2, a simple yet effective glioma segmentation framework that combines lesion- and boundary-aware single-cohort training, an RC-aware joint learning objective, and treatment-status-guided model routing at inference. During pooled training, the joint objective applies both standard and boundary-weighted binary cross-entropy to all four region channels, while employing Dice supervision for channels with non-empty targets and a controlled false-positive penalty for channels with empty targets across the mini-batch. This design explicitly accommodates the absence of RC in pre-treatment cases. Our approach uses only the standard nnU-Netv2 data augmentation and a single specialized model per case, without additional synthetic tumor generation or multi-model ensembling. RC-aware nnU-Netv2 ranked second in the BraTS 2025 Lighthouse Challenge Task 1 (BraTS-GLI 2025), with segmentation performance comparable to that of the first-place method, which relied on on-the-fly GliGAN-based synthetic tumor augmentation and multi-model ensembling. On the official blind test set, our method achieved mean lesion-wise Dice scores of 0.7878, 0.8709, 0.7923, and 0.8715 and mean NSD@1.0 scores of 0.8309, 0.8712, 0.7980, and 0.8336 for enhancing tumor, RC, tumor core, and whole tumor, respectively. In paired post-treatment analysis, the RC-aware joint model improved mean lesion-wise Dice by 0.042 (95% CI: [0.028, 0.057]) and mean NSD@1.0 by 0.043 (95% CI: [0.028, 0.059]) compared with the joint baseline.

Frequency-domain physical neural networks (PNNs) have emerged as a promising analog computing paradigm, achieving a reduced number of devices, enhanced robustness, and high inference accuracy. However, existing demonstrations across various physical domains, such as optics, acoustics, and electronics, are mostly specific to the target devices or platforms, usually requiring hardware-dependent models and post-fabrication parameter tuning. Here, we demonstrate cross-platform implementations of frequency-domain PNN across different wave-based computing platforms using identical, pre-trained parameters. By exploiting shared second-order nonlinear processes, the PNN model is implemented on optical, microwave, and acoustic-wave platforms without the need of fine tuning the parameters. Evaluated on a unified four-class classification task, the frequency-domain PNN achieves high inference accuracies of 97.6% in optics, 98.4% in electronics, and 98.2% in mechanics. Further, we systematically compare related performance metrics across these three distinct physical domains, paving the way to a universal neural network across different physical platforms.

Quantized large language models are increasingly deployed on edge devices for their low latency and energy efficiency. However, model quantization weakens alignment safeguards, leaving qLLMs (quantized large language models) highly vulnerable to jailbreak attacks. To address this challenge, we present MOMAT (Mixture of Multiple Atlases), a hardware-enhanced safety framework that combines structured knowledge retrieval with low-power defense acceleration. Each atlas represents a semantic cluster of harmful or benign sample sets and policy templates, enabling domain-localized Retrieval-Augmented Generation guarding that mitigates the curse of dimensionality and the resulting semantic sparsity problem in large, heterogeneous safety databases. MOMAT retrieves top-$k$ similarity features from all atlases for each prompt and evaluates them using a lightweight MoE (Mixture of Experts) detector, while a CiM (Compute-in-Memory)-accelerated similarity engine performs fast, low-power atlas-local retrieval. MOMAT's CiM-based retrieval accelerates a 100-query batch from 15,052.44 ms to 3,207.21 ns (a $4.69 \times 10^6\times$ speedup) and reduces energy from $8.1 \times 10^7$ $\mu$J to 3.32 $\mu$J, yielding an approximately $2.5 \times 10^5\times$ energy reduction over DRAM-based (Raspberry Pi) baselines. Red-team evaluations across standard benchmarks show that MOMAT matches the defense performance of state-of-the-art methods while avoiding benign overkill and providing substantial efficiency gains, demonstrating that CiM-based modular defenses can make edge-deployed qLLMs both safer and more energy-efficient. We will release the full 223.2k-sample dataset to foster future research.

