Papers

Large language models (LLMs) are increasingly expected to orchestrate diverse tools over long-horizon tasks. Existing agents typically use Direct Tool Calling (DTC), where the model invokes tools sequentially, and tool responses accumulate in the model context, which can lead to context rot and orchestration breakdown. In contrast, Programmatic Tool Calling (PTC) invokes tools through executable code and stores intermediate results in runtime variables rather than appending them to the context. Prior work has explored PTC mainly as an inference-time optimization, but its benefits for agentic training remain unclear, and suitable verifiable training tasks are scarce. In this work, we introduce ProgAgent, a full PTC training stack, comprising a pipeline for synthesizing verifiable long-horizon tasks, a lightweight PTC runtime, and a two-stage training recipe. Using only 215 tasks instantiated at four environment scales, ProgAgent yields higher pass rates than the evaluated baselines of comparable size across four benchmarks. On LOCA-bench, ProgAgent-14B achieves a 30.7% pass rate at a 64k-token environment state, where these baselines score zero. Cross-harness experiments show that PTC-trained models outperform DTC-trained models under both evaluation interfaces, suggesting that the benefits of PTC training extend beyond the interface.

Predictively-oriented (PrO) filtering for state-space models was proposed in Shen et al. [2026] as a promising route to avoiding over-confident inferences in settings where either the dynamical model, the measurement model, or both, could be misspecified. Our focus is on nonlinear state-space models, where the local linearisation used to construct the PrO filter in Shen et al. [2026] deteriorates as nonlinearities enter either the dynamical model or the measurement model. Our proposed solution is based on unscented transforms, an established tool from classical filtering which provides a more accurate mechanism to propagate Gaussian uncertainty through a nonlinear functional, and which includes the cubature rule as a special case. We introduce two variants: replacing the linearisation in the predict step alone, which we call UKF-PrO, has a computational cost comparable to the existing PrO filter; replacing it in the update step as well, UKF-PrO+, requires a substantially more expensive joint optimisation. Performance is empirically assessed on the Lorenz96 model under conditions of increasing difficulty, in which the state-space model is systematically misspecified; we find that UKF-PrO with the parameters of the cubature Kalman filter offers consistent improvements, which become substantial under strong nonlinearity, whereas the default unscented parameters are unreliable, and that the full update-step extension with the same parameters does not justify its additional cost.

Camera-first agents must act on the visual world as it unfolds. We argue that the central obstacle for future camera-first agents is not recognition ability - multimodal models (vision-language and live/streaming alike) generalize well through language and remain the right tool for semantic reasoning - but time. Many visual targets are ephemeral: they appear, are query-relevant, and leave within a short window. Binding memory to the same end-to-end path as reasoning means that by the time a monolithic pipeline completes, the evidence is gone. We formalize this as Ephemeral-Target Interaction (ETI) and a Perceptual Deadline Contract separating a hard memory deadline from a soft reasoning deadline, grounded in the bounded, passive nature of sensory and working memory. We propose a two-deadline architecture: an always-on, bounded, non-semantic fast path; an active remembering mechanism that commits a moment to durable memory before it is overwritten (event-driven in v1); and a deferred deliberation layer (multimodal model + tools). We instantiate the contract on real MOT17 track windows, a consistency check, and calibrate the single-path baseline from measured frontier-model latency: even a short answer takes 1.2-9.3 s, and an agentic identify-and-search longer. Answers can be late; evidence cannot be missed.

Virtual Reality (VR) requires strict Quality-of-Experience (QoE) guarantees, making radio resource management harder. Static slice allocations are wasteful, while dynamic approaches often rely on heuristic penalty functions lacking an empirical basis. We introduce a QoE-driven network-slicing framework for Open RAN in which resource allocation is guided by empirically derived multi-user VR traffic statistics. Closed-form n-fold convolutions model the aggregate data rate and inter-arrival time of n concurrent users, validated empirically for up to four. The fitted data-rate cumulative distribution function is checked against the measured throughput to give a percentile-based satisfaction indicator; the inter-arrival law sets the slice targets. A near-real-time RIC xApp closes a loop over the E2SM-KPM and E2SM-RC interfaces to apply slice-level Physical Resource Block (PRB) quotas. On an over-the-air testbed it reclaims up to 41% of the cell's committed PRBs from a greedy neighbouring slice while never reducing the protected VR slice. An analytical study then estimates the PRB demand and per-user QoE across user counts and channel qualities, giving the admission boundary beyond which the target QoE is unattainable. A controlled comparison further shows that a fixed-rate synthetic load overstates the VR slice's PRB footprint by 68%, showing the necessity of data-driven slice dimensioning.

