Papers

Future wireless networks require learning agents to adapt across heterogeneous channel conditions, traffic patterns, quality-of-service (QoS) requirements, objectives, and operational constraints. Reusing decision knowledge across such tasks is challenging because conventional multi-task and transfer reinforcement learning methods primarily share or transfer policies, coupling transferable knowledge with task-dependent action mappings. This paper proposes FARM (Fundamental Agentic Reward Model for Multi-task Wireless Network Optimization), a reward-space transfer framework that shifts cross-task knowledge reuse from policy space to trajectory-level decision evaluation. FARM introduces an Agentic Reward Model (ARM) that learns a task-conditioned reward prior from heterogeneous source-task trajectories and provides auxiliary guidance for task-specific policy optimization. In Stage I, ARM jointly models task conditions, temporal trajectory dependencies, and objective-dependent reward structures while each source task retains its own controller. In Stage II, the learned reward prior is frozen and reused to guide the adaptation of a target-specific controller for previously unseen tasks, without transferring source-task policies. Experiments on heterogeneous multi-access edge computing (MEC) tasks show that FARM achieves a mean late-stage gain of 29.8% over Single-task SAC on unseen Rate-Latency targets, compared with 16.1% for CRA Transfer, and reaches a 46.2% gain on the moderate-OOD FAR-M case. Further analysis shows that both Mamba and Transformer trajectory encoders support Reward-Space Transfer, while Mamba provides improved robustness as longer history dependencies are introduced.

Video Object Segmentation (VOS) in complex and long videos is increasingly important for real-world applications, where target objects often appear only intermittently within long temporal horizons. However, existing benchmarks largely focus on temporally salient objects that remain visible for most of the video. To address this gap, we introduce FaVOS (A Benchmark for Video Object Segmentation with Fractional Temporal Visibility), a benchmark designed to evaluate VOS methods under low temporal visibility. We show that, in this regime, the standard J&F metric can collapse VOS evaluation into absence classification, because empty predictions receive high rewards on target-absent frames. Consequently, even a trivial empty-mask predictor can outperform strong models such as SAM 3, revealing a fundamental mismatch between current metrics and practical VOS performance. To mitigate this issue, we propose Volumetric J&F, which evaluates mask sequences as spatio-temporal volumes and reduces the dominance of target-absence rewards while preserving sensitivity to segmentation quality and temporal structure. Project page: https://aidaslab.github.io/FaVOS.

Using frontier agent harnesses to tackle mathematical research problems has emerged as an effective means of advancing mathematics. However, solving frontier problems in mathematics may require a massive number of agents working in parallel for extended periods to construct proofs, thereby generating an enormous volume of intermediate proof results. Organizing these intermediate results throughout a long-horizon proof-search process and reusing knowledge gained from prior explorations remain major challenges. We present Ansatz, a mathematical research agent built around Continual Graph Memory, a graph-based, evolvable, cross-problem mathematical research memory system that explicitly organizes the entire proof search process and reuses information from exploration trajectories of previous problems. Specifically, we develop a unified graph memory that represents all intermediate exploration results, including facts, plans, and counterexamples, together with edges that explicitly represent the relationships among them; dependency-aware retrieval supplies precisely targeted local context; an evidence-sensitive curator updates the research frontier and distills lessons from prior attempts; and scoped recall surfaces earlier statements and negative findings for local re-proving rather than uncritical reuse. Experiments cover runs across all ten First Proof Second Batch problems, together with four component studies. Ansatz reports closure on all ten research tasks, demonstrating its ability to sustain and resume long-horizon mathematical search. Beyond these problems, Ansatz also produces solutions to the Jamison caterpillar conjecture and Erdős Problems 289, 348, and 488 without human intervention, and makes partial progress on several open problems, illustrating its strong ability to solve open mathematical research problems.

Cross-fitting is routine in much of applied research. While conventional confidence intervals that ignore cross-fold dependence are asymptotically valid in several settings, they undercover in many applications that share a common form of nonregularity: from the classic cross-validation problem of testing whether a fitted model outperforms another, to testing for heterogeneous treatment effects with machine learning, to estimating the value of a potentially non-unique optimal treatment regime. Exploiting a new locality condition, I show that a large class of cross-fitting estimators still satisfies a central limit theorem despite the nonregularity, but with an asymptotic variance that must be adjusted for the cross-fold correlation. Then, I propose a method for estimating this correlation and construct new confidence intervals that attain asymptotically nominal coverage. Finally, I show that the proposed confidence intervals attain approximately nominal coverage in a simulation study with random forests and neural networks.