Three-dimensional (3D) information is encoded in four-dimensional scanning transmission electron microscopy (4D-STEM) through parallax between the virtual images formed at each detector pixel. In this paper we connect the encoding and retrieval of depth-dependent information in 4D-STEM to two other widely-used 3D imaging methods, tomography and light-field photography, and derive its 3D phase contrast transfer function. Viewing the 4D data as a collection of angular slices yields a collection of sinograms, providing an intuitive visual representation of the 3D information transfer. Within the small-angle approximation, tilt-corrected bright-field (tcBF) depth sectioning is equivalent to non-iterative tomographic reconstruction. Alternatively, the 4D data also can be mapped to light-field (or plenoptic) imaging, which acquires spatially resolved diffraction data in parallel using an array of lenslets. Adapting a light-field imaging algorithm, we show that tcBF depth sections can be obtained by extracting two-dimensional slices from a single four-dimensional fast Fourier transform (4D FFT) of the dataset, yielding a substantial speedup when reconstructing a large number of depth slices. These differing perspectives distinguish the geometric aspects of depth resolution from wave-optical effects that are present in phase-contrast imaging. Depth-resolved aberration-corrected bright-field (acBF) imaging yields improved 3D reconstructions compared to tcBF, removing contrast oscillations and reducing the variation in response with depth. We demonstrate acBF depth sectioning using simulated and experimental data and resolve distinct layers in stacked oxide films and nanoparticle assemblies. These non-iterative volumetric reconstructions from a single 4D-STEM dataset show promise for improved imaging of thick, weakly-scattering samples and for reconstructing low-dose acquisitions.

Sparse view 3D reconstruction is an important and common scenario in multimedia applications, such as augmented reality/virtual reality (AR/VR) content creation, cultural heritage digitization, and certain robotic applications, where only a limited number of randomly captured views may be available. However, sparse views contain only limited 3D information, posing two major challenges:1) too few images are available for matching, making it difficult to build multi-view consistency; 2) insufficient view coverage leads to a lack of information in under-sampled regions, resulting in missing parts of object structure. Existing methods mostly still rely on limited reprojection errors and regularization terms, which are prone to overfitting to a single view and inconsistent appearances across views. In geometrically under-sampled regions, they often rely on heuristic density control, lacking reliable guidance and often resulting in blurring and structural holes.To address these issues, this paper proposes Hierarchical Gaussian Fields (HierGF), which revisits sparse-view reconstruction from a hierarchical geometry-perception perspective and converts limited observations into reliable self-generated supervision beyond fixed priors and heuristic density control. In particular, we transform coarse 3D geometric information and additional 2D generative priors into structured pseudo-supervision through a two-stage geometry-perception backbone network, thereby enhancing multi-view consistency with very few input views. In addition, we introduce a learnable confidence network to guide gradients toward cross-view consistent content, and a geometrically consistent densification module to improve the reconstruction of multi-view alignment and under-sampled regions.

Correlated oxides can switch to a low-resistance state when exposed to a voltage, making these insulator-to-metal transition materials promising candidates for memristor technology. Switching in materials such as VO2 occurs because of a thermal-runaway instability that leads to the abrupt formation of a hot current-carrying metallic filament within the insulator. We study, theoretically and experimentally, this bias-induced switching as a non-linear dynamics stability problem. We construct Landau free energy-like Lyapunov functions to show that both the on-state ("set") and the off-state ("reset") transitions can be described as separate saddle-node bifurcations. Starting from the reaction-diffusion equation for the heat transport, the abrupt snap-back (reset) to the insulating state is systematically formulated as a mean field Landau-like dynamics with the filament width playing the role of the order parameter. The formulation predicts an abrupt first-order phase transition with a break-off of a finite width filament across the current-controlled reset transition. The formulation is quantitatively compared against our electrical and optical microscopy experiments on VO2 thin film devices. Multi-cycle switching dynamics experiments further reveal the emergence of a stable stochastic regime following an initial electro-spatial forming phase. Beyond establishing a cross-disciplinary dynamical framework that may be readily compared to instabilities in various reaction-diffusion scenarios, the work also provides a predictive foundation for the rational design and optimization of Mott-transition-based technologies.

