Papers

Reliable decision-making requires more than accurate prediction: a model must preserve its beliefs, apply the relevant utilities, and recognize when the information needed to justify an action is missing. We study whether language models can learn this decision procedure from supervised fine-tuning and generalize it across domains and differing natural-language expressions of the decision challenge. Across 20 datasets, we explore challenges of belief instability and decision-making errors by first eliciting probabilities of outcomes and then varying only the utilities and the framing of the decision problems, while holding the evidence fixed. We train models to preserve elicited beliefs while selecting the action that maximizes expected utility, and evaluate transfer to unseen application domains, held-out framings, and different classes of payoff structures. We further introduce incomplete-information settings in which required utilities are withheld and replaced with irrelevant text, testing whether models can distinguish missing decision-relevant information from merely additional context. We find that targeted fine-tuning substantially improves coherent decision-making and that in many situations, learning transfers across domains and framings to situations unobserved during training. Further, models trained for decidability learn to identify when action cannot be justified based on missing information. Finally, we show the value of a routed system that considers separately the recognition of decision completeness and utility-sensitive decision execution.

Task-oriented conversational agents remain fragile under real world conversation scenarios as they rarely follow a predictable script, especially when users exhibit non-cooperative behavior. Existing function-calling benchmarks often emphasize successful, cooperative interactions and underrepresent adversarial conversation trajectories, thereby limiting the training resources available for developing robust agents. We present ToolRACER, a synthetic data generation pipeline that coordinates user, assistant and tool emulation models to generate and validated multi-turn interactions between a user and an agent. Using ToolRACER, we construct ToolRACERBench a robust multi-turn conversation benchmark spanning six domains, ranging over 55 varied personas, generating a validated corpus of 5.6K conversation trajectories, with approximately 66% of conversations containing failure-prone conversation scenarios. We inject adversarial behaviors, producing validated conversational interaction trajectories that capture realistic, robust scenarios. We evaluate models trained on ToolRACERBench against internal benchmarks, as well as on function calling benchmarks such as $\tau^2$-bench, BFCLv3 and ACEBench to evaluate agentic accuracy and robustness. Models trained on ToolRACERBench improve end to end agentic accuracy across $\tau^2$-bench and ACEBench, demonstrating significant gains when mixed with in-domain dataset in small language models for agent capability tasks.

It is shown that the mapping class group of any non-sporadic surface admits a continuum of equivariant coarse median structures, thus improving a previous theorem of the author. The same holds for a wide class of right-angled Artin and Coxeter groups, including examples of any dimension at least two, and RACGs which are not quasi-isometric to RAAGs; the coarse median structures in these cases are not induced by cocompact cubulations. The results are instances of a general procedure, which uses quasicocycles to flexibly modify the structure of a (suitable) hierarchically hyperbolic group.

The computational cost of Whittle maximum-likelihood parametric spectrum estimation (PSE) for short-dwell weather-radar Doppler moment retrieval is addressed. PSE can reduce finite-sample bias, especially for short coherent processing intervals and broad spectra, but repeated likelihood evaluations make direct use expensive. A fast implementation of the finite-sample Gaussian Doppler model is introduced by writing the likelihood in terms of closed-form autocorrelation lags and synthesizing the model spectrum and analytic derivatives by fast Fourier transforms (FFTs) on the same N-point grid as the periodogram. The estimator is extended to jointly estimate total power, normalized mean Doppler frequency, and normalized spectral width. Monte Carlo simulations with 1000 noise realizations show that the fast PSE keeps the accuracy advantage of the original PSE, while objective-level timing confirms the reduction from dense lag-to-frequency synthesis to FFT-based evaluation.

Solar inertial modes are increasingly viewed as probes of the solar interior, with recent helioseismic evidence indicating that these modes may be sensitive to the internal magnetic field. We investigate the role of magnetic fields on inertial mode excitation and dynamics using global nonlinear MHD simulations with differential rotation forced toward a helioseismically derived profile. Initial poloidal fields of varying strengths and topologies self-consistently generate different toroidal field configurations through the $\Omega$-effect. In all cases, baroclinic instability excites high-latitude inertial modes, with weak fields enhancing its growth rate above the hydrodynamic value. For stronger fields, the response depends on topology and depth. As a consequence, the polar vortex emerges earlier and modifies its relative amplitude, handedness, and hemispheric asymmetry. Only magnetic topologies exhibiting toroidal bands at the upper convection zone develop a second, magneto-shear ("edge") instability at the equatorward edges of these bands. Powered mainly by the toroidal field, this instability sets in earlier and grows faster for stronger initial fields. It excites equatorial modes absent in the hydrodynamic simulation, with fast magneto-Rossby signatures approaching the classical Rossby dispersion relation at tachocline depths. Alfvénic signatures and tentative slow magneto-Rossby branches also appear. Inertial-mode excitation and frequencies thus depend on magnetic topology and spatial distribution, not merely field strength.