Multimodal contrastive learning typically relies on manually specified scoring functions, such as cosine similarity, to quantify the similarity between representations from different modalities. Several such functions have been recently proposed to extend contrastive learning beyond two modalities. In principle, however, any function that assigns high compatibility to matched samples and low compatibility to mismatched ones could serve as a scorer. Motivated by this broader view, we instantiate the scorer with a Transformer and find that even a randomly initialized model, when kept frozen, can outperform strong handcrafted multimodal contrastive baselines, whereas optimizing the scorer jointly with the encoders is less effective. Building on this observation, we introduce the Prior-Fitted Scorer (PFS): a Transformer scorer pre-trained on synthetic multimodal data generated from structural causal model (SCM) priors before transferring it to real-world data. The SCM exposes the scorer to diverse dependency structures, including shared, pairwise, and private information, providing a structured prior over higher-order cross-modal similarity. Across four real-world datasets, frozen Transformer scorers are competitive to contrastive baselines, and prior-fitting further improves the random initialization. More broadly, our results suggest that multimodal contrastive learning need not rely on manual similarity functions: Transformers can provide strong inductive biases, while causal pretraining can further shape these biases.

Bengali has no grammatical gender, but honorificity is obligatory: every third-person reference must choose ordinary or respectful register. English has no such signal, so the system must choose a level of deference that the source never specified. We use this property to measure how large language models (LLMs) allocate respect in English–Bengali translation. Our evaluation set contains 2{,}016 sentences across 72 occupations. Each appears in seven sentence frames under four pronoun conditions. Masculine and feminine versions differ by one word, forming counterfactual pairs. We report a single Honorific Bias Score (HBS) for each model on a 0–100 scale, covering both the gender and occupation axes. A lower score means that respect is allocated more evenly. For the occupation analysis, we compute each occupation's honorific rate across the seven no-pronoun sentence frames. The standard deviation of these 72 occupation-level rates ranges from 17.2 to 35.5 percentage points across models, showing substantial variation in how honorifics are allocated across occupations. Occupation is associated with 3.2% to 51.5% of the variance in register choice, compared with at most 1.2% for gender. Gender effects are much smaller overall: only two of the nine models show significant directional gender gaps, and they favor opposite genders. HBS ranges from 19.9 to 70.1 across the nine models, showing substantial variation in how evenly the models allocate respect.

Coding agents increasingly maintain repositories over time, where each change becomes the starting point for subsequent tasks. Yet existing benchmarks either start tasks from developer-written code or cannot separate later-task difficulty from the effects of an agent's earlier changes. We ask whether agent-written code makes subsequent maintenance harder and whether the cost stems from inherited test failures or the code itself. We introduce SWE-Maintain, comprising 100 three-event maintenance episodes from 50 real-world Python repositories, with hidden tests certified across each sequence. Each later task is evaluated twice with the same agent and tests: from the developers' preceding patches (Developer-History) and from the agent's own changes (Agent-History). Across five coding systems, two backbone models, and 1,996 paired later events, Agent-History reduces later-task resolution by 5.6 percentage points on average. When success additionally requires all earlier tests to pass, the gap reaches 14.0 points. Two-thirds of Agent-History starting repositories already fail earlier tests, and only 1.9% of events starting from those states end with all accumulated tests passing. When the inherited code passes earlier tests and the next task is not already solved, the disadvantage disappears (+2.7 points; 95% CI [-0.5, +6.1]). These findings suggest that the main downstream cost of agent-written changes lies in inherited test failures rather than a measurable maintenance disadvantage in code that preserves earlier tested behavior. Evaluations should therefore include agent-produced history, and coding agents should preserve earlier behavior across changes.