In-context learning (ICL) enables language models to perform new tasks from demonstrations without weight updates. However, every ICL inference requires processing the full set of examples, resulting in inefficient deployments, and how ICL works mechanistically is not fully understood. Prior work compresses ICL into fixed activation vectors extracted from specific layers or positions, but these input-independent interventions fail on complex tasks where the output depends on fine-grained interactions with the input. By analyzing the ICL forward pass, we show that each attention head's output is an affine transformation of its context-masked counterpart, and that the parameters of this transformation are empirically stable across samples for a given task. Building on this, we introduce Task Operator (TO), which replays this transformation as an analytically derived update to the attention output projection. Across lexical, algorithmic, and reasoning tasks, TO achieves the best overall performance among prior methods and substantially narrows the gap between zero-shot inference and ICL. We further show that the extracted knowledge concentrates in a task-specific sparse circuit across layers and positions, and that averaging operators from disjoint demonstration batches enables effective many-shot scaling without expanding the context window. Our code is available at https://github.com/gzxiong/task_operator.

Comets can supply volatile and prebiotic material to rocky planets, but survival of fragile molecules depends on atmospheric entry and impact. Compact planet chains can transfer comets inward at low velocity. We test whether modest planetary eccentricity preserves this route. We follow 64,000 comets from the region between the two outermost planets of an eight-planet chain to an Earth-mass planet on a temperate orbit around a $0.1 M_\odot$ star. A second suite of simulations isolates the last transfer to the innermost planet across initial planet eccentricities $e_{p,0}=0$–$0.05$ and separations of 10 and 30 mutual Hill radii; a $\Delta=12$ comparison tests the influence of the near-3:2 period ratio at $\Delta=10$. At $e_{p,0}=0.05$, the fraction of comets intercepted by the outermost planet falls by 11 per cent and 6 per cent more reach the inner region. The innermost planet's impact probability changes by less than 3 per cent, while the fraction of impacts below 15 km s$^{-1}$ falls from 25 to 18 per cent. The fraction of all injected comets producing impacts below 15 km s$^{-1}$ consequently decreases by 30 per cent. Measurements of the comet-planet speed before the final close encounter show that eccentricity preferentially removes the slowest approaches. Applying an impact-chemistry model gives a 31 per cent lower HCN survival-equivalent yield. Modest planetary eccentricity can therefore increase inward transport while reducing the prebiotic material delivered under chemically favourable impact conditions.

We argue that as video generation extends to longer durations, subject consistency should be evaluated over dynamic subject sets. We therefore introduce DynSC-Eval, an evaluation framework that dynamically tracks eligible subjects throughout their visible lifespans and measures local continuity and global identity preservation using six complementary object-level metrics, with explicit detection of inconsistency events. To validate its effectiveness, we design synthetic experiments that actively inject inconsistency events, demonstrating both the sensitivity of DynSC-Eval and the limitations of existing metrics. Evaluations of diverse models on 5s, 15s, and 60s video generation further reveal substantial subject consistency differences that are obscured by conventional metrics. Beyond evaluation, we construct rewards from DynSC-Eval and apply DiffusionNFT post-training in an autonomous-driving testbed. On 5s generation, our approach reduces the six inconsistency metrics by an average of 13.82% for Wan-2.1-1.3B and 5.66% for SANA-2B, with improvements also observed on the I2V model ReSim. Qualitative comparisons further demonstrate the effectiveness of our method. We then extend generation to 10s and 30s through curriculum learning and show that consistency optimization remains effective while largely preserving other capabilities.