As autonomous coding agents get increasingly deployed, the risk that accidental or adversarially injected misspecifications in tasks lead to dangerous agent behavior is critical to address. Prior work has shown that agents given such tasks rarely flag the conflict and instead cheat, editing tests or hard-coding expected outputs, and the actions taken to cheat can cause real damage, such as deleting a security defense to make a corrupted test pass. It remains unclear whether such conflicts can be established with independently verifiable evidence before the agent acts. We present SpecGuard, which detects and formally certifies these conflicts between task intent and tests. Given only the task description and codebase, SpecGuard autoformalizes the intended behaviour into a Lean 4 specification. The tests are formalized independently, and the Lean kernel checks whether any implementation could satisfy both formalizations, producing a machine-checked certificate when none can. On conflicted SWE-bench tasks, SpecGuard detects up to 72.8% of conflicts and formally certifies up to 51.1%, with a nearly five-fold lower conflict miss rate than model-based judgment. SpecGuard provides a pre-execution safety check that identifies reward-hacking opportunities through formal certification of task-level conflicts, before any agent behavior is observed. Our code is available at https://github.com/prmbiy/specguard.

Pentamode lattices are architected materials characterized by an elastic response dominated by a single deformation mode. The low stiffness of the other deformation modes, however, may result in limited stability under compressive prestress, thereby limiting their mechanical performance. In this work, we develop a general analytical framework to determine the buckling limits of prestressed 2D and 3D isotropic pentamode lattices with generic link geometries. Two distinct instability mechanisms are considered: a global instability associated with the loss of stability of the effective medium, and a local instability associated with finite-wavelength modes at the scale of the lattice cell. Closed-form expressions for the corresponding critical pressures are derived, and the buckling limit is obtained as the lower of the two. The analytical predictions are validated through numerical simulations of different PM lattices, showing good agreement over the configurations considered. The results reveal a trade-off between approaching the pentamode limit and maintaining sufficient buckling resistance, but also show that appropriate link designs can substantially increase the buckling limit while retaining the desired pentamode behavior.

Huffman coding is one of the oldest and most fundamental problems in computer science. Given a string of length $\TextLength$ over a general alphabet, the goal is to assign a binary codeword to each character so that no codeword is a prefix of another and the total encoded length of the string is minimized. Huffman coding is widely used in practical compression systems, including file compression. As modern datasets continue to grow, it is natural to study whether a Huffman code can be constructed efficiently in the massively parallel computation (\MPC) model. The celebrated Huffman coding algorithm admits two straightforward \MPC implementations: for any constant $\epsilon\in(0,1)$, one uses $O(\TextLength^\epsilon)$ memory per machine but requires $\Theta(\log \TextLength)$ rounds, while the other runs in $O(1)$ rounds but requires $\Theta(\sqrt{\TextLength})$ memory per machine. We give the first nontrivial \MPC algorithm for Huffman coding that bypasses both limitations. For every constant $\epsilon>0$, our algorithm uses $O_\epsilon(\log\log \TextLength)$ rounds and $\softO(\TextLength^\epsilon)$ memory per machine, while its total memory and total computation are $\softO(\TextLength)$. This provides a rare example in which an exact solution to a problem whose classical algorithm is sequential in nature can be obtained in a sublogarithmic number of \MPC rounds.