Federated learning (FL) lets distributed intrusion-detection systems (IDS) train on private traffic, but exposes the global model to poisoning and backdoor attacks under non-IID client data. A popular defense family scores or filters clients using a small trusted "root" dataset and the similarity of client behaviour to a trusted reference. I re-examine these trust-based defenses under a protocol designed to remove common evaluation artifacts: train-only feature standardization, a server root set disjoint from client data, multi-seed (n=6) runs with paired significance tests, formal (varepsilon, delta) accounting for DP-SGD, and a defense-aware adaptive attacker. Across three NIDS datasets (CSE-CIC-IDS2018, NF-ToN-IoT, NF-UNSW-NB15-v2) I find: (1) a representation-similarity aggregator (RepGuard) significantly improves classification utility and seed-stability over FLTrust under heavy non-IID poisoning, and - under severe heterogeneity or many adversaries - also lowers backdoor ASR substantially (e.g., 0.48 vs FLTrust 0.86 at Dirichlet alpha=0.1); (2) at strong trigger magnitudes no defense reduces backdoor attack success rate (ASR approx 0.95); (3) a controlled trigger-magnitude sweep shows defenses help only for small triggers (ASR 0.10 for RepGuard at magnitude 0.5, vs. 0.40 undefended) and collapse to ASR approx 1.0 by magnitude 2; (4) a triggered-canary check that assumes knowledge of the attacker's trigger - the assumption behind much prior "success" - lowers ASR from 0.49 to 0.12, i.e., trigger knowledge, not representation scoring, suppresses the backdoor; (5) correctly-accounted DP-SGD (varepsilon in [0.3, 16]) costs little utility but does not mitigate the backdoor. I release the protocol so future FL-IDS defenses can be evaluated without the leakage and oracle assumptions that inflate reported robustness.

AI coding agents now write much of the code that ships, yet whether that code is secure remains unclear. We introduce the Last Secure Code Benchmark, a comprehensive evaluation of secure code generation by coding agents. Built from 150 real vulnerabilities, its 450 tasks span seven widely used languages (Java, C, C++, JavaScript, Python, PHP, and Go), 24 CWEs including 21 of the 2025 CWE Top 25 and 8 of the Top 10 KEV weaknesses, more than any benchmark we compare with, and three vibe-coding scenarios: filling in a function, writing a file, and building a repository's source tree from scratch. Every task asks only for functionality, through a specification that never mentions security, and hidden functional and security tests grade the result. Across six frontier agents, 73.6% to 91.3% of tasks pass functionally, yet even the strongest, Claude Fable 5, produces code that is both functional and secure for only 25.1% of tasks, and the others for 16.4% to 24.7%. The best model only passes 12% of the vulnerability scenarios across all vibe coding setting, posing strong concern on the security of the code written by vibe coding.

Reconfigurable intelligent surfaces (RIS) combined with rate-splitting multiple access (RSMA) provide a promising solution for improving the spectral efficiency (SE) of sixth-generation (6G) communication systems. However, most existing studies neglect the practical coupling between the RIS phase shift and amplitude, which may cause phase-shift mismatch and significant performance loss. In this paper, we investigate the multiuser sum-rate maximization problem for practical RIS-enhanced RSMA systems considering amplitude-phase coupling. To address the resulting non-convex problem, an effective variable separation algorithm is proposed. The proposed algorithm is compatible with RIS reflecting optimization under the continuous phase shift or any amplitude-phase coupling relationship and can also be directly extended to practical RIS-enhanced NOMA and OMA systems. Simulation results demonstrate performance gain of over 190% compared with the ideal-RIS-based benchmark, while the proposed RSMA scheme significantly outperforms the corresponding NOMA and OMA schemes.

We study randomized row-action methods for problems with linear observations, convex constraints, and non-Euclidean geometry. Our starting point is a unified formulation in which the data-fitting term acts in observation space, while the feasible set and the regularizer encode structure in variable space. It covers constrained least squares, nonnegative and simplex-constrained models, regularized data fitting, and equality-constrained Bregman selection problems. The structure of these problems leads naturally to two algorithmic branches. The first uses primal Bregman-proximal row steps for soft data-fitting problems. The second uses dual Bregman–Kaczmarz steps for problems with hard equality constraints. We first derive a common one-step inequality. We then give deterministic and stochastic convergence results for the soft branch under relative smoothness. For constrained least squares, the normalized row oracle has exact first- and second-moment identities. They yield an expected ergodic bound in the inconsistent case, an $O(1/K)$ bound in the interpolating case, and linear convergence of the actual iterates under a residual error bound. For equality-constrained selectors, we prove linear convergence in expected Bregman distance under a residual-to-Bregman error bound. We recall global sufficient conditions for this bound and derive local constants for the nonnegative quadratic geometry and the entropy geometry on the simplex. Numerical experiments illustrate the noisy stepsize trade-off and the local entropy selector theory, and show that entropic row-action updates can recover interpretable factors in a simplex-structured nonnegative matrix factorization problem.