On cooling liquids towards their glass-transition temperature, the dramatic increase of their relaxation times is accompanied by changes in particle dynamics at two distinct temperatures: A high-T crossover, where particles become temporarily caged by their neighbors, and a low-T crossover, where the cage escape mechanism changes. While there is some evidence that the former is associated with a change in local particle arrangement, no structural modification has so far been detected across the low-T crossover, fueling scepticism about the relevance of structure for glassy dynamics. Here, we introduce a novel four-point correlation function which allows to determine a structural length scale characterizing cage anisotropy. Extensive molecular dynamics simulations reveal that this scale, as well as the mean structural length scale of the glass-former, extends into the medium range, i.e., significantly exceeds the particle size. Strikingly, the difference between these two scales - a measure of the degree of cage anisotropy - peaks at both crossover temperatures. We discuss how the presence of these peaks enables understanding the nature of the change in the microscopic transport mechanism at the two temperatures, thereby linking both dynamical crossovers to a single structural observable, the anisotropic medium-range order. This fundamental insight demonstrates that structure, extending well-beyond the local cage, is an essential ingredient for understanding the relaxation dynamics of deeply supercooled liquids.

Density peak clustering (DPC) connects each observation to its nearest neighbor of higher density and identifies cluster centers as high-density observations with unusually large nearest neighbor uphill shifts. The resulting uphill paths from observations to cluster centers, however, can be irregular and unstable in low-density regions, making the clustering assignments sensitive to local perturbations and obscuring the population geometry of the DPC graph. In this paper, we introduce gradient-guided density peak clustering (GGDPC), which performs a gradient ascent step before each nearest neighbor uphill search. We develop a stability theory that relates the GGDPC graph to the gradient ascent flow of the population density. In particular, we establish consistency of GGDPC under five complementary criteria: recovery of local modes, adjusted Rand index, dendrogram (cluster tree), path length, and waterfall measure. Together, these results provide new statistical, geometric, and topological interpretations of DPC-type clustering algorithms.

We prove that at every infinite cardinality, every group which is neither abelian of exponent greater than two nor generalized dicyclic admits a graphical regular representation, settling the infinite-group part of Watkins's conjecture. We also determine the Cayley index of every infinite group: it is $1$, $2$, or $8$, according to its algebraic type, and in every case the index is attained by a connected Cayley graph. For every infinite group $G$ of cardinality $\kappa$, we construct $2^\kappa$ pairwise nonisomorphic Cayley graphs with exactly the unavoidable inverse-pair symmetries, diameter two, and $\kappa$ common neighbors at every distinct pair. The principal tool recovers a continuous ordinal hierarchy from an alternating adjacency baseline with bounded-degree errors: robust finite patterns identify the initial classes, successive twin quotients recover the layers, and their finite exception packets determine the translation action. The reconstruction applies without a group action and is stable under additional layerwise bounded-degree edits. Further results give closed Cantor-cube families with prescribed finite data in the regular cases, sharp cofinality-dependent graph properties, and optimal three-valued shortest-path metrics.

World models, which simulate an environment and predict how it changes under actions, are increasingly used in real-world applications such as robotics. Complex systems call for the same tool because the effect of an action is not immediate. Seeding nodes for a campaign, or immunizing nodes against an epidemic, changes little on its own; what matters is the outcome that unfolds over the steps that follow. Designing an algorithm that selects such actions to maximize expected performance on a task is inherently iterative, and every candidate must be scored by the outcome it produces. Obtaining that outcome has relied on simulation, whose cost becomes a bottleneck when candidates are evaluated over many sampled trajectories. We propose an action-conditioned Network World Model that learns a network's diffusion dynamics under interventions over time, applies each action to the network, and predicts the outcome that follows. It serves as a fast evaluator inside an algorithm design loop in which a coding agent designs and refines executable algorithms using feedback from full rollouts, action-level credit, and counterfactual probes over alternative interventions. Across eight network tasks and five diffusion models, the designed algorithms match or exceed the strongest reported baseline in 138 of 141 settings while enabling up to 14.5 times faster rollouts than Monte Carlo simulation. Code will be released upon acceptance.

For a fixed digraph $F$, let $\operatorname{ex}_2^+(n,F)$ be the maximum of $\sum_{v\in V(D)}d_D^+(v)^2$ over all $n$-vertex $F$-free digraphs. Ai et al. [arXiv:2606.03520, 2026] asked for which self-converse tournament $F$ one can determine $\operatorname{ex}_2^+(n,F)$. Let $TT_r$ denote a transitive tournament on $r$ vertices and $\text{RT}_5$ denote a regular tournament on 5 vertices. Note that both $TT_r$ and $\text{RT}_5$ are self-converse. Iľkovič [arXiv:2609.05042, 2026] determined $\operatorname{ex}_2^+(n,F)$ for $F=TT_4$. In this paper, we determine $\operatorname{ex}_2^+(n,F)$ for $F$ being $TT_r$ or $\text{RT}_5$, and the latter confirms a conjecture due to Iľkovič.