Processing-in-Memory (PIM) on monolithic 3D (M3D) DRAM is a promising answer to the memory wall for data-intensive dynamic programming (DP), yet extracting its performance today demands hand-written kernels: the programmer must pick a tile size, place data across non-uniform-latency memory tiers, partition work between heterogeneous processing units, and insert the right broadcasts and queues by hand. We present GenMLIR, an MLIR compiler that automates this mapping for tier-aware 3D PIM. GenMLIR encodes the semi-ring generalized grid update, the algebraic form shared by all-pairs shortest path (APSP) and sequence alignment, as a first-class IR abstraction, and lowers it through four GenDRAM-aware pass groups that derive blocked tiling, tier-aware placement, tile-to-PU assignment, and explicit communication. We also characterize precisely which DP recurrences the abstraction admits and which it does not. On the GenDRAM architecture, GenMLIR-generated code runs up to 5.8 (APSP) and 16.7 (alignment) faster than a lowering without compiler support, and 1.2-1.4 faster than a standard affine-tiling PIM compiler we also implement, reaching 90-100% of an achievable-performance bound. It expresses these workloads in 4-9 lines instead of 300-500 and compiles in a negligible fraction of runtime.

While generative AI models can produce visually faithful artwork, they often fall short in conveying emotional authenticity–a key driver of human expression. In visual storytelling, particularly comic storyboarding, this gap becomes pronounced: effective storyboards require both technical knowledge (e.g., anatomical accuracy or scenic composition) and emotional insight (from lived experience). We explore how human-AI collaboration can support emotion-driven creativity, where the user's feelings guide generation and emotional resonance is the goal. We present EmoToon, a technology probe that helps non-professional artists generate sketch-like storyboards and iterate on visual ideas. In a controlled study with N=25 participants, we find that AI assistance significantly improves emotional expression, aesthetic quality, and exploration, but reduces users' creative ownership. Our findings offer broader insights for human-AI co-creation in the domain of comic storytelling, emphasizing balance between output quality and user freedom, and raising new challenges for image generation models in this domain.

The Double-Partition Coloring Problem (DPCP) is a recently introduced generalization of several coloring problems on graphs and hypergraphs, including the Partition Coloring Problem, the List Coloring Problem, and the Conflict-Free Coloring Problem (CFCP). In this work, we further investigate the structural and computational properties of the DPCP and develop exact and heuristic solution approaches. We derive structural results that allow the detection of infeasible instances and the reduction of their size. We propose one-step and two-step heuristics for obtaining high-quality solutions. These procedures are incorporated into an enhanced branch-and-price algorithm based on a set covering formulation of the DPCP. The algorithm further includes new strategies for solving the NP-hard pricing problem. Computational experiments show that the proposed branch-and-price algorithm and a compact integer programming formulation complement each other: while the compact formulation is generally more effective on sparse instances, branch-and-price performs particularly well on denser and larger instances. In particular, these are the first exact algorithms to be computationally evaluated on various CFCP instances.

Contemporary data in many scientific domains, such as images, media streams, biomedical omics, are naturally modeled as probability distributions in Wasserstein space instead of points in Euclidean space. Consequently, clustering methods for distributions are in high demand, as they provide exploratory frameworks to group objects with similarity. Clustering distribution-valued data in Wasserstein space often requires choosing the number of clusters in advance. We propose Relative Wasserstein Spatial Depth (RWSD) as a selection criterion for optimal number of clusters. It compares the depth of each distribution in its assigned cluster with its depth in competing clusters. We pair this criterion with Wasserstein K-means or a proposed K-medians algorithm. Under suitable assumptions, we establish consistency of the cluster centers and of the selected optimal number of clusters. We also give an error rate under two-stage sampling and characterize the robustness of RWSD under any fixed partition. In extensive simulations, RWSD selects the intended number of clusters under unequal separations, heterogeneous dispersion, heavy tails, and outliers, where silhouette score or the Davies-Bouldin Index often fails. Applications to MNIST images and flow cytometry data illustrate the practical value of RWSD. In some cases where a cluster is not geodesically convex and lacks a meaningful depth center, the RWSD can fail.

Multilingual LLM benchmarks omit Slovak, a morphologically rich West Slavic language of five million speakers, or cover it only by machine translation. We present sk-bench, a native-first Slovak benchmark with 30 datasets (33 scored task variants) across ten skill categories. Eleven resources are introduced or first packaged for generative-LLM evaluation, including IFEval-SK with Slovak-adapted instruction checkers and native Chiby/SKJ1 resources for Slovak grammar and morphology. We evaluate 55 open- and closed-weights models under one harness. The best open model trails proprietary APIs by 12.6 points. Model rankings are similar for native and translated closed-form data ($\rho\geq0.98$), though translation separates the strongest models less well. By contrast, human-authored and LLM-generated QA questions rank models differently ($\rho=0.72$). For Qwen3-14B, continued Slovak pretraining lowers the overall score by 13.9 points. A small instruction set restores three quarters of that loss. Test-time reasoning improves scores by 8.5 to 12.5 points for models of 9B and above. Together, these findings suggest four design lessons for other under-resourced languages: use native data where translation fails, plan instruction repair after language adaptation, enable test-time reasoning before scaling up, and avoid overinvesting in target-language prompts. We release the data and code at https://github.com/slovak-nlp/sk-bench