Multimodal 3D Human Pose Estimation (3D HPE) combines complementary information from RGB, LiDAR, and mmWave radar, but models trained on correlated observations from the same individuals, raise privacy risks overlooked by record level analysis. We present a unified framework for multimodal 3D HPE that couples kinematics-induced sensor fusion with subject level privacy auditing and private training. First, our multimodal model aligns modality specific joint representation, injects skeletal structure and adaptively aggregates complementary sensor evidence for accurate pose prediction. Second, we formulate a black-box subject membership inference attack for 3D HPE, complemented by an empirical pointwise maximal leakage analysis, which characterizes how individual attack score outcomes change inference about the membership outcome. Third, we instantiate user-level differential privacy via Action Temporal Stratification, a population weighted within-subject sampling strategy that enforces action and temporal coverage. We evaluate our framework on the MM-Fi dataset across three diverse experimental protocols. Source-code will be released upon acceptance.

An alchemical resource theory admits a nonfree state that can be exactly replicated by free operations and whose finitely many copies enable exact simulation of arbitrary quantum instruments. The only known nontrivial examples, imaginarity and parity asymmetry, exhaust the qubit case (arXiv:2609.12988) and exhibit an all-or-nothing property: any nonmaximal resource exactly simulates only free unitaries. We introduce aquaternionicity for even-dimensional qudits $d\ge4$, based on a quaternionic structure represented by an antiunitary $\Theta$ with $\Theta^2=-I$. It is extremal in two senses: every pure state is alchemical, and alchemicality is maximally abundant under mixing. Within this framework, for any nontrivial $d$-dimensional resource theory and pure alchemical state $g$, the admissible pure-mixing set satisfies $\dim\mathcal M_g\le 2d-4,$ which aquaternionicity saturates for every pure state. Conversely, these two extremal properties characterize the quaternionic structure. Because every pure state is alchemical, the initial resource can be chosen to match the known noise structure. Moreover, when the noisy resource states remain within a common isotropic subspace, the same free processor works exactly for all of them while returning the resource unchanged. This enables recalibration-free exact simulation, including universal quantum computation, and is realized for standard dephasing and dissipative noise, where aquaternionicity substantially outperforms imaginarity and parity asymmetry.

Real-time conversational agents require speaker diarization that streams and runs on a CPU. Most systems apply an utterance-level speaker encoder to short, heavily overlapping chunks, which wastes computation and leaves the model optimized for the wrong task. We instead turn a state-of-the-art speaker recognition architecture into a causal frame-level encoder that reads the stream once and emits one embedding every 80~ms from a bounded two-second window of past audio, and pair it with online clustering that gates every update on the self-similarity of the stream. Trained only by distillation from an utterance-level teacher on simulated and out-of-domain mixtures, and evaluated with one fixed set of hyperparameters, the system is the most accurate streaming diarizer on low-overlap benchmarks at sub-second latency, degrades far less than cache-based systems as the number of speakers grows, and runs five times faster than real time on a single CPU thread.

We characterize lower curvature bounds on metric surfaces by the Bishop–Gromov inequality for metric balls.

