Papers

Large language models (LLMs) and agentic AI systems are creating rapidly growing inference energy demands as model sizes grow and reasoning trajectories extend. While in practice, many queries do not require the capabilities of the largest available model, and routinely directing such queries to a high-capability model can introduce unnecessary, considerable computation and energy consumption. In this paper, we investigate whether adaptive routing across a heterogeneous pool of LLMs can reduce this energy burden without substantially compromising task performance. We design a language-model-based router that reads in each query and selects an answer model from a fixed candidate pool. The candidate models are first profiled through an offline tournament that records their correctness, latency, power, and GPU energy for each query. Using these measurements, the router is trained through supervised fine-tuning followed by group relative policy optimization (GRPO) with the tailored paradigms. Results demonstrate that learned routing can selectively allocate expensive model capacity based on query context and improve the accuracy-energy tradeoff in multi-LLM serving. Across seven benchmark tasks, we also observe a sharp accuracy-energy phase transition among routers, providing practical insights into improving energy efficiency while maintaining LLM performance.

Over-the-air federated learning lets edge devices transmit their local updates simultaneously, reducing the communication overhead. The resulting waveform, however, has a peak-to-average power ratio (PAPR) that grows with the model dimension, and keeping the amplifier in its linear range leaves two remedies: clipping the peaks or backing off the transmit power. Neither remedy is without cost: i) the clipping distortion appears at the receiver as a bias that cannot be removed, and ii) back-off keeps the signal intact but degrades the average signal-to-noise ratio (SNR). Independent of this trade-off, the transmission remains uncompressed, spending one channel use per model parameter, which keeps large-model training out of reach. To address these challenges, we propose AirGC-CD, an over-the-air scheme that precodes each local update with a partial Gaussian circulant matrix before clipping. In AirGC-CD, the precoder's output is exactly Gaussian regardless of the update's sparsity, so the clipping function is designed for a known distribution instead of inheriting it from the data. This enables the clipping to be inverted on average by a single scalar Bussgang gain in closed form, and we prove that the resulting aggregate is exactly unbiased, with clipping adding only variance. The clipping ratio is then the only free parameter left, trading the variance of the clipping against the SNR loss from back-off, and we derive its near-optimum in closed form. Since the precoder is linear, it also acts as a compressor, reducing the transmission from the model dimension d to the sketch dimension m at a cost of only O(dlog d) via two fast Fourier transforms, whereas a Gaussian sketch costs O(md). Experiments on five image datasets show that AirGC-CD outperforms baseline over-the-air FL schemes in most settings, particularly at low SNR, while using fewer channel uses per round.

The distinction between the spirit and letter of the law is a central issue across research and everyday life, and a growing concern for building safe, intelligent machines. What is this distinction based on, and how can we develop machines that follow the intention behind a rule? We used targeted adaptation that made large language models prioritize the spirit or letter of the law. With minimal modifications, our method significantly changed LLM behavior across diverse measures, novel vignettes, real-world scenarios, and influential legal cases. An analysis of model internals revealed a low-dimensional space with three interpretable dimensions matching a formal pre-specified framework for the geometry of legal concepts. These findings show how legal thought in LLMs may be organized and directed.

Enterprise integration traffic in large Enterprise Resource Planning (ERP) systems move through Electronic Data Interchange (EDI) and the Intermediate Document (IDoc), and the operational tools that surround that traffic remain fragmented. Status records are written for a technical audience, exception handling is performed one document at a time, and process health is read manually from long lists. This work proposes an enterprise process control tower that lets an Information Technology (IT) team observe, diagnose, and improve business workflows from a single operational layer. The design links process health observation, semantic translation of technical status text, machine-learning diagnosis of exceptions with model explainability, grouped recovery actions, and key performance indicators inside one governance framework. The proposed framework is evaluated on a synthetic IDoc event dataset of 84,000 records spanning twelve months, of which 15,120 error records are used to train a five-class remediation-routing model that maps each exception to a standard recovery action. A rule-based router that reproduces current production heuristics serves as the primary baseline. The proposed gradient-boosted diagnosis engine reaches a macro-averaged F1 score of 0.883 on the held-out test partition, against 0.742 for the rule-based router, while Shapely-based attributions expose the drivers of each routing decision to operators. Results are derived from synthetic data and are presented as an internal-consistency demonstration rather than real-world validation. The contribution is an integrated tool design that places diagnosis and governance on the same plane as observation.