Multimodal large language models (MLLMs) can reproduce sensitive, copyrighted, or unsafe information from their training data. Effective unlearning must remove target information while preserving non-target content, even within the same response. Existing methods typically apply uniform forgetting objectives to complete responses and rely on fixed, off-policy supervision, resulting in coarse token control and misalignment with the model's current behavior. We propose CUSE, a Counterfactual-guided Unlearning framework via on-policy Self-distillation for multimodal large language models. It first constructs counterfactual inputs by removing target-related evidence and uses the resulting changes in token probabilities to identify which tokens require stronger forgetting. It then distills token-level forgetting and retention targets on responses sampled from the current policy, with a frozen copy of the original model serving as the teacher model augmented with the corresponding forgetting- and retention-oriented privileged contexts. This design jointly determines what to forget and what to preserve. On MLLMU-Bench, CUSE reduces forget-set ROUGE-L by 33.6% while achieving 1.2% higher retain-set ROUGE-L than the strongest baseline.

Model merging combines fine-tuned experts into one network without retraining, and closed-form methods compute each merged layer from statistics of its input activations. However, these statistics are measured on a single expert, while in the merged model, each layer receives activations from the merged layers that precede it. Recent work re-measures these statistics in the merged model but still forms them from a single network. We formulate closed-form methods as a linear system and show that they use the same matrix on both sides. We then introduce CrossMean, which replaces this matrix on one side with a bilinear statistic that pairs merged and expert activations. From the same statistics, we also derive a label-free criterion that indicates the task sets in which accounting for the merged activations improves the merge more. Empirically, task sets with a lower criterion value show a larger improvement. On the eight-task benchmark of each vision backbone that we tested and on BERT encoders, CrossMean improves on merging with statistics measured on the experts, and with all linear layers merged, it also improves on re-measuring these statistics in the merged model. These results suggest that merging benefits from pairing merged and expert activations.

Piezoelectricity arises from the coupling between electric polarization and mechanical deformation, described by a third-rank tensor whose coefficients are constrained by crystal symmetry. In natural crystals, atomic periodicity and energetic stability restrict accessible symmetries, limited to 20 distinct anisotropies that can exist. Here we establish a general framework, based on graph representation, for controlling the full effective piezoelectric tensor through architecture by resolving how macroscopic symmetry, local polarization and deformation of structural elements act together. This enables each coefficient to be traced to the members that reinforce or cancel it and be controlled through targeted geometric changes. We further uncover frequency-selective tensor reconfiguration, in which excitation frequency switches the dominant piezoelectric coefficients of a single architecture. These principles open a broader class of applications in which a single piezoelectric architecture can deliver programmable sensing, actuation and mode conversion across multiple directions and deformation states. We realize a multi-degree-of-freedom actuator that integrates distinct actuation modes in one block and reduces volume by 90% relative to conventional multi-stack assemblies.

Nheengatu (yrl), the Amazonian Língua Geral, has 6–10 thousand speakers and is co-official in São Gabriel da Cachoeira (Brazil), yet remains poorly served by evaluated Portuguese–Nheengatu MT resources. We present the largest parallel corpus assembled for the language (14.5k Portuguese–Nheengatu pairs), anchored by the 2023 official translation of the Brazilian Federal Constitution; the first MT evaluation set built from transcribed spontaneous speech; and, to our knowledge, the first open-weight MT models dedicated to bidirectional Portuguese–Nheengatu translation. Three findings organize the paper. First, translation into Nheengatu on held-out constitutional articles plateaus on development data regardless of training length or model size, and much of the remaining error sits in conventional legal terms that the written sources do not attest consistently; a terminology success rate over a frozen lexicon shows that glossary mechanisms at training time and lexical constraints at decoding each raise term realization, the latter at a cost in repetitive output that chrF++ barely registers. Second, a corpus expansion that adds transcribed conversation raises speech chrF++ by 24–27 points while leaving written domains unchanged, a register gap that written-only benchmarks cannot see. Third, a larger NLLB model helps translation into Portuguese but not into Nheengatu on legal or speech text. For Portuguese-to-Nheengatu translation, our final systems reach 31.7–32.8 chrF++ on article-disjoint constitutional text, 64.1–64.5 on the near-domain set (62.1 after removing items with target-side content overlap), and 45.2–47.5 on speech.