We give quantum algorithms for Hamiltonian simulation, linear differential equations, and linear systems with square-root dependence on sparsity. For an $s$-sparse Hermitian matrix with row and column Euclidean norms bounded by $\eta$, evolution for time $t$ to error $\epsilon$ uses $\mathcal{O}(\sqrt{s}\tau+\sqrt{s\tau\log(1/\epsilon)}+\log(1/\epsilon))$ queries, where $\tau=\eta|t|$. The same bound holds for time-dependent Hamiltonians with coherent time-indexed sparse access and certified time discretisation. Our construction combines an exact entrywise representation with Cayley transducers. A joint lower bound nearly matches the time, sparsity, and precision dependence, ruling out a uniformly additive bound in $\sqrt{s}\tau$ and $\log(1/\epsilon)$. For dissipative linear differential equations we obtain $\mathcal{O}(\bar g[\sqrt{s}(\tau+\sqrt{\tau})+1]\log(C\bar g/\epsilon))$ queries, where $\tau$ is the evolution time multiplied by a norm bound, $\bar g$ bounds the ratio of combined input and source norms to the final-solution norm, and $C$ is an absolute constant. We sharpen this bound by separating Hamiltonian and dissipative components, and treat general constant generators. Lower bounds capture joint parameter dependence for time-dependent equations and precision hardness for constant positive semidefinite dissipation. For linear systems, a Cayley walk and Chebyshev filter give $\mathcal{O}(\kappa\sqrt{s}\log(1/\epsilon))$ queries when $\|A\|\le1$ and $\|A^{-1}\|\le\kappa$, and we give an alternative proof of a matching joint lower bound. We also give gate-efficient implementations of time-independent Hamiltonian simulation and the dissipative ODE solver, including Lipschitz time dependence, retaining the query bounds with polylogarithmic overhead in inverse precision given efficient entry decoding, arithmetic, and required source preparations.

Many policies assign treatment according to whether a score crosses a cutoff. Regression discontinuity (RD) designs identify the effect of treatment assignment, but the threshold policy itself may also reshape individuals' incentives, inducing behavioral responses that affect outcomes even holding treatment status fixed—a channel that conventional RD designs cannot capture. We develop a framework that exploits randomized variation in policy thresholds, together with rank-invariance-type restrictions, to identify the treatment-assignment and incentive-response effects. We illustrate the empirical relevance of this distinction using data from a merit-based scholarship experiment in Malawi. In this illustration, the conventional RD estimate is positive, while the incentive-response effect is negative and more than twice as large in magnitude. These findings suggest that threshold policies may generate unintended adverse behavioral responses, potentially reflecting discouragement induced by demanding thresholds, and caution against relying solely on conventional RD estimates when evaluating threshold policies.

We study weighted bilinear Hardy-Steklov operators on the half-line for $1<p_1,p_2<\infty$ and $0<q<\infty$. The operator-theoretic core is an exact output-localisation principle. For a bounded bilinear map into $L^q$, compact middle truncations identify compactness with vanishing of a directed output-tail invariant and, with constant one, identify that invariant with the essential distance and a radius-based Hausdorff measure of noncompactness. Under the additional hypothesis of local finite-rank approximability, the same invariant also equals the finite-rank distance and governs the rank-approximation conclusions. We verify these hypotheses for the Hardy-Steklov map and then translate the tail invariant through the five weighted boundedness regions. In Regions $I,\mathrm{II}_1,\mathrm{II}_2,\mathrm{III}_1$, boundedness must be supplemented by lower- and upper-endpoint vanishing, and the essential distance is comparable to the larger endpoint defect. In the strict subcritical Region $\mathrm{III}_2$, boundedness alone implies compactness. This yields a sharp structural transition between the equality boundary $P=Q$ and the summable regime $P>Q$. The quasi-Banach range $0<q<1$ is treated throughout with the powered metric $d_q(f,g)=\|f-g\|_q^q$.

Timing errors represent a fundamental challenge in high-speed analog-to-digital converters (ADCs). This paper presents a frequency-interleaved architecture with inherent robustness to timing errors. The frequency interleaving is achieved by an embedded analog Fourier transform (AFT), allowing accurate channel spacing. The charge-based passive AFT is able to share the capacitor array with the capacitive DAC of the subsequent data conversion. A circuit implementation for the complex-valued irrational coefficients in the AFT matrix is proposed, which eliminates the device-ratio-induced systematic errors. Measured at 10b 13GS/s, the calibration-free ADC achieves 45.8-dB SNDR and 56.1-dB SFDR at Nyquist. The ADC consumes 26.4mA from a single 1V supply, resulting in a Walden FoM of 12.7 fJ/conv.-step.