In a spiked covariance model, let $\mathcal{H}$ and $\mathcal{B}$ be selected sample and population eigenvector matrices. Their alignment matrix $\Xi=\mathcal{H}^\top\mathcal{B}$ is a latent oracle diagnostic of direction-specific correspondence beyond subspace overlap. We study its uniform pathwise recovery in a nested high-dimensional, low-sample-size experiment with fixed sample size and growing dimension. At every fixed model, an observable spectral contrast consistently estimates the row Gram matrix $\Xi\Xi^\top$. For two or more spikes, however, no measurable estimator sequence achieves uniformly reliable almost-sure path recovery of $\Xi$ over a family having common spike strengths and a common signal subspace but different ordered eigenframes. For every $0<\epsilon<1/\sqrt{2}$, a finite orbit forces each estimator to have some orbit point at which the approximate path-recovery probability is at most a variational bound $u(\epsilon)$, where $u(\epsilon)$ tends to zero as $\epsilon$ tends to zero. The worst-model exact-recovery probability over a countable orbit is zero. Thus the Gram summary is recoverable model by model, while the full alignment matrix is not uniformly recoverable along the nested path.

A fundamental problem in the field of anisotropic macromolecular materials concerns how the flow-induced orientational order connects microscopic structure to macroscopic rheological stress and optical anisotropy. This study considered cellulose nanocrystal suspensions as rigid-rod model systems for the purpose of isolating orientational effects from chain stretching. Simultaneous rheo-optical measurements of birefringence, orientation angle, and shear stress were combined with a Fokker-Planck orientation model and stress decomposition. Our results show that, at low values of the Peclet number, birefringence and total stress exhibit proportionality similar to the stress-optic rule. However, the concentration dependence indicates that this proportionality reflects stress partitioning rather than a unique material coefficient. With increasing flow strength, orientational saturation is accompanied by a reduced relative Brownian stress contribution and a corresponding change in the total stress-based relation. Referencing the optical response to an estimated Brownian stress contribution yields a substantially more unified response across concentrations and flow conditions, supporting a stress-partitioning interpretation of apparent stress-optic rule breakdown.

We show that the equations of motion for magnetically charged $p$-branes in $(n+p+2)$-dimensional AdS reduce to coupled one-dimensional Toda-type equations, and use this structure to construct static toroidal black branes with $T^n$ symmetry and $p$ translation-invariant directions. In the pure-Maxwell case we obtain an exact $\sinh$ solution–a $(p+2)$-dimensional black brane times an $n$-torus, asymptotically $\mathrm{AdS}_{p+2}\times T^n$–whose dilatonic analogue, with a scalar coupled through $e^{a\phi}$, has a running torus modulus and a non-trivial dilaton. A perturbative expansion reduces the linearized system to a pair of Pöschl-Teller equations, one elementary and one solved by associated Legendre functions of non-integer degree, so that the general solution is non-elementary; at weak coupling $\epsilon=a^2$ we construct the $O(a^2)$ backreaction explicitly. From the surface gravity, horizon area, and holographic renormalization we obtain the Hawking temperature, entropy density, and mass density, verifying the first law $d\mathcal{M}=T_H\,ds$ and the traceless conformal stress tensor of the dual $(p+1)$-dimensional theory. At $O(a^2)$ the first law persists, but the dilaton source shifts the Stefan-Boltzmann exponent to $p_{\rm eff}=p+a^2/2$, breaking the clean power law $s\propto T_H^p$ while the normalizable dilaton hair decouples from the entropy.

The notion of extremal subsets was introduced by Assaf, Dranowski, and González as a natural generalization of Demazure crystals. In this paper, we show that the crystal basis of a finite-dimensional irreducible module over a quantum group of finite type has a unique maximal proper extremal subset, whose complement is the set of elements that generate the whole crystal basis as an extremal subset. We conjecture that this complement consists of all nonzero elements obtained by successively applying the lowering Kashiwara operators to a distinguished element, which is constructed from the lowest element of the crystal basis. We prove this conjecture in type A.