In this work, we present an amplitude extrapolation ansatz for geometry optimization that uses calculations from previous steps to construct an initial guess for the quasi-Newton solver at each new geometry, significantly accelerating convergence of the coupled-cluster and Lambda solver alike. The ansatz incorporates a necessary amplitude transformation to mitigate discontinuities associated with orbital energy crossings at the Hartree-Fock level of theory, while remaining computational and memory efficient. The method is tested for a diverse set of medium-sized molecules to demonstrate a systematic and significant reduction in the number of coupled-cluster iterations compared to the conventional method.

Objective: Data-driven models that estimate physiological states, particularly biological joint moments, are widely used in exoskeleton control. However, these estimators are often coupled to device-specific sensor configurations, limiting controller transfer and the use of open-source biomechanics datasets. Methods: We proposed joint kinematics as an intermediate representation that decouples hardware-specific sensing from downstream biological joint moment estimation. A joint-moment estimator using joint angles and angular velocities was trained exclusively on open-source biomechanics data and evaluated using kinematics from a hip exoskeleton, knee exoskeleton, and inertial measurement unit (IMU) sensor suite during level-ground, ramp-ascent, and ramp-descent walking. Results: The estimator achieved an root mean square error (RMSE) of 0.17 Nm/kg and coefficient of determination (R2) of 0.79 using hip exoskeleton kinematics, 0.19 Nm/kg and 0.62 using knee exoskeleton kinematics, and 0.15 Nm/kg and 0.85 using kinematics derived from IMUs across bilateral hip, knee, and ankle joints. Conclusion: Joint kinematics enabled an estimator trained only on open-source data to operate across distinct wearable platforms. Significance: This framework may reduce target-device data collection and support transferable biological joint moment estimation for exoskeleton control and wearable biomechanics.

Collective molecular excitations govern energy transport in light-harvesting and optoelectronic materials. Theory predicts that excitonic band topology can generate edge-confined states, offering a route to direct energy flow in molecular systems, yet experimental evidence is lacking. Here we use scanning tunnelling microscopy-induced luminescence to image Frenkel excitons in two-dimensional zinc phthalocyanine molecular crystals with nanometer resolution. By varying cluster size, we observe the transition from discrete quantum-confined excitations to collective excitonic bands and directly resolve their real-space wavefunctions. We identify an excitonic state localized at the crystal perimeter and show, through comparison with microscopic theory, that it originates from the momentum-resolved non-trivial topology of the bulk excitonic bands. Our results establish molecular crystals as a platform for engineering topological excitonic states and provide a strategy for controlling energy transport through excitonic band engineering.

In molecular crystals, molecular excitations can hybridize into collective Frenkel excitons. Here we show that doubly degenerate excitations in planar two-dimensional molecular crystals generate excitonic bands with nontrivial momentum-resolved topological characteristics and edge states appearing only for selected crystallographic terminations. Using a model inspired by aggregates of zinc phthalocyanine molecules deposited on insulating surfaces, we demonstrate that the existence of edge excitons depends sensitively on the crystallographic orientation of the boundary and is governed by the Zak phase of the underlying bulk bands. The resulting edge states exhibit momentum-locked elliptical polarization textures originating from the internal structure of the molecular excitations, potentially enabling directional optical coupling and excitation transport. Our results establish a route toward topological and polarization-textured excitonic states in two-dimensional molecular materials.