Collaboration depends on shared context, and technical documentation is one way that context persists across people and AI teammates. Specificity, the amount and exactness of detail expressed in language, shapes what information documentation captures and how precisely that information is communicated. This work audits sentence-specificity scoring artifacts on technical documentation and tests whether scores applied only after generation help choose among fixed LLM-generated revisions. Across Wikipedia and three technical-documentation corpora, the fixed general-domain predictor SpeciTeller and the pinned post-publication author-repository implementation of Ko et al.'s target-adapted predictor produce different corpus orders and same-sentence rank agreement from -0.066 to 0.510. Strict filtering and token-length adjustment change these patterns without reconciling them. In the Gemma set, SpeciTeller ranking raises direction-valid selection from 71.7% to 83.3% (+11.7 points; 95% source-case bootstrap interval +1.7 to +21.7); in the GPT-OSS-120B set, SpeciTeller ranking raises direction-valid selection from 51.7% to 56.7% (+5.0 points; 95% source-case bootstrap interval -6.7 to +16.7), and every primary single-score GPT-OSS-120B interval includes zero. These findings tie score interpretation and decision value to the predictor and candidate set.

A chatbot that says "I will remind you tomorrow" will not run again until the user writes. We call such a promise an empty commitment: a promise of action after the current turn that nothing in the agent's tools or runtime can carry out. Unlike a broken promise, its emptiness is decided by the agent's configuration at the moment of speaking, so it can be detected from a single turn, before deployment or at run time. We define empty commitments on top of commitment semantics, with three failure types, an anchoring condition for promises that a tool could make real, and an outcome taxonomy that separates these failures from honest deferrals and from over-refusal. We build a checker (a setup-blind detector, deterministic feasibility rules, and a response judge) validated against 400 human labels. On a controlled benchmark of 293 follow-up requests across five setups that add one persistence affordance at a time, four open-weight models of 8-14B parameters fail on 45.9% of responses when no tool exists and nothing is stated. A frontier model fails on 4.4%, but it gets there by deferring and asking, not by using the tools it has: promises made without the enabling call remain in every model. Telling the model its runtime, the cheapest fix, cuts open-weight failures nearly in half where nothing is doable and changes nothing where a scheduler exists; a directive capability card removes most failures at the largest cost in over-refusal; running the checker in the loop and rewriting flagged replies removes more at a smaller cost. Code, prompts, model outputs, and human labels are released.

Dynamic treatment regimes (DTRs) are sequential decision rules that individualize treatments to each patient at each treatment stage adapting to their past clinical course. Existing literature typically accommodates each individual's medical history, but overlooks a patient's preferences. We propose a method that incorporates a patient's latent preferences through data augmentation into a tree-based reinforcement learning method to estimate optimal dynamic treatment regimes for multi-stage, multi-treatment settings. For each patient at each stage, we derive the posterior distribution of preferences given responses to a questionnaire, and then subsequently weight multiple outcomes with the estimated preferences to identify the optimal stage-wise personalized decision. For multiple stage situations, we grow a decision tree at each stage and implement the algorithm recursively using backward induction. Our proposed method, named Augmented Patient Preference incorporated Reinforcement Learning (APP-RL) is robust, efficient, and leads to interpretable DTR estimation. The finite-sample performances of the proposed method has been thoroughly evaluated through simulation studies.