Stochastic partial differential equations (PDEs) govern critical engineering and geophysical systems but are challenging to use for real-time control under parametric uncertainty. We present a unified framework that couples Physics-Informed Neural Networks (PINNs) with Polynomial Chaos Expansion (PCE) to construct a fast and differentiable surrogate. The PCE representation enables analytical propagation of parametric uncertainty and computation of the corresponding moments without requiring Monte Carlo sampling. We provide an error decomposition for the PINN-PCE surrogate that separates PCE truncation, stochastic quadrature, and PINN approximation errors. Embedding this surrogate into a stochastic model predictive control (SMPC) scheme enables finite-horizon control updates based on analytic mean and covariance predictions. We further show how the surrogate approximation error can be incorporated into tightened probabilistic constraints. The approach is validated on three benchmarks: the Korteweg-de Vries equation, Burgers' equation, and the two-dimensional incompressible Navier-Stokes equations, representing dispersive, convective, and convective-diffusive dynamics. Across all cases, the surrogate enables real-time control updates while maintaining prescribed risk levels and closely matching the corresponding high-fidelity solvers at substantially lower computational cost.

A dominating set $D$ of a graph $G$ is a fair dominating set if every two vertices outside $D$ have the same number of neighbors in $D$, and the fair domination number $\mathrm{fd}(G)$ is the minimum cardinality of such a set. Caro, Hansberg and Henning, who introduced this parameter, proved that $\mathrm{fd}(G)<17n/19$ for every maximal outerplanar graph $G$ of order $n\geq3$, and asked whether this bound is asymptotically best possible. We show that it is not the case by proving $\mathrm{fd}(G)\leq(7n-3)/8<7n/8$ for every maximal outerplanar graph $G$ of order $n\geq3$, and we exhibit an infinite family of maximal outerplanar graphs with $\mathrm{fd}(G)/n\rightarrow7/9$, so that the best asymptotic constant lies between $7/9$ and $7/8$.

We present SineStim, a portable, 12-channel current stimulator for transcutaneous spinal cord stimulation and peripheral electrical stimulation at kilohertz frequencies. Each channel delivers independent, current-controlled sinusoidal waveforms with amplitudes from 0 to 50 mA (0.1 mA resolution) and frequencies from 0 to 50 kHz (0.1 Hz resolution), with burst modulation modes supported. A custom output stage with a high compliance voltage of $\pm$ 120 V was designed and developed, with galvanically isolated channels and independently programmable stimulation parameters for each channel. The stimulator performance was tested on both passive loads and human subjects. Benchtop characterisation on resistive and resistive-capacitive loads demonstrated a total harmonic distortion between 1 - 8 % across typical operating conditions. Multichannel functionality was demonstrated in two-channel human forearm stimulation experiments. Burst-modulated waveforms with differing channel amplitudes modulated inter-finger force ratios, and channels with small frequency offsets elicited temporal interference force patterns at the beat frequency. Smoothly enveloped bursts of kilohertz sinusoidal waveforms produced negligible stimulation artifacts at steady state in concurrent surface electromyography (sEMG) recordings at motor threshold, in contrast to conventional biphasic square wave stimulation. SineStim delivers precise, isolated, multichannel kilohertz stimulation through a portable device, with performance demonstrated on both benchtop loads and human participants. By combining multichannel spatial control with minimal-artifact sEMG compatibility, SineStim enables future precision non-invasive neural stimulation paradigms based on multichannel optimisation and real-time closed-loop control capabilities not available via existing single-channel stimulators.

Forecasters often know an event is imminent but not the shape, size, or timing of its effect. We introduce Event Signature Transfer (EST), a training-free, model-agnostic operator that turns a completed past event into an explicit forecast scenario. EST removes a source event's own trend and seasonality, then scales and retimes the remaining event signature onto a native forecast, preserving the forecast's linked structure and reducing to it exactly at zero strength. Because it reads only output quantiles, EST applies to any quantile forecaster, with no training, no model internals, at transfer time. Across twelve real episodes and ten synthetic scenarios on Chronos-2, TimesFM-2.5 and Toto-2.0, manually configured EST reduces real-episode WQL by 21.7-90% in-sample. On Chronos-2, it leads eleven of twelve matched comparisons against covariate conditioning, activation editing and raw replay. The operator builds a scenario; it does not estimate its likelihood.