In the Wilson formulation of lattice gauge theory, the fundamental degrees of freedom are group-valued link variables, while the action is a sum over the trace of the plaquettes, the smallest Wilson loops. For generative sampling methods such as normalizing flows, this poses a challenge: the distribution of an individual plaquette is easy to model, but mapping sampled plaquettes to the links is the obstruction. We explore a statistical way around this in two dimensions for the $\mathrm{SU}(N)$ gauge theory with the Wilson plaquette action. The action can be written in terms of holonomy variables, which determine the links once a consistency condition at the four corners of an extended lattice is satisfied. Using a boundary condition that only affects the holonomies, we trade this consistency condition for a conditional sampling problem, which reduces to sampling $X,Y\in\mathrm{SU}(N)$ given the group commutator $Z=XYX^\dagger Y^\dagger$, where $Z\in\mathrm{SU}(N)$ depends on the four corners. Plaquettes are sampled independently, and the resulting configurations carry weights due to the additional conditional sampling. We obtain the density of $Z$, which determines these weights, in closed form for $\mathrm{SU}(2)$ and $\mathrm{SU}(3)$. A normalizing flow models the single-plaquette distribution for $\mathrm{SU}(2)$ and $\mathrm{SU}(3)$; for the group-commutator problem we use a closed-form sampler for $\mathrm{SU}(2)$ and a trained normalizing flow for $\mathrm{SU}(3)$. In our tests on $2\times2$ and $32\times32$ lattices at two couplings each, per-plaquette acceptance rates exceed $99\%$ and the effective sample size of the final weights is moderate, typically above one half.

Large language models (LLMs) and vision-language models (VLMs) are usually frozen after deployment, so they do not learn from the cases they solve. This is especially concerning in medicine, where new clinical evidence, updated guidelines, and new therapies can change established practice. Fine-tuning can update the model, but it requires access to model weights and additional training. Parameter-free methods avoid training, but they may overfit a fixed validation set, lack reliable domain knowledge, or lose visual details by saving experience only as text. To address these limitations, we present a model-agnostic framework that allows frozen LLMs and VLMs to learn from deployment experience through three forms of external expertise: a Skill that guides reasoning and tool use, a Knowledge Memory that stores reliable facts supported by earlier cases or trusted external evidence, and a Multimodal Knowledge Base that keeps visual examples and guides the model to relate each retrieved case to the current image. Instead of relying on a fixed validation set, a validation strategy keeps an update only if it helps on new cases without degrading performance on earlier ones. Across six benchmarks covering clinical diagnosis, clinical workflows, medical reasoning, and medical and non-medical visual reasoning, and with four open-weight and closed-source base models, our framework improves performance during online deployment by up to 34.2% over the base model on medical tasks, generalizes to unseen cases, transfers to other models without further optimization, and works in non-medical domains.

We revisit a standard accepted practice in the continuous diffusion language model literature of fixing conditioning prompt tokens clean during training. We make a very simple modification: also noise the conditioning prompt tokens during training. We demonstrate that under this modified training objective, we achieve better generalization in combinatorial reasoning tasks such as Sudoku and N-Queens, with the largest gains on harder variants ($3.73\% \to 24.65\%$ solve rate on Sudoku Hard), and increased diversity of generated solutions ($50.60\% \to 73.79\%$ coverage on 10x10 N-Queens). We also show measurable improvements to natural language generation quality in modest dataset regimes with Gigaword summarization, but notably demonstrate that gains do not transfer to all natural language tasks (e.g open ended dialogue generation). Our method is a single line change to the training objective, requires no additional inference costs by default, and provides the flexibility of classifier-free guidance inspired guided sampling. Our \href{https://github.com/LateralIntelligence/noise-your-prompt}{code} is publicly available.

Vision-language-action (VLA) policies typically feed dense visual patch tokens into a language-action backbone, preserving scene context but offering no explicit mechanism to regulate how strongly different visual tokens influence policy computation. We introduce DIVA, a Dual-Space Intent-Aware Visual Attenuation module with an anchor-then-attenuate design. DIVA combines high-level task intent with low-level visual evidence to estimate patch-wise relevance anchors, then applies them in two complementary spaces: it reweights projected visual tokens before backbone entry and persistently attenuates low-relevance visual states within the backbone. DIVA preserves the full visual token sequence and requires no external grounding supervision. On LIBERO, DIVA improves OpenVLA-OFT from 96.6% to 98.0% average success and raises its zero-shot LIBERO-Plus score from 69.6 to 72.6. Real-world experiments further show consistent gains under task-irrelevant visual perturbations, supporting the robustness of intent-aware visual attenuation beyond simulation.