In this paper, we investigate refined Bohr-type inequalities involving Schwarz mappings for holomorphic mappings defined on the unit ball \( B_X \) of a complex Banach space. We first establish refined Bohr inequalities for scalar-valued holomorphic mappings \( F : B_X \to \mathbb{U} \). We then extend some results to vector-valued holomorphic mappings involving Schwarz mappings of the form \( F : B_X \to \overline{\mathbb{U}}^{\,n} \), \( n \ge 2 \), where improved Bohr radii are obtained under suitable norm conditions. Furthermore, we study the Bohr radius involving Schwarz mappings for holomorphic mappings whose ranges are contained in the closure of the unit ball of a JB$^*$-triple. Our results generalize the results of Jia et al. \cite{JLP}, Liu et al. \cite{LLP2021} and Liu et al. \cite{LSX2018}.

We present a root isolation algorithm for an analytic function. The method combines interval subdivision with a heuristic test based on local cubic Hermite interpolation. Whenever the function is sufficiently close to its cubic Hermite interpolant at a few sample points, we assume that the critical points of the cubic interpolant give a valid root isolation for the original function. Such an algorithm cannot, in general, provide a rigorous certification of the computed intervals. Nevertheless, extensive numerical experiments show that it is highly efficient and almost always produces the correct results. We hope that a probabilistic or exact algorithm can be derived from this work.

We study the focusing inhomogeneous nonlinear Schrödinger equation with an inverse-square potential $$ i\partial_t u+\Delta u-a|x|^{-2}u+|x|^{-b}|u|^{\alpha}u=0, \qquad (t,x)\in\mathbb{R}\times\mathbb{R}^{N}, $$ in the mass-supercritical and energy-subcritical regime $$ \frac{4-2b}{N}<\alpha<\frac{4-2b}{N-2}, $$ where $N\ge3$ and $0<b<2$. Two main lines of research have developed largely independently for this model: scattering theory below the ground-state threshold and the variational analysis of solutions with prescribed mass. In this work, we establish a connection between these two approaches.

Compiling GUI procedures that agents execute repeatedly into programs can reduce their token costs. However, measuring payback and deciding when to compile have two challenges. First, compilation costs are uncertain because attempts can require repair and still fail to produce a usable program. Second, future reuse is unknown because tasks may stop arriving or GUI drift may stop the program from working. To address these challenges, we propose PACE (Payback-Aware Compilation from Experience), a system with a measurement protocol and an online compilation algorithm. The measurement protocol records successful and failed compilation costs, and compares agent and program execution costs on matched task inputs to estimate per-use savings and payback counts. Using these measurements, the online algorithm compares estimated future savings with compilation costs, including failed attempts, based on past task arrivals and compilation outcomes. It checks execution and compilation charges against a cumulative budget determined by observed task arrivals before allowing either action. Under stated action-cost assumptions, total cost after each arrival is at most $1+\epsilon$ times the cost of running every task with the agent. For successful compilation attempts, estimated payback counts excluding source agent runs are 2-16 uses. In simulations using recorded task arrivals, PACE reduces token costs by 17.3% compared with ReAct, 24.9% with the AutoRPA adaptation, and 17.3% with the ToolPro adaptation on average ($\epsilon=0.25$).

Random sampling in DNA sequencing produces repeated reads, increasing retrieval latency and sequencing cost. We study the coverage-depth problem for full-message recovery in distributed DNA storage under noiseless uniform sampling, where strands are partitioned among $M$ containers and one strand is independently sampled with replacement from each container per round. For arbitrary linear codes and ordered partitions, we derive exact formulas for the recovery-time distribution and expectation. We prove that MDS codes, whenever they exist, are optimal for every fixed partition, and establish a universal lower bound on the expected total read cost together with its equality conditions. For MDS codes, we identify container-size regimes that yield genuine savings in total reads and regimes that provide only parallelism without changing the asymptotic sequencing cost. For simplex codes, we prove that the $q$-ary simplex code is, up to isomorphism, the unique single-container minimizer among codes with the same parameters, resolving a recent conjecture by Bertuzzo, Ravagnani, and Yaakobi. We further construct a partition attaining the minimum total read cost and derive bounds for intermediate and balanced partitions. These results clarify when distributed sampling reduces latency alone and when it also reduces sequencing cost.