Recent advances in natural language processing have led to molecular Large Language Models (LLMs) with strong performance across diverse chemistry tasks. However, they still struggle to capture fine-grained structure-property relationships, particularly how small, localized modifications alter a molecule's behavior. To address this limitation, we introduce MolSC, a dataset of substituent contributions, defined as property changes induced by attaching specific substituents to molecular scaffolds. Curated from manually annotated bioactivity records, MolSC spans structural-alert liability, target-specific bioactivity, and physicochemical descriptors, and contains 181K substituent-level examples for training. We further propose MolSC-Bench, a held-out evaluation benchmark of 1,541 examples disjoint from MolSC at the scaffold, substituent, and molecule levels. Our experiments show that existing molecular LLMs and strong proprietary models such as GPT-5.2 and Gemini-3-Flash show limited reliability in substituent contribution prediction. In contrast, training on MolSC substantially improves this ability and achieves strong performance across diverse downstream molecular tasks. These results highlight substituent contribution learning as a key component of fine-grained molecular understanding.

This paper studies exact controllability of linear mean-field stochastic differential equations with backward-structure. We establish Hautus criteria and uncover a sharp dichotomy between the zero initial time and positive initial times. At time zero, exact controllability is characterized by a Hautus condition for the mean dynamics, with additional control directions generated by the centered stochastic dynamics. At any positive initial time, exact controllability further requires a stochastic Hautus condition for the centered system. Unlike the classical eigenvector-based condition, this stochastic criterion is formulated in terms of positive-semidefinite eigenmatrices of an associated Lyapunov-type operator. The dichotomy arises from the triviality of the initial sigma-field. We also clarify the controllability relations among the corresponding ODE, SDE, and mean-field SDE systems.

This thesis proposes a framework for "responsible intelligence" to address AI's critical challenges in safety, ethics, and cultural sensitivity. It advances three core areas: First, it improves domain adaptation in specialized fields using active learning and graph-based knowledge to reduce hallucinations. Second, it enhances ethical rigor via a novel decoding-time alignment mechanism that proactively blocks harmful text generation in real-time. Finally, it ensures cultural and multilingual safety through language-specific steering that respects diverse linguistic and social norms. Ultimately, this work provides a blueprint for building next-generation AI that is contextually knowledgeable, ethically sound, and culturally adaptable.

Classical preliminary-test and Stein-type estimators interpolate between restricted and full regression fits, but OLS and chi-squared calibration fail when $p\gg n$ and the restriction is data-adaptive. We propose an honest sample-separated framework built around a common selected-null law. Independent selection data extend a mandatory core; separate estimation data pair a complete-design regularized full-model (FM) fit with a fresh exact-null submodel refit. A maximum residual-association statistic assesses all excluded coordinates. Projected Gaussian draws reproduce its conditional null distribution under homoskedastic Gaussian errors, yielding a finite-sample-valid rank test. The inverse moment of the same design-specific law replaces $q-2$ and calibrates preliminary-test (PT), Stein-type (S), and positive-part Stein-type (PS) rules. We derive exact endpoint-risk formulas, honest conditional validity, null-law concentration, inverse-moment consistency, an explicit selection-failure remainder, and endpoint adaptivity without uniform-dominance claims. Gaussian experiments with 2,000 replications and up to 30,000 predictors compare Ridge and LASSO FMs, tuned by frozen-fold cross-validation, under a common submodel, test, and calibration. PS provides large near-null coefficient-risk gains and approaches the relevant full-model risk under strong departures. A separate CPSS-LASSO/CPSS-MCP audit evaluates data-adaptive selection, while a split-sample DepMap study with 19,152 predictors illustrates prediction gains and sensitivity to error assumptions. The HDMaxShrink R package implements the procedure.