In 1997 Guo Wuwen proved that if $(\lambda_1,\lambda_2,\dots,\lambda_n)$ is the spectrum of a nonnegative matrix with Perron eigenvalue $\lambda_1$ and $\lambda_2$ real, then $(\lambda_1+t,\lambda_2\pm t,\lambda_3,\dots,\lambda_n)$ is the spectrum of a nonnegative matrix for every $t\ge 0$. Whether the same holds for symmetric nonnegative matrices has remained an open question. We give an affirmative answer, in a slightly stronger form: if $A$ is a symmetric nonnegative matrix with eigenvalues $\lambda_1\ge\dots\ge\lambda_n$ and $d_1\ge|d_2|+\dots+|d_n|$, then there is a symmetric matrix $B\ge A$ with eigenvalues $\lambda_1+d_1,\dots,\lambda_n+d_n$, and $B$ has the same diagonal as $A$ when $d_1+\dots+d_n=0$. This leads us to call a symmetrically realizable list Guo-extreme if none of Guo's moves can be reversed within the set of symmetrically realizable lists. Every symmetrically realizable list is a Guo-extreme list plus a vector from the cone of Guo's moves. Our second main result is a certificate, modelled on a theorem of Laffey on extreme symmetric matrices: if $A$ is irreducible and every nonnegative symmetric matrix that commutes with $A$ and vanishes wherever $A$ is positive is a multiple of the identity, then the spectrum of $A$ can be moved by a small amount in every trace-preserving direction (in every direction, if $A$ has a positive diagonal entry) and remain symmetrically realizable; in particular it is not Guo-extreme.

We give an explicit treatment of the third tensor power of the Clifford representation on $n$ qutrits, for $n\geq2$. This representation has five pairwise inequivalent irreducible constituents, with multiplicities $1,1,1,1,2$. The decomposition reduces the Clifford twirl of a tensor cube to four real parameters. Two of them are spectral moments of the input. The other two are cubic invariants expressed as sums of cubes of the Weyl characteristic function and the discrete Wigner function. On trace-one Hermitian matrices, the invariant polynomials of degree at most three are spanned by the constant function and four algebraically independent functions. These functions do not separate Clifford orbits. For pure inputs a single parameter suffices. We recover its range and the characterization of its maximum by stabilizer states. We bound the stabilizer fidelity, determine full separability across the three copies in the symmetric family, and recover the criterion for a complex projective $3$-design. We classify the jointly invariant quantum Boolean functions. For stabilizer testing and for testing unknown observables we give explicit bounds on copies and queries. We also give character formulas for the finite projective Clifford groups and record the complete character table for two qutrits.

Human-in-the-loop optimization refers to systems that present choices to a human decision-maker and query their judgment to optimize system parameters iteratively. The user interface typically provides either an absolute rating scale or asks for a preferred option. However, these approaches can potentially suffer from calibration problems, inconsistencies, and completeness violations of user preferences in such a sequential evaluation. This paper revisits the design space of opinion measurement interfaces for human-in-the-loop optimization. We conducted a study (N=360) using text summarization and image color enhancement tasks to evaluate human sequential feedback in six representative opinion measurement interfaces. Based on our analysis, we recommend using a listwise approach to support sequential opinion measurement, which counterweighs individual rating and preferential choice limitations and conveys more information for subsequent use. We further discuss the trade-offs in different user interface designs and provide guidelines to inform future research.

The predominance of English-language Web content creates an information-access barrier for speakers of low-resource languages such as Amharic. This paper presents the design, imple-mentation, and evaluation of a bidirectional Amharic-English cross-language question answering (CLQA) system that allows a user to formulate a natural-language question in either Amharic or English and retrieve an answer from documents written in the other language. The system integrates a statistical machine translation (SMT) component for query translation, a maximum-entropy named entity recognition (NER) component for Amharic text processing, semantic-based indexing, question analysis, passage retrieval, and answer extraction. A domain-diverse Amharic-English parallel corpus was used to train the SMT component, while monolingual cor-pora supported language modeling. Documents were normalized, tokenized, stemmed, indexed, and semantically enriched before retrieval. The original evaluation used precision and recall as the principal effectiveness measures. For Amharic-to-English retrieval, the system achieved 72% precision and 79% recall; for English-to-Amharic retrieval, it achieved 65% precision and 70% recall. These findings show that a modular translation-and-retrieval architecture can provide useful cross-language access even under low-resource conditions. This revised article preserves the original experimental claims while situating them in the context of subsequent multilingual transformer and Amharic QA research.