In directed logic (DL), electronically controlled optical elements serve as photonic gates. Such electro-optical elements are of mixed nature: they have two inputs–electronic and optical–but only one, optical, output. This makes cascading such gates without repeated conversion between optical and electronic representations cumbersome. This problem can be largely overcome using the nested cascading scheme proposed by Shamir and Hardy [Opt. Express, 17, 150 (2007)]. Although promising as a solution to the cascading problem, the Shamir-Hardy scheme has so far been proved or implemented only for a few simplest cases. Here, we develop a general rigorous theory of nested cascading, valid for an arbitrary number of gates. We propose a variant of the algorithm which easily extendable to large number of gates, and rigorously prove its validity. Furthermore, we analyze how nested cascading scales with the size of the corresponding Boolean formula. We show that good (linear) scalability is guaranteed in many important cases, while the average scaling with respect to the corresponding Boolean formulas is only moderately polynomial, with an exponent of approximately 3/2. Yet, the worst-case scaling remains exponential with respect to more general Boolean circuits allowing sharing and reuse of intermediate results.

We show that $\mathbb{Z}_2$ is a quantum normal subgroup of $SU_q(2)\times SU_q(2)$ and that the quotient group is isomorphic to $SO_q(4)$, thereby establishing that $SU_q(2)\times SU_q(2)$ is a double cover of $SO_q(4)$.

Invariant suites and runtime monitors increasingly gate agent deployment decisions, and the evidence offered for any particular suite is almost always a single observation: it passes an implementation believed correct and fails one believed broken. We measure what that observation is worth. Applying mutation analysis to an invariant suite for a multi-session agent state-projection layer, we first find that this standard validation certifies a suite in which a first-order mutant removing event-identity deduplication survives every check. We then freeze the repaired twelve-check suite, record its hash, and run it once against ten mutants specified by an adversarial reader who designed none of its fixtures: it kills five. Instrumenting the five survivors against the reference shows they fail in two distinct ways, not one. Three are never activated, because no fixture supplies an input on which the mutated code behaves differently at all. The other two corrupt internal state that no oracle in the suite can observe. The two modes need different repairs, and neither is visible from a pass/fail report. Treating the missing inputs as a coverage question, we enumerate seven discriminating dimensions of the input space, register in advance which are uncovered and which survivors they should explain, and add one fixture per uncovered dimension while reusing the existing oracles verbatim. All five survivors then die, each to the check written for its predicted dimension. We report this as a repair result on the same challenge set rather than a second held-out estimate, and give the artifact, including the frozen hash, the registered predictions, all mutants and the run logs, so the distinction is checkable.

An extension with temporal modulation is presented of the canonical 2 + 1-dimensional solitonic Dym equation as originally introduced by Konopelchenko and Dubrovsky. Application is made of admitted Painleve' II symmetry reduction to derive exact solution to a class of associated 2 + 1-dimensional nonlinear moving boundary problems of generalized Stefan-type.

Standard factual benchmarks assume clean monolingual prompts and exact-match scoring, two assumptions that break simultaneously in real-world multilingual deployment, from retrieval-augmented generation pipelines returning mixed-language passages to users pasting multilingual web content. We introduce Multilingual Distractor Interference (MDI), a lightweight and fully replicable evaluation protocol requiring no new data or annotation, in which factual questions are preceded by a semantically irrelevant foreign-language sentence, and evaluate five instruction-tuned LLMs across TruthfulQA and TriviaQA under eight distractor conditions (40,000 evaluations). Our central finding is a metric confound: for Llama-3.1-8B under a Hindi distractor, 58% of responses switch to Devanagari script, yielding a raw hallucination proxy of 0.710, but manual review reveals that 120 of 148 script-switched responses that were correct under clean conditions remain semantically correct despite being written in the wrong script, reducing the adjusted semantic hallucination rate to 0.470. All other models respond through abstention escalation with no hallucination increase. A paragraph-length English distractor triggers near-universal abstention (0.806-0.998) across all models, consistent with reading-comprehension confusion, a failure mode with direct consequences for multilingual RAG pipelines. TruthfulQA multiple-choice accuracy is unaffected under all single-sentence conditions. These results show that exact-match hallucination rates in mixed-language settings should be decomposed into script-switching and semantic error components before drawing conclusions about model reliability.