A right sub-gamma bound is described by a quadratic proxy $v$ and a pole $c$, and the pair is not unique: enlarging either preserves it. Fixing $v$ at the variance $V$ removes the ambiguity at quadratic order but is a convention, not a consequence. For centered infinitely divisible laws with finite nonzero variance we determine the entire boundary of the feasible set, the map $v\mapsto c_*(v)$. A Beta$(1,2)$ multiplier applied to the normalised Kolmogorov canonical measure turns feasibility into a one-dimensional comparison and yields an exact variational formula: the frontier is convex, nonincreasing, and its feasible set is convex. The abscissa of convergence of the moment generating function imposes a $v$-insensitive floor on $c_*$, so a law whose variance-exact pole sits on it has a flat frontier, whereas the third-cumulant obstruction exists only at $v=V$. Whenever that pole lies strictly above both floors the frontier drops strictly as soon as $v>V$; if the control is also purely local with $\kappa_3^2/(9V^2)>\kappa_4/(12V)$, the drop has a square-root profile, and in the remaining boundary case a cube-root profile, with derivative $-\infty$ at $V$ either way. We then characterise when the optimised frontier reproduces the exact Chernoff deviation: equality holds at a level exactly when some pair on the frontier is tangent to the cumulant generating function at a Legendre maximiser for that level. At the levels tabulated in a worked example $v=V$ costs four to thirty percent; for a centered exponential a gap persists at every level, reaching thirteen percent.

The equilibrium oblique-detonation polar is a fundamental gasdynamic relation. This relation links the deflection angle, the wave angle, the flow state and the heat release. However, the wave angle is usually determined by numerical iteration. There is no explicit formula to determine the wave angle from a prescribed deflection angle, referred to as the inverse problem, even for the classical oblique-detonation model: a calorically perfect gas with a constant specific-heat ratio and a fixed heat release. This is inconvenient in practical applications. In this work, we solve this inverse problem for the classical model and derive explicit formulae for the wave angle. Inspired by the derivation of the oblique-shock inverse relation, the oblique-detonation inverse relation also reduces to a cubic equation. Heat release only changes the cubic coefficients, so the equation can still be solved explicitly. We also derive the complete analytical branch structure. We further find that the detachment point and the downstream total-sonic point can be determined explicitly by solving a cubic and a quadratic equation, respectively. We extend the explicit formulae to the two-$\gamma$ model and apply them to oblique-detonation calculations with equilibrium chemistry. These explicit formulae for the wave angle and critical points promote the theoretical understanding of oblique detonations and support their practical calculation.

Referring surgical video instrument segmentation (RSVIS) aims at segmenting the instrument in a surgical video, given a textual description. Despite recent progress, current models are trained and assessed on relatively small-scale benchmarks, hindering the development of more general RSVIS. In addition, existing benchmarks support only the single-target expression that refers to one instrument in the video, while overlooking multi-target and no-target referring expressions, restricting the applicability of RSVIS in practical scenarios. Addressing these issues, we propose LD-RSVIS, a new benchmark aiming to facilitate more robust and general RSVIS. Specifically, LD-RSVIS consists of 3,536 surgical videos with 1.09 million frames and covers a broad set of 30 instrument classes from 25 various procedures. By including abundant videos and classes, LD-RSVIS could benefit large-scale training and evaluation of more general RSVIS methods. Besides, unlike existing datasets, LD-RSVIS offers diverse referring settings, including no-target, single-target, and multi-target expressions, which enables the development of more practical RSVIS models in real applications. In order to ensure high-quality annotations, all videos in LD-RSVIS are manually labeled with multiple rounds of inspection and refinement. To our knowledge, LD-RSVIS is the largest and most diverse benchmark for RSVIS. To analyze LD-RSVIS and to provide comparison for future research, we evaluate 12 representative methods, and the results reveal that more efforts are required for improvements. To encourage future research, we present a simple yet effective RSVIS method, dubbed Cascade-RSVIS, that first mines target-specific cues using the complementary multi-cue text information and then employs such cues and textual information for segmentation, achieving promising performance. Our benchmark and code will be released.