Given a smooth domain $D\subset\mathbb S^{N-1}$, $N\geq 3$, and a positive constant $H$, we study radial graphs over $D$ with constant mean curvature $H$ which meet orthogonally the boundary of the cone spanned by $D$. The corresponding problem is given by a quasilinear elliptic equation on $D$ with homogeneous Neumann boundary condition. We investigate the structure of the solution set under smooth perturbations of the domain which are parametrized by an open set of $C^{3,\alpha}$-diffeomorphisms of the sphere. After pulling the equation back to a fixed reference domain, we apply a classical transversality theorem and prove that, in the class of domains for which $N-1$ does not belong to the Neumann spectrum of the Laplace-Beltrami operator, there is a residual set for which every solution is nondegenerate. A consequence of our result is that, for every such domain and every $R>0$, the set of solutions whose $C^{2,\alpha}$-norm is bounded by $R$ is finite.

Hybrid language models interleave Gated DeltaNet (GDN) layers, whose recurrent memory has a fixed size, with full-attention layers, whose key-value (KV) cache grows with the context and dominates serving memory. In this paper, we ask whether the recurrent memory already knows what the attention memory will need. We first show that which GDN heads of Qwen3.8-27B keep long memories is fixed by the weights rather than by the document. Next, we define the future utility of a KV entry by how much removing it changes the attention output over the model's own continuations, and find that the write gate, decay and write magnitude that GDN layers emit during prefill anticipate this utility before any query is seen. Because the relation is spread across many GDN heads and layers, and a linear map recovers its trend but less of its high-utility tail than a nonlinear one, we propose GateSight, which learns it with lightweight per-head multilayer perceptrons that read these signals and no attention score. Their predicted utility decides which entries each attention layer keeps, and how much each layer keeps is set by the model's native attention output gate, whose mean differs by more than an order of magnitude across layers. Finally, at serving time, GateSight scores the cache once after prefill and physically releases discarded KV entries. Experimental results on SCBench and RULER-4096 with Qwen3.8-27B and, with the same recipe, the sparse-attention model Qwen3.8-Flash-Next, up to the 262K-token limit of the models, show that with a tenth of the attention KV cache, GateSight is the most accurate of five query-agnostic methods on three of the four model-benchmark pairs and leads KVzip on all four, by 9.5 to 47.2 points. Notably, it scores a full-length context with negligible overhead relative to prefill.

Versatile quadruped locomotion depends on a tight coupling between what the robot perceives and what its body can execute. Yet existing visual controllers typically organize depth history using fixed temporal patterns, overlooking that the same temporal span corresponds to different spatial ranges as the robot's speed, posture, and gait phase change. Consequently, the same visual history may be insufficient in one motion state but outdated or misaligned in another. To address this limitation, we propose TempoLoco, a body-aware dynamic visual perception framework that organizes visual memory according to the robot's current locomotion state. TempoLoco learns multiple complementary temporal views from a long depth buffer, encodes them within a shared feature space, and dynamically aggregates them using proprioceptive history and multi-scale terrain evidence. This enables a single policy to align visual anticipation with the robot's current motion capabilities across complex terrains, including stairs, platforms, gaps, discrete obstacles, rough ground, and slopes. From the experiment analysis, TempoLoco achieves a 78.4% success rate, compared with 58.2% for the strongest fixed-history baseline and 62.1% for uniform multi-scale fusion. In separate capability evaluations, it completes 41/45 simulated trials across nine terrain types and 23/25 real-world trials on a Unitree B2 with a fixed arm, demonstrating robust and deployable locomotion through body-aware dynamic visual perception.

We investigate signal transmission in a network of unidirectionally coupled plasma oscillators, where only the first oscillator is driven by a weak periodic input. As the signal propagates through the network, nonlinear interactions generate higher-order harmonic components. Spectral analysis identifies five distinct transmission patterns for the fundamental frequency and odd harmonics, and four patterns for even harmonics, encompassing damped propagation, undamped transmission, and localized amplification. The transmission characteristics are governed by both the coupling strength and system parameters. Furthermore, we analyze how variations in these key parameters shape the overall signal dynamics.