We analyze spacetime shortcuts generated by longitudinal contraction and transverse expansion of the intrinsic spatial metric at fixed reference volume. Smooth bounded ultrastatic corridors are globally hyperbolic and geodesically complete and produce an exact travel-time advance between exterior clocks. An axial maximum makes the axial light ray globally minimizing; a strict maximum on an interior interval establishes Olum's endpoint-surface criterion for exactly matched corridors. Exact Einstein-source identities distinguish integrated Eulerian matter energy from asymptotic mass: infinite tubes have negative integrated energy, whereas finite corridors admit competing gradient contributions. A transverse Ricci identity proves local null-energy violation in every nontrivial localized finite corridor. The axial averaged null energy is strictly negative for localized axial-maximum profiles, and quantitative bounds relate its magnitude to the reference time advance and peak strain. We also obtain the complete finite-profile Ricci spectrum and the axial passenger tidal tensor. These results provide explicit causal and source constraints on a prepared geometry without specifying its physical formation or matter dynamics.

In reinforcement learning with verifiable rewards, a benchmark's verifier supplies the reward. Reward-hacking audits mostly look for runs that pass without doing the task. But a run can do the task, cause a side effect, and pass a verifier that checks only the requested change. Do RL reward verifiers notice such a run? We answer with mutation testing: hand-written edits to a correct run leave the work missing or wrong, or add an unrequested change. A pass is a false positive only if our check finds the end state wrong. Every false positive came from an unrequested change, none from missing or wrong work. WorkArena's form verifier, called once after submission, passed forms with and without an unrequested field in all 36 valid pairs. Called after every step, it passed 7 of 11 runs filling that field last, all text or reference fields. In AppWorld, repeating a run's last write changed nothing in 33 of 48 tasks, but in every task of two templates the tests passed a duplicate write. Three recent RL methods train on one; their reward formulas give the duplicate full reward. One count assertion per template rejected all six duplicates and passed all six correct runs.

We present a hybrid quantum-classical framework that detects affinity and romance-investment fraud by modelling the cognitive biases in a manipulative conversation. In our proposed framework, cognitive biases central to this fraud class are carried by dedicated qubits in a structured parameterized quantum circuit, together with a frame qubit makes the encoding sensitive to the temporal order of manipulative reframing, and a narrative qubit that aggregates co-occurrence through a trainable entanglement layer. The circuit parameters are trained jointly with a classical reinforcement-learning agent that decides, turn by turn, whether to flag the conversation, modeled as an optimal stopping problem. We evaluate the model's performance on synthetic conversations that include hard negatives, legitimate but urgent, and legitimate but pushy sales conversations.

The following article attempts to define a new digital signature scheme based on a variant of the MinRank problem under appropriate parameters. First, the key generation process will be defined, followed by the message signing and verification processes. The speed of signing and verification will be analyzed, and finally, the signature will be evaluated in terms of security and robustness against known attacks, which will lead to a set of requirements that we will consider when selecting certain parameters related to the size and number of variables. It should be noted that this signature, although simple in design, can serve as the foundation for an entire family of digital signatures based on the central idea that inspired its creation.

Language models are more and more often asked for structured output: JSON that follows a schema, or a tool call with typed arguments. A small machine, an automaton, enforces the format by forbidding the tokens that would break it. We observe that this machine has a rare property: from any of its states, each token leads along exactly one path. Graphs in which only a few paths join any two points are a classical object of complexity theory, and our theoretical result settles an open question about them: one can decide whether such a graph connects two points while verifying that it really has few paths, with very little memory. Precisely, the problem lies in the classes ReachUL, LOGDCFL, C=L and SC2, and needs only O(log2 n/ log log n) space, below the classical O(log2 n) of Savitch's theorem. The constructions behind the proofs become an inference engine: text the format forces is written without running the model, the mask is recomputed on the GPU without any table, recursive formats use a small stack, every output stays valid under a token limit, and independent fields are decoded in parallel and verified. On one 16 GB Apple M2 Pro with Qwen3.5-2B and 4B, against MLX with llguidance, the standard setup for this hardware, schema-constrained extraction finishes 1.2- 1.3x sooner with the same answers, a grammar costs 3 MB instead of up to 1.5 GB, one server holds sixteen grammars where tables run out of memory, and sixteen tool-calling agents finish 2.5x sooner.