Many models of collective behavior write headings, beliefs, and meanings in one experimenter-defined frame. Organisms do not live there: each represents the world in a private space, and comparison requires translation. Consensus becomes an existence problem before it becomes a dynamical one. Reciprocal pairwise translations need not compose consistently around a loop. This return transformation (the holonomy) can expose a mismatch no isolated reciprocal pair can carry. We develop a gauge-covariant theory in which physical predictions are invariant under private relabeling. For reciprocal unitary translations, the kernel of the connection Laplacian is isomorphic to the joint fixed space of loop holonomies: a collective can hold in common exactly what all its loops leave unchanged. A blind-subgroup criterion identifies invisible defects. Common-frame comparison, such as in traditional models of collective behaviour, occupies the flat sector. For independent uniform translations from a finite group $G$, a connected graph with $E$ relations among $N$ individuals is flat with probability $|G|^{-(E-N+1)}$. Exact cycle spectra determine persistence under linear relaxation. The consequence is operational as well as structural. For unrecorded routes under a specified model with isotropic Gaussian source and readout noise, we derive the smallest mean-squared error achievable by any decoder. At fixed signal-to-noise ratio, the jointly preserved feature fraction fixes its long-route limit. Networks with identical loop angles can differ in recoverable content when their preserved axes differ. The relational geometry of a collective sets both the states it can share and the content it can recover.

Chain-of-thought (CoT) can sound plausible yet be unfaithful to the model's underlying reasoning. Most prior work probes CoT faithfulness through input–output behavior or input attributions, leaving internal computation largely underexplored. We instead cast faithfulness as internal concept grounding: Does a large language model's (LLM) CoT reasoning engage the same internal concepts that support the LLM's direct prediction, and do the shared concepts causally drive its answer? Encoding a prediction pass and a CoT pass with a single shared sparse autoencoder (SAE), a reliable approximator of the latent concepts LLMs use, makes their internal concepts directly comparable. We introduce three correlational metrics of concept-level alignment and a causal metric, $\Delta p$, which ablates the shared concepts and measures the drop in answer probability. Across five LLMs and four datasets, concept alignment is generally high, as indicated by the correlational metrics; yet these only identify which concepts are shared, not how much they causally contribute. $\Delta p$ fills this gap: causal faithfulness varies substantially with model depth, peaking at mid-to-late layers rather than the final ones, and model scale reshapes the layer-wise profile. Moreover, causally important shared concepts are not always verbalized in the CoT. These dissociations suggest that faithfulness cannot be reliably assessed from surface-level or representational correspondence alone; assessing it requires causal tests of whether the internal concepts underlying a CoT actually drive the model's prediction.

Understanding the physical world requires more than object recognition, scene description, and short-term visual prediction, as real-world physical systems involve multiple continuous fields, latent causal mechanisms, partial observations, and intervention-sensitive dynamics. We propose FireWorldBench, a benchmark for evaluating complex physical world intelligence in multimodal large language models and agents through coupled-field fire dynamics. Fire provides a canonical stress-test environment, where multiple interacting physical fields jointly shape observable states and temporal dynamics. FireWorldBench is organized along two complementary axes, a physical capability axis and a fire scenario task axis, jointly covering physical-state understanding, temporal dynamics, causal mechanisms, and intervention reasoning. The benchmark comprises 520 fire-world entries, including 494 controlled simulation worlds and 26 real-world-aligned event groups, spanning 47 scene archetypes across 7 environment families. These entries combine structured textual observations, multiple 2D physical-field visualizations, and 3D event-level scene modeling, yielding 9,074 text-image interleaved question-answer pairs across choice-based and open-ended report-generation formats. FireWorldBench evaluates whether models can infer latent physical states, explain underlying mechanisms, forecast coupled-field evolution, and assess intervention consequences from multimodal partial observations, providing a challenging testbed for complex physical world intelligence.