Safety monitors help safeguard language-model agents interacting with external tools and environments, but conservative monitoring can generate many false alarms, consuming extensive review resources and weakening trust in alerts. Because false and genuine alarms often remain interleaved in native monitor scores, obtaining a reliable cutoff still requires substantial manual verification. In practice, a small set of verified-safe non-alarmed trajectories may be available while alarms remain unlabeled, naturally casting false-alarm auditing as a positive-unlabeled (PU) ranking problem. The key challenge is monitor-induced selection, since observed safe references are accepted by the monitor, while the hidden safe alarms of interest are precisely those it incorrectly flags, making the observed positives poorly representative of the positives to be recovered. To address this challenge, we propose a two-stage framework in which Trust-aware PU Supervision adapts safe references toward the alarm domain and protects plausible false alarms from excessive negative pressure, while Reliability-gated Rank Distillation consolidates consistent ordering preferences from multiple PU reference models into a single student. Consensus-guided Structural Refinement then improves the student ranking using hierarchical safe-reference support, alarm relations, and predicted reference consensus. The framework requires no alarm safety labels for fitting and leaves the underlying monitor unchanged. Across mainstream safety monitors, our method achieves a macro AUPRC of $0.6444$, outperforming eight evaluated PU baselines by 5.27–16.98 absolute percentage points; compared with PULDA, the strongest evaluated PU baseline, it recovers 33.3% more false alarms at a 5% review budget.

Motif-scaffolding systems commonly follow a generate-then-filter paradigm, in which candidate proteins are generated independently and structural evaluation is used primarily for terminal screening or ranking. This paradigm underuses evaluation: failed predictions contain state-specific evidence about whether a design requires repair of motif geometry, global foldability, or other structural constraints. We introduce ELMS (Evidence-based LLM-guided Monte Carlo Search), an evaluator-in-the-loop search framework for motif scaffolding that turns such evaluator feedback into targeted design actions. Effective reuse of structural feedback is nontrivial because different scaffold states exhibit different failure modes, and repeatedly refining a single trajectory can prematurely commit computation to an unproductive region of sequence space. ELMS therefore retains evaluated scaffolds as persistent search states: a Critic Agent diagnoses state-local structural failures, a Policy Agent selects targeted operators with execution parameters, motif-locked operators realize legal sequence modifications, and MCTS determines which historical states should receive further design effort. Under the standard GeomMotif protocol (100 candidates per task), ELMS achieves Successful rates of 86.41% on single-motif tasks and 84.57% on paired-motif tasks, exceeding the strongest prior baseline by 19.3 and 21.9 percentage points, respectively. On MotifBench, under a matched 100-candidate search budget, it solves 26.7 of 30 tasks on average (88.89% Task Success), compared with 16.0 tasks (53.33%) for the strongest baseline. These results establish ELMS as an effective approach for converting structural evaluation from a terminal filter into actionable guidance for iterative motif scaffolding.

We establish local and global spectral bounds for Hermitian $A_\alpha$-matrices of mixed graphs. Using the first three spectral moments, we obtain an upper bound for the largest eigenvalue as the largest real zero of an explicit cubic polynomial. Vertexwise estimates for the extreme eigenvalues yield new lower bounds for the spectral spread, including a bound that strictly improves an existing degree-based estimate for ordinary graphs. A local two-dimensional compression produces a spread bound involving neighbour-degree data and gain-weighted triangles. This bound is exact for every mixed orientation of a star and is independent of a known Zagreb-index bound. We also derive two complementary upper bounds for sums of the smallest eigenvalues. As further consequences, the spectral estimates provide a computable convergence guarantee for Richardson graph filtering and a stability certificate for residual graph-neural-network layers on directed networks. Numerical examples illustrate the sharpness and mutual incomparability of the proposed bounds.