Autonomous and agentic clients can distribute reconnaissance across many identities so that each request remains valid, low-rate, and benign-looking while the population collectively acquires broad system knowledge. We formalize this threat as Distributed Collective Reconnaissance (DCR) and present SwarmReconGuard, a reproducible black-box benchmark in which the defender observes only service-boundary telemetry. The Docker-isolated study evaluates 11 benign and attack behaviors across 10-10,000 virtual identities, comprising 440 test runs and 3,666,300 requests, with complete telemetry integrity. We compare semantic, Gaussian, conditional, graph, kernel, hybrid, and CUSUM-based detectors. Gaussian likelihood-ratio detection achieves 100$\%$ detection with 0$\%$ observed false positives on known attacks but only 3$\%$ on unseen policies. CUSUM yields 36.1$\%$ overall detection at 1.25$\%$ false positives, while hybrid CUSUM reaches 85.7$\%$ detection with 0$\%$ observed false positives at 10,000 identities. Results expose a major policy-generalization gap and motivate exposure-aware, scale-aware defenses.

In the context of locally convex vector valued integrable tensors on open sets of a $n$-dimensional manifold $M$ developped in our previous work \cite{sil09261621}, in this paper we construct a framework in which to address both a variational problem associated with a Lagrangian $\mathscr{L}$, and a conservation law on-shell associated with actions $\phi$ and $\omega$. $\mathscr{L}$ is a $C^{p}$-map valued in a quasi-complete Hausdorff lcs $\mathcal{Z}$ and defined on an open set $O$ of a normed space, $\phi$ is a $C^{p}$-action parametrized by an open $A$ of a normed space $T$, acting over the space $\mathfrak{Z}^{O}$ of fields valued in a normed space $Y_{0}$ and defined on $W$ domain of a chart $\xi$ of $M$; $\omega$ is a $C^{p}$-action on $W$ parametrized by $A$. Via a result in \cite{sil09261621}, we construct $\mathfrak{A}:\mathfrak{Z}^{O}\times A\to\mathcal{M}(W,\mathcal{Z})$ where $\mathfrak{A}(u,\eta)$ is the action functional associated with $\mathscr{L}$, $u$ and $\eta$. Here $\mathcal{M}(W,\mathcal{Z})=\mathcal{L}_{s}(\mathcal{H}(W),\mathcal{Z})$ is the lcs of measures on $W$ and valued in $\mathcal{Z}$. By letting $\mathcal{A}(u):A\to\mathcal{M}(W,\mathcal{Z})$ be such that $\mathcal{A}(u)(\eta)=\mathfrak{A}(u,\eta)$, we state that $\mathcal{A}(u)$ is differentiable off-shell, find its differential map off and on-shell. We establish that if the differential map of $\mathcal{A}(u)$ on $0$ is $0$, then $\mathscr{N}[u]$ is conserved on-shell, and if $p=\infty$, then the divergence of $\mathscr{N}[u]$ equals $0$, where the divergence of a lcs valued vector field is defined and analyzed in \cite{sil09261621}. Here $\mathscr{N}[u]$ is a $\mathcal{L}_{b}(T,\mathcal{Z})$-valued $C^{(p-1)}$-vector field of $M$ on $U=\xi(W)$.

Learning a directed acyclic graph (DAG) from observational data is a challenging combinatorial problem due to the exponential growth in the number of candidate parent-set configurations. Existing exact score-based methods often require computationally intensive combinatorial search, whereas constraint-based methods can become unreliable or computationally demanding as graph size and conditioning-set complexity increase. We develop a non-parametric hybrid framework, referred to as DECO (Directional Evidence-guided Configuration Optimization), that extracts dependency and directional evidence from observation data to construct admissible parent sets prior to exact optimization. It reduces the optimization search space by eliminating empirically unsupported parent configurations while preserving flexibility for all plausible edge orientations. Theoretical analysis establishes an exponential reduction in the admissible parent-set configuration space and quantifies how bounded edge-level omission affects the probability of retaining the true parent structure. Experiments on benchmark Bayesian networks and synthetic discrete and continuous DAGs demonstrate substantial search-space reduction while achieving competitive structure-recovery performance, with favorable structural Hamming distance across many evaluated settings. These results show that directional evidence can provide an effective preprocessing mechanism for reducing the computational burden of exact DAG learning without requiring a fixed parametric structural~model.

We study the homogeneous Kuramoto model on random 3-regular graphs. A connected graph $G$ is globally synchronizing if, for almost every initial condition, the gradient flow converges to a fully synchronized state, in which all phases coincide. It has been conjectured by Bandeira [Oberwolfach Rep. 21 (2024) 1163-1226] that a random 3-regular graph $G_{2n,3}$ is globally synchronizing with high probability. We disprove this conjecture in a strong way: $G_{2n,3}$ is not globally synchronizing with probability tending to 1.