In this paper, we study the unramified Iwasawa module over the cyclotomic $\mathbb{Z}_2$-extension of the real quadratic field $\mathbb{Q}(\sqrt{p_1p_2p_3p_4})$, where $p_1, p_2, p_3$, and $p_4$ are distinct odd prime numbers. We give a criterion for the finiteness of an unramified Iwasawa module of a number field. The criterion is based on the nonexistence of a $2$-group with a certain prescribed quotient. Using this criterion, we construct an infinite family of such real quadratic fields with unramified Iwasawa module of type $\mathbb{Z}/4\mathbb{Z} \oplus \mathbb{Z}/2\mathbb{Z} \oplus \mathbb{Z}/2\mathbb{Z}$ and $2$-class group of type $\mathbb{Z}/2\mathbb{Z} \oplus \mathbb{Z}/2\mathbb{Z} \oplus \mathbb{Z}/2\mathbb{Z}$. This gives the first example of an infinite family of real quadratic fields whose ideal class group has $2$-rank three and whose cyclotomic $\mathbb{Z}_2$-extension is totally ramified and satisfies Greenberg's conjecture.

Threat modelling for wireless and mobile networks is dominated by static, catalogue-driven methods that fix the adversary in advance and never ask whether an attack is feasible against the defence actually deployed; agent-based resilience studies share this limitation, optimising a defender against an exogenous threat profile. We instead endogenise the adversary. Network entities run a decentralised consensus agent-based model (DC-ABM) defence, while an adaptive adversary population co-evolves by allocating a bounded attack budget across vulnerability-mode partitions. The equilibrium is a rankable, feasibility-grounded threat model expressed through three structural metrics: degeneracy-weighted path robustness, trust-weighted functional substitutability, and robust degeneracy. We prove that the DC-ABM defence contracts below its Byzantine breakdown threshold and that the induced adversary payoff is convex against that threshold, so the emergent attack drives a subset of partitions to breakdown in priority order. We further prove that this attack concentrates on the least substitutable partitions, that severity diverges as the induced corrupted-mass fraction approaches breakdown, and that operators can shrink the attack surface at deployment time along a closed-form connectivity-substitutability exchange rate.

Small-object detection in UAV imagery is challenged by weak visual evidence, ambiguous boundaries, dense object distributions, and complex backgrounds. Effective detection therefore requires long-range contextual information for target-background discrimination while preserving explicit local two-dimensional structures for accurate localization. These requirements arise at different stages of the detection pipeline and are not naturally addressed by a uniform feature-processing strategy. We propose Hybrid Dual-domain Mamba-YOLO (HDMamba-YOLO), a stage-wise heterogeneous SSM-CNN detector organized according to a perception-reconstruction-alignment-interaction rationale. EfficientVMamba-based EVSS establishes long-range contextual perception in the backbone, while PhasePatchMerging2D provides phase-aware hierarchical transitions. DST-Wrapper and Native C3k2-ASSAF then perform perception-to-reconstruction transition and repeated local two-dimensional reconstruction during FPN/PAN aggregation. DySample provides content-adaptive cross-scale resampling, while OS-CVTIA introduces macro-micro interaction and task-specific modulation for localization and classification. On VisDrone2019, HDMamba-YOLO-B achieves 42.737% mAP50 and 25.713% mAP50:95 with 10.042M parameters and 29.879 corrected GFLOPs. HDMamba-YOLO-Lite achieves 41.140% mAP50 and 24.741% mAP50:95 with 5.344M parameters. Under the unified AI-TOD evaluation protocol, HDMamba-YOLO-B obtains 21.621% AP and 47.881% AP50. Controlled ablations further support the stage-wise allocation of state-space perception, convolutional reconstruction, dynamic alignment, and task interaction for UAV small-object detection.

We study the online deployment of mobile ad hoc networks in unknown orthogonal environments, formalized as the Partially Observable Cooperative Guard Art Gallery Problem. We give a full proof that CADENCE algorithms achieve full coverage while maintaining a connected visibility graph using at most $n/2 + h - 2$ agents in orthogonal worlds with $n$ corners and $h$ holes. We further evaluate deployment-order heuristics that reduce agent count and deployment time in practice.