Joint-Embedding Predictive Architectures (JEPAs) are a promising paradigm for representation learning by predicting future latent states without reconstructing observations. Recent work has adapted JEPA-style latent prediction to offline zero-shot control through action-conditioned temporal prediction, enabling representations to capture long-horizon dynamics from fixed behavioral data. However, transition-level predictive objectives supervise actions in isolation and provide limited information about which actions are preferable when similar states admit different goal-conditioned outcomes. We introduce TwinJEPA, a framework for learning action-preferred predictive representations by augmenting JEPA-based control with offline-mined action-preference supervision. TwinJEPA identifies approximately matched states across offline trajectories and constructs preference pairs via goal-conditioned reward relabeling. It then learns two complementary objectives: reward-gap regression, which preserves the magnitude of outcome differences, and preference classification, which captures the relative ordering of alternative actions. Both objectives are used only during training and incur no additional inference-time cost. We evaluate TwinJEPA on long-horizon navigation and continuous-control benchmarks with state- and pixel-based observations. TwinJEPA yields positive benchmark-level mean differences across all matched state-based evaluations, while analyses across domains and observation modalities indicate that larger gains tend to arise when local action alternatives provide more informative outcome contrasts. The results suggest that local action-comparison supervision can complement temporal prediction by encouraging JEPA-based representations to retain both long-horizon temporal structure and fine-grained distinctions among locally observed actions for offline zero-shot control.

This paper investigates non-parametric regression estimation when the explanatory variable takes values in a separable Hilbert space and observations are sampled over an increasing regular spatial lattice. Under a rigorous field-to-field independence setup between covariates and errors, we explore the structural and asymptotic concentration properties of the functional $k$-nearest-neighbour ($k$-NN) estimator. By establishing a sharp pathwise deterministic sandwiching framework for the data-driven random bandwidth, we successfully decouple the local infinite-dimensional small-ball profile from the polynomial covariance decay of the neighborhood indicators. Pointwise convergence rates are derived across short-range, critical, and long-range spatial regimes, revealing a combined penalty term that reflects both covariate spatial interaction and response error memory. Furthermore, uniform consistency over compact subsets is established for unbounded sub-Gaussian error processes through metric entropy and indicator boundary shell chaining. Structured polynomial simulations confirm our theoretical rates and exemplify the precise mechanics of the spatial long-range bottleneck.

Agent harnesses play a critical role in defenses by enforcing safety constraints to prevent unsafe actions. However, rapidly emerging attacks outpace manual harness adaptation, motivating automated harness evolution. Yet the signals available for harness evolution are often sparse, such as brief descriptions or a few attack examples in threat reports and preprints. To address this limitation, we introduce HASTE, a multi-agent framework that evolves agent harnesses from sparse threat evidence through an adversarial interplay between safety-specification generation and attack-case generation. Safety specifications guide harness updates toward addressing identified safety vulnerabilities, while attack cases probe for remaining safety vulnerabilities after each update. By feeding evaluation outcomes back into both processes, HASTE enables harness evolution against emerging attacks beyond the initially observed evidence. Experimental results across multiple backbone models, attack types, and evidence forms show that HASTE consistently reduces attack success rates while preserving benign-task utility. The code is available at https://github.com/xxiqiao/HASTE.

This paper provides a complete characterization of the spectrum, point spectrum, continuous spectrum, and residual spectrum of bounded convolution operators acting on the non-separable space $\sigma_\infty$, the dual of the Hahn sequence space. Because standard duality and separability-based techniques fail in this setting, a linear functional approach is employed with the fine spectral decomposition of the unilateral shift operator. Furthermore, as a major application of this methodology, we establish a complete description of the fine spectrum for operator polynomials acting on general sequence spaces without requiring separability. This result provides a unified framework that subsumes classical spectral determinations for general banded triangular matrices across various sequence spaces and enables the explicit computation of fine spectrum on non-separable spaces.

Jazz pianists develop distinctive traits that experienced listeners can often identify within seconds, yet the features underlying this recognition resist formal description. We study jazz pianist style through the lens of a pretrained symbolic music transformer, showing that its learned representations already encode pianist identity well enough for highly accurate classification across two benchmarks. We then augment the transformer with cross-attention over learned pianist identity embeddings, enabling it to generate music conditioned on a specific artist's style. Two evaluation protocols confirm that the generator captures meaningful stylistic structure: a sliding-window classifier consistently attributes conditioned continuations to the correct artist, far above unconditioned baselines; and a classifier trained entirely on synthetic generations identifies real pianists across 12 classes with 87% chunk-level and 95% song-level accuracy. Finally, we repurpose the classifier to locate the most characteristic moments within a performance, surfacing the specific musical gestures that distinguish each pianist's voice.