Random Fourier features approximate a shift-invariant kernel in $\mathbb{R}^d$, such as the Gaussian kernel $\exp(-\tfrac12\|x-y\|_2^2)$, using independent samples from its spectral measure to construct an unbiased estimator for the kernel value with subgaussian concentration. However, generating $m$ such features requires $O(md)$ time. A long line of work on structured random features has accelerated this computation for the Gaussian kernel and other radial kernels, albeit with weaker concentration guarantees. We close this gap by showing a random feature map that can be computed in near-linear time $O((d+m)\log d)$ and yields near-subgaussian concentration. We obtain it through a new and tighter analysis of the Fastfood feature map of Le, Sarlós and Smola (ICML 2013), using Brascamp-Lieb inequalities and the Schoenberg representation of radial kernels.

Credit assignment in modular financial reinforcement learning is difficult to evaluate because an attributed signal may improve a component loss without adding decision-relevant information or improving economic outcomes. We propose an evaluation framework combining equal-information reconstruction controls, fixed-utility comparisons and prespecified downstream guardrails. The framework is instantiated in an attribution-guided modular trading architecture and tested in three controlled studies covering update allocation, supervision-target representation and portfolio risk control. We find that residual-based allocation is exactly reproduced by an ordinary loss-budget control under the disclosed interface; direct shortfall supervision improves component prediction but fails decoded-outcome preservation; and component-wise critic supervision worsens risk-condition certainty equivalent and drawdown relative to scalar supervision under fixed utility. These results identify three distinct evaluation failures: redundant attribution signals, representation-induced prediction trade-offs, and misalignment between component supervision and downstream risk outcomes. A preserved historical engineering case further shows why module-level learning claims require run-level provenance. The contribution is a reusable methodology for evaluating credit assignment in expert systems operating under partial observability, delayed feedback and economic risk constraints.

The Distributed Flexible Job Shop Scheduling Problem (DFJSP) requires the joint determination of factory assignment, machine selection, and operation sequencing to minimize the makespan. This paper presents an exact SAT-based approach built on an efficient SAT formulation that exploits problem structure through propagation, symmetry breaking, and incremental SAT solving to effectively address these coupled decisions. The proposed method is evaluated on 76 standard benchmark instances from the Fattahi, Brandimarte, and rdata datasets under two- and three-factory configurations. Together with previously established results, the new SAT optimality certificates bring the total number of proven-optimal instances to 72 out of 76, including three newly certified instances; for two of these instances, SAT also improves the previous best-known makespans. Across the benchmark configurations, SAT achieves the best results on four of the five configurations and remains competitive with strong exact approaches such as CP and LBBD1_CPIL, as well as existing metaheuristic methods. These results demonstrate that the proposed SAT framework is an effective exact approach for the DFJSP, combining strong solution quality with rigorous optimality certification and offering a complementary exact method for future DFJSP studies and evaluations on additional benchmark instances.

Inspired by the great success of foundation models in the development of artificial general intelligence, Graph Foundation Models (GFMs) have recently attracted significant attention and have the potential to advance graph intelligence. Exhibiting such intelligence requires effectively learning knowledge from large-scale, multi-domain graph data during pre-training. Increasing domain diversity during pre-training is expected to yield richer and more generalizable knowledge. Most existing methods pursue this goal by designing domain alignment strategies that help a single graph model learn from multiple domains. However, we observe that existing methods struggle to benefit from increasing domain diversity. Through empirical and theoretical analysis, we reveal two key limitations. Firstly, pre-training in most GFMs is confined to a single message-passing paradigm. Since propagation flattens relations among neighboring nodes, the graph model can hardly perceive the knowledge beyond propagation, namely the attribute-predictable component of these relations. Secondly, domain alignment strategies often rely on randomly initialized parameters that are jointly updated with the graph model using immediate loss feedback, leaving early knowledge acquisition vulnerable to cross-domain interference. To address these issues, we propose FLO, a Fast–sLow reciprOcal approach for multi-domain graph pre-training. Specifically, a propagation-free attribute branch predicts the graph model's relations without message passing and returns these predictions to refine the model. We further stabilize knowledge acquisition by pairing a fast process that adapts to the current domain with a slow process that integrates knowledge across domains. On 12 datasets, FLO benefits from increasing domain diversity and achieves the best average performance in cross-domain and cross-task transfer.