clifford_qc is a Python research toolkit for constructing quantum models, comparing variational and subspace eigensolvers, and estimating measurement resources. A common sparse Pauli algebra connects Hamiltonians, density operators, gates, and candidate-selection observables; dedicated interfaces handle circuits, execution, and statistical estimates. A worked gradient example illustrates shared measurements, and a molecular workflow separates reusable preparation, solving, and reference validation. Matrix-element banks support virtual subspaces, exact Ritz residuals, and folded-spectrum roots. Process-isolated benchmarks using the legacy density reference show that packed coefficients and streaming reduce observed peak memory on the larger tested inputs at increased runtime. Measurement studies separate setting counts, shot requirements, and device cost: fully commuting readout certifies a fixed first-order Ritz functional with fewer shots, while anticommuting cliques help some fixed reference energies and not others. Entangling-gate time and connectivity can offset shot savings or make a protocol inadmissible under illustrative device models. Further studies expose subspace bias, indeterminate encoding comparisons, failed approximate-restriction screens, and high residual-measurement costs under the tested uniform allocations. A provider workflow, scoped cost ledger, portable evidence export, and independent checks connect calculations to their recorded assumptions. Generated tables, provenance records, and manuscript checks connect claims to inputs. These small-system results support reproducible method development and establish neither hardware performance nor a general simulation speedup.

Agent runtimes that plan before acting run every step as soon as it is ready. Neither these runtimes nor their benchmarks distinguish a step that is ready from a step the request needs, and the difference is large: across 303 requests on 154 production workflows, an eager runtime that stops once the goal is met spends 62.2% of its CPU on work the request never needed. LazyAgent is, to our knowledge, the first complete formulation of demand-driven execution for agent runtimes. Its rule: run a ready step only if the current request needs it, and keep that answer exact as the plan changes. A demanded set derived from the program decides, in one traversal with no model calls, so LazyAgent spends nothing outside the request. Because demand is defined exactly, its behavior is proved: builders know before any run where demand must save and where it must tie. The bill is known before a plan runs, so LazyAgent can act as a query optimizer to choose among candidate workflows. Measured live against that eager runtime, LazyAgent saves 51.7% and 52.1% of container time on release gates where a frontier LLM writes the patches, 12.8% of tokens on DataSciBench plans at matched quality, and 24.0% of dollars inside a live LangGraph application, with no losses. When the LLM re-plans after real code failures, it solves every solvable goal with 11.7% fewer tokens per goal. On the production workflows it saves 42.0% of CPU, ties exactly where a request needs the whole graph, and adds at most 0.43% overhead. The same set steers every physical operation: projection saves 80-89% of a shared CI step, where the identical eager control saves 0.0%. Public agent benchmarks are designed the same way as eager runtimes, so they cannot show this difference. We compose our test programs from their tasks and call on industry to build benchmarks that measure how efficiently agent workflows run.

A quantum inspired framework was developed for polarized radiative transfer in the atmosphere to jointly model multiple scattering, absorption, and emission processes. The radiative field is represented by the Stokes vector and discretized using an n-directional two-dimensional spatial lattice on 3-dimensional uniformly distributed zenith and azimuth angular vectors. The formulation of this quantum computation system is based on the Lattice Boltzmann Method, which incorporates a physics-based polarized phase matrix. A classical lattice solver employing identical spatial and angular discretization and physical modeling as in quantum algorithm is implemented. The corresponding quantum algorithm is evaluated using the IBM Qiskit quantum simulator, enabling direct comparison of numerical accuracy to a classical simulation reference. The results demonstrate that fully physics-based polarized radiative transfer can be embedded within a quantum lattice framework, which provides an initial benchmark toward scalable quantum atmospheric radiative transfer and future quantum climate modeling

Agentic retrieval-augmented generation (RAG) enables language models to adapt retrieval based on previously retrieved evidence, but it remains unclear whether such adaptive orchestration consistently outperforms well-designed static pipelines. We conduct a controlled comparison of agentic and static RAG for Taiwanese historical question answering, sharing the same generator and hybrid retrieval backend. Despite similar aggregate performance, the two pipelines differ on 70.83% of questions, with their advantages largely canceling out when averaged. An oracle that selects the better response per question improves the composite score by 0.2417 over the better individual pipeline, revealing substantial headroom for question-level selection. We therefore introduce a post-hoc selector that compares the two responses and their cited evidence, significantly outperforming either individual pipeline and recovering 60.34% of the oracle headroom. These results show that aggregate comparisons can obscure meaningful question-level differences between retrieval strategies, suggesting that exploiting their complementarity may be more fruitful than seeking a universally superior pipeline.