Science & Mathematics

Designing capable neural policies for multi-agent pathfinding (MAPF), which map local observations to action distributions, represents an important challenge in multi-agent learning, with practical relevance to scalable collective automation problems such as warehouse logistics. In contrast to prior studies focusing primarily on individual policy design, this study explores a meta-level Mixture-of-Experts (MoE) approach to learning-based MAPF. Specifically, given a collection of compact yet high-performing experts specialised for different map distributions, our architecture learns a router that selectively fuses their outputs according to each agent's local situation. We instantiate and validate this MoE approach with a commonly used graph neural network-based policy as well as a newly developed recursive action-token refinement model. Combined with supervised and reinforcement learning fine-tuning of the experts, our experiments show that MoE generally improves upon the pretrained policy and outperforms existing leading methods. At the same time, our findings reveal that appropriately trained, parameter-matched monolithic models still outperform MoE, highlighting a nuanced role for map-distribution MoE for MAPF.

Graph-to-table systems combine three design choices: an encoder that maps relational evidence to feature rows, a context constructor that selects labelled examples, and a tabular decoder. We investigate whether these components can be evaluated independently through matched comparisons that change one component while holding the rest of the pipeline and the evaluation constant. Across logistic regression, CATBOOST, and TABPFN-3 on CLUTRR, 57 GraphLog worlds, and 57 knowledge graphs, we show that conclusions about one component can change when the others change. On CLUTRR, absolute positional encodings maintain all distinctions among the evaluated queries and fit the short-path setting nearly perfectly; yet, they generalise worse than relation-count features because the supported prefix acts as a shortcut that breaks down on longer paths. On GraphLog, encoder preference depends on the decoder, while the relation-count representation degrades as CatBoost tree depth increases. Across different knowledge graphs, adding the frozen ULTRA score to local features is consistently useful, whereas adding local features to the score has a smaller and less stable effect. A pointwise TABPFN-3 readout using the full encoder underperforms native ULTRA on 48 of 57 datasets. These results show that encoder assessment depends jointly on context and decoder, while panel-level conclusions can also depend on evaluation and aggregation protocols.

Large language models remain unreliable on long-horizon tasks requiring procedural reasoning. We argue that a key cause is the absence of explicit procedural memory: accessible, modifiable knowledge of how tasks are accomplished, distinct from semantic and episodic memory. Goal-Oriented Graphs offer such a representation, encoding goals and prerequisite dependencies as inspectable symbolic structure. We examine their construction, the challenges they pose, and principles for integrating them with neural reasoning.

Chain-of-thought (CoT) enables constant-bit-size transformers to perform general-purpose computation through generated tokens (Li and Wang, 2025a), whereas looped transformers carry out the same computation by recurrence in hidden states (latent thought). We construct a family of constant-bit-size looped transformers that interpolates between CoT and latent thought. For a prescribed loop count $L=L(n)$ and a fixed integer $k\ge1$ satisfying $s(n)^{1/k}=Ω(L)$, our transformer simulates a time-$t(n)$, space-$s(n)$ Turing machine using a context window of $O(s(n)/L)$ and $O(t(n)s(n)^{1/k}/L)$ generated tokens, reducing both resources by a factor of $L$ relative to the corresponding unlooped queue-machine simulation. The construction also holds with softmax attention at constant bit-size, under the finite-precision kernel specified in this paper.

Multimodal Mixture-of-Experts (MoE) reasoning models rely on long chain-of-thought (CoT) for complex tasks, yet often overthink easy inputs, increasing token cost and latency, and sometimes degrading accuracy. Existing methods optimize reasoning length during training or impose external control at inference, but leave the underlying reasoning mechanism a black box. We find that fast and slow thinking exhibit distinct expert-routing patterns: holding the input fixed and varying only the prompted reasoning tempo produces stable routing shifts. Suppressing slow-thinking experts shortens the reasoning chain, whereas suppressing fast-thinking experts lengthens it, providing causal evidence that these experts influence reasoning depth. Based on this finding, we propose TempoMoE, a training-free, two-stage framework. It identifies fast- and slow-thinking experts via tempo-contrastive routing analysis and redistributes routing weights within the original top-k set according to sample difficulty. It amplifies fast-thinking experts and suppresses slow-thinking experts on easy inputs and reverses this adjustment on hard ones, enabling difficulty-adaptive control over reasoning depth. We evaluate TempoMoE on Kimi-VL, Qwen3-VL, and ERNIE-4.5-VL across six benchmarks. Compared with Vanilla, TempoMoE consistently improves accuracy while reducing output tokens, with accuracy gains of up to 2.6 points and token reductions of up to 35.4%.