We characterize extreme points of the (positive) unit balls of Marcinkiewicz spaces with respect to their natural quasi-norms. Having these results at hand, we show that every positive surjective isometry on a noncommutative Marcinkiewicz space is of elementary form. In particular, a linear mapping on a noncommutative Marcinkiewicz space over an atomic von Neumann algebra with all atoms having the same trace is a positive surjective isometry if and only if it is the restriction of a Jordan $*$-isomorphism which preserves the singular value functions. We also study (not necessarily positive) surjective isometries on weak $L_p$-spaces affiliated a $\sigma$-finite non-atomic von Neumann algebra for $1<p<\infty$, Marcinkiewicz sequence spaces, and weak $\ell_p$ operator ideals, $p>0$, as well as one-parameter groups of isometries on the latter two.

Multi-agent communication aims to help agents benefit from one another's information. Yet improvements in system performance leave a fundamental ambiguity: do they reflect effective communication, a favorable agent architecture, or simply additional reasoning? Because communication methods are commonly evaluated within the systems they were designed for, these factors are difficult to disentangle. Final accuracy further merges corrected errors and corrupted answers into a single outcome, obscuring how communication changes decisions. We introduce Independent–Communicate–Revise (ICR), a controlled framework that evaluates communication as answer revision following independent reasoning. ICR fixes initial reasoning trajectories, measures correction and preservation conditional on both agents' initial correctness, and uses a no-message revision control to quantify gains beyond additional reasoning. Across four reasoning benchmarks, our audit of textual and latent communication reveals that similar aggregate accuracy can conceal substantially different revision behaviors. Compared with transmitting answers alone, full reasoning increases correction while reducing preservation on all four benchmarks, so richer messages amplify beneficial and harmful influence alike. Receiver-policy comparisons on MedQA and GPQA-D further show that a structured verification policy shifts every channel toward greater preservation and lower correction, while its effect on selectivity varies across channels and tasks. These findings challenge treating communication quality as an intrinsic property of a channel. ICR therefore recenters evaluation on selective revision, providing a unified framework for examining how message content and receiver policies jointly produce benefits and harms.

We study a two-sided matching problem between families and daycare centers in which a family may have multiple children with joint preferences, and some children may already be enrolled in daycare centers while seeking transfers. Our model combines and generalizes two well-studied matching frameworks: matching with couples and school choice with initial enrollments. The interaction between joint family preferences and initial enrollments creates fundamental challenges for the existence of desirable outcomes and the design of computationally efficient algorithms. Since stable matchings need not exist, we decompose stability into individual rationality, fairness, and non-wastefulness, and study the trade-off between fairness and non-wastefulness from two complementary directions. To preserve non-wastefulness, we first consider master-list-based approaches and show that they are insufficient in the presence of initial enrollments. We then introduce resettlement, which protects displaced families by allowing them to return to their initial assignments, and develop a greedy improvement algorithm that establishes the existence of matchings satisfying non-wastefulness together with resettlement-based fairness. To preserve fairness, we first examine a generalized cutoff approach and then develop a fairness-preserving non-wasteful algorithm that guarantees fairness together with a relaxed notion of non-wastefulness. We further characterize the computational complexity of these solution concepts. While several fairness notions can be verified in polynomial time, deciding whether a non-wasteful matching satisfying them exists is NP-complete. Resettlement restores universal existence, but computing resettlement-based outcomes is PLS-hard, and verifying the stronger dominance-based fairness notion is coNP-complete.

The correction of systematic errors in TESS light curves is crucial for all astrophysical analyses employing these observations. Here we present a data analysis and a software package, SysCoCoPy, to investigate and directly compare the performance of the Presearch Data Conditioning (PDC) correcting method from the Science Processing Operations Center (SPOC) pipeline and three correctors developed by the community. We incorporate these three correctors based on their implementations in Lightkurve and are particularly interested in the ability of the four correctors to remove scattered light contamination from the Earth and the Moon, which is a key systematic for TESS. We implemented these correctors in SysCoCoPy with a framework that allows an automatic optimization of their parameters to scale the analysis towards increasingly larger samples. SysCoCoPy provides qualitative and quantitative products for the comparison of individual cases as well as statistical results for selected samples. We currently find that while our automatic parameter optimization provides a significant number of successful scattered-light corrections for two of the correctors with a design that favors this purpose, an statistical analysis of our largest sample indicates that PDC is presently more robust, with a larger overall success level of the metrics used.