We consider the stability of light rings in four-dimensional circular stationary axisymmetric spacetimes. Using a Hamiltonian formalism for null geodesics and the Killing symmetries of the spacetime, we show that the transverse Hessian of the geodesic potential, which controls the light ring's stability, reduces there to a pure curvature expression. Its eigenvalues are $\mu_{\pm} = \Phi_{0 0} \pm \lvert \Psi_{0} \rvert$, where $\Phi_{0 0}$ is a Ricci scalar and $\Psi_{0}$ a Weyl scalar of a complex null tetrad adapted to the light ring. Through the Sachs optical equations, these quantities are responsible for focussing and shear of a null geodesic congruence. Stability requires both eigenvalues to be positive, so a light ring is stable if and only if $\Phi_{0 0} > \lvert \Psi_{0} \rvert$ there. Assuming the Einstein field equations with generic stress–energy tensor $T_{a b}$, we recast this as $4 \pi T_{a b} p^{a} p^{b} > \lvert \Psi_{0} \rvert$, where $p^{a}$ is the light ring's tangent vector: for stability, the focussing by matter must dominate the local tidal shear. The trace of the Hessian alone provides an inexpensive test of stability, requiring no knowledge of the Weyl curvature, since a light ring at which the strict null energy condition is violated cannot be stable. Applying the test to specific matter models, we recover known results for vacuum and electrovacuum, with and without a cosmological constant. We also demonstrate that stable light rings are forbidden for a minimally coupled real scalar field with any self-interaction potential, since such a field must be invariant under both isometries. We show that a complex scalar field escapes this obstruction whenever it carries harmonic dependence on the Killing coordinates.

World Action Models (WAMs) commonly couple video and action experts through dense, layer-wise Mixture-of-Transformers (MoT) attention, exposing action prediction to irrelevant visual features while enforcing matched expert depths. This coupling hinders selective visual-feature use and introduces redundant computation. We introduce Bridge-of-Experts (BoE), instantiated as BridgeWAM, which aggregates visual, language, and proprioceptive priors through 32 Latent Bridge Queries (LBQs) into a shared, limited latent space. Joint video-action supervision lets action gradients guide the Video Expert toward representations better suited for action prediction, while the Action Expert learns to selectively retrieve task-relevant information through cross-attention. Decoupling expert depths and reusing LBQs during action denoising yield a 5.06B model with a 30-layer Video Expert and a two-layer Action Expert, achieving 98.25% success on LIBERO and 72.24% pooled success on LIBERO-Plus, demonstrating effective task execution and strong generalization with an ultralight action head. Ablations support joint representation adaptation and unified LBQ conditioning. Relative to Fast-WAM, BridgeWAM reduces single-sample inference FLOPs by 28.72% in simulation profiling and real-robot latency to 77.1 ms (a 2.47-fold speedup). Qualitative experimental results further indicate preferential attention to target objects and task destinations across perturbations.

Extreme low-bit weight-only quantization is an effective approach to reducing the deployment costs of large language models. Below 3 bits, scalar quantization often suffers severe accuracy loss, while vector quantization (VQ) better preserves accuracy through more expressive joint encoding. However, existing VQ methods still exhibit substantial degradation on real-world tasks, and two design challenges remain underexplored: (1) protecting sensitive weight channels within jointly encoded vectors; and (2) allocating a fixed bit budget among representation components, such as direction and radius, to suit different weight matrices. To address these challenges, we propose SAGE: Split Adaptation and Grouping by Energy for weight-only vector quantization. First, Energy-Stratified Grouping (ESG) couples channel grouping with activation-weighted fitting, grouping channels from different energy tiers within each subvector to prioritize sensitive channels. Second, Adaptive Bit Split (ABS) balances direction-codebook capacity and radius precision per matrix, selecting the split with the lowest activation-weighted reconstruction error under a fixed budget. Across benchmarks covering knowledge, mathematics, code generation, instruction following, and tool use, SAGE achieves state-of-the-art performance in the 2-bit quantization setting. On Qwen3-30B-A3B, it retains 95.3% of the BF16 model's average benchmark score. With our custom-designed CUDA kernel, SAGE delivers up to a 2.55-fold non-streaming end-to-end speedup over FP16. The code is available at https://github.com/Zincwolf/sage.