Plug-and-Play (PnP) methods using continuous-time deep generative models as learned data distribution priors enable synthesizing high-quality solutions of model-based inverse problems. The quality of the solutions depends critically on the interplay and consensus between the inverse problem solver and the generative model sampler. We revisit a recently introduced flow-based PnP framework, which we refer to as PnP-CF (Consensus Flow), and extend its application to nonlinear inverse problems, including those with learned neural-network forward operators. PnP-CF combines a flow-matching denoiser with (i) ADMM-like dual variable updates and (ii) a time-adaptive pullback onto the intermediate flow probability path that combines a stochastic and a deterministic component. We discuss how these two features stabilize the algorithm and promote consensus, and demonstrate on a set of experiments that PnP-CF not only solves standard linear inverse problems with quality at least comparable to state-of-the-art baselines, but also extends gracefully to nonlinear inverse problems, including forward operators realized by pretrained neural networks, for which existing methods produce unsatisfactory results or fail completely.

Code generated by large language models often passes every functional test while remaining insecure. These security hallucinations are common, written with high confidence, and repeated across samples, so neither the generator's own uncertainty nor static analyzers reliably catch them. We introduce SPLASH, which is guided by a new measure called audit pressure. It contrasts teacher-forced log-probabilities for the generated program under a security review and a style review. Aggregated to statements and programs, and grounded in a probabilistic model of competing edits, audit pressure allows SPLASH to choose the secure program among a model's candidate programs and locate the vulnerable statement, with no model training, CWE-specific rules, or internal access to the generating model. On 63,684 programs generated by 12 models for 899 tasks, SPLASH picks secure code in 74% of cases, places the vulnerable statement first in 58% of insecure programs, and chooses better than a frontier LLM judge.

Large language models (LLMs) have recently shown strong potential for time series reasoning, while models that reason over time series are usually with long chains of thought. However, this process often consumes thousands of tokens, directly leading to substantial latency during inference. In contrast, System 1 models such as Jev make decisions instantly and can naturally be an alternative, but raw time series are not compatible with the embedding space of these models, thus hindering the reasoning performance, especially when input series contain thousands of numeric points. In this paper, we make a very early attempt to propose a time series System 1 model, namely TimesJev. Specifically, we first formalize System 1 sufficiency in time series reasoning, which motivates a compressed representation for time series with key frequency patterns preserved. To construct this representation, we derive a dimension reduction objective that combines (i) a sufficiency divergence that measures the core information lost during compression, and (ii) a realizability divergence that quantifies the incompatibility of the System 1 model to the optimal prediction given this representation. We optimize this objective by learning time series compression with a gating mechanism for sufficiency, and then aligning the selected representation with the embedding space for realizability. Finally, experiments on 13 datasets demonstrate that TimesJev outperforms reasoning models by over 62% with 62.9% less inference time than their fine-tuned versions on average.

Thin cylindrical objects offer little space for planar markers. We propose a dual-colour-stripe marker that uses the known curvature of an object to improve pose measurement with a single camera. The boundary between two colours provides a long centre line, while dots and printed edges supply features at different depths. We analyse how this depth separation improves the estimation of rotation about the cylinder axis, or roll. For a camera-facing marker at 15 cm, the Cramér-Rao lower bound on roll standard deviation is 2.00°for a flat marker, 0.38°for a single curved marker and 0.21°for a wrap-around marker. Monte Carlo roll RMS errors are within about 10% of the corresponding standard-deviation bounds. A first-order relation connects the sagitta of the curved marker to the roll sensitivity provided by a stereo baseline. The analytical and simulation results explain how curvature improves monocular measurement and provide a basis for designing marker geometry.

Surface irrigation planning requires screening many operating conditions and spatial configurations, while high-fidelity shallow-water-equation (SWE) simulations are expensive. We study tabular surrogate models for predicting maximum water height from drone-derived field morphology, terrain, hydrological, spatial, and scenario descriptors, where each row represents a spatial cell under one irrigation scenario. To assess deployment feasibility, we evaluate gradient-boosted trees, neural tabular models, AutoML, and tabular foundation models (TFMs) under two regimes: same-sector generalization, where a model predicts unseen scenarios after observing a limited set of configurations in the same sector, and sector-out transfer, where knowledge learned from source sectors is transferred to an unseen sector. The main findings are twofold: a) under the evaluated fixed 8/32 same-sector split, surrogates achieve strong performance, especially with raw spatial coordinates; b) sector-out transfer is substantially harder because the target sector has different coordinates and geometry. Sector-frame normalization of spatial coordinates improves transfer across similarly structured sectors, but does not solve cross-shape generalization. Overall, tabular surrogates appear useful for same-sector scenario screening under this fixed split, while robust transfer to unseen sectors remains open.

Building capable language models for India requires scalable multilingual data and robust evaluation. We introduce Nemotron-Pretraining-Indic-v1, a synthetic pretraining corpus of approximately 2.83 trillion tokens covering all 22 scheduled Indian languages in native script and twelve in romanized form. It is the largest openly released Indic pretraining corpus. Using our GPU-accelerated pipeline, we translate curated English documents, assess translation quality, and repair low-quality outputs while preserving technical content and document structure and maintaining source traceability. Through continued-pretraining experiments across twelve languages, we show that combining our synthetic corpus with natural Indic text improves knowledge and reasoning over a public-corpus baseline. Additional experiments show how data volume, script coverage, and Indic training share affect English performance. We also introduce the Indic-LLM-Eval-Suite, a multilingual benchmark suite for evaluating Indic language models across chat, instruction following, reasoning, knowledge, code, and translation. The suite combines existing Indic benchmarks with task-aware translations of English benchmarks, using automated quality assessment, targeted correction, and native-speaker review in a human-in-the-loop pipeline. We use the suite to comprehensively benchmark contemporary open-weight Indic and multilingual models across capabilities and languages, revealing substantial variation across tasks and languages. Together, these resources provide a scalable foundation for developing and evaluating more capable Indic language models.

Single-photon avalanche diode (SPAD) arrays offer photon-level sensitivity for sensing in optical integrated sensing and communication (O-ISAC). However, after each detection, a SPAD is blind for a dead time, which saturates the array at high received power. Existing O-ISAC analyses assume linear receivers and do not capture this effect. In this paper, we study ranging with a SPAD array in a monostatic O-ISAC system based on direct-current-biased optical orthogonal frequency-division multiplexing (DCO-OFDM). Sampling the array once per dead time, we model the number of pixels that detect a photon in each sampling interval as a binomial random variable whose success probability accounts for the intensity variation of the echo within a dead time. Based on this model, we derive the Cramér-Rao bound (CRB) for range estimation. We show that the SPAD count model of optical wireless communication, which assumes a constant photon rate over a dead time, causes a range bias that depends on the received power. Moreover, unlike for a linear photodiode, saturation degrades the ranging accuracy beyond a sensing-optimal transmit power. The rate-constrained operating point is then the larger of this power and the power needed for the target data rate. Numerical results validate the analysis against a continuous-time photon-counting simulation and show that the proposed model is needed for ranging.

Every scientific text is embedded in the writing norms of its time: it reflects earlier publications while also shaping the texts that follow. The growing use of LLM-based writing tools introduces a new factor in this evolving system, but the linguistic changes coinciding with their emergence remain insufficiently understood. We investigate whether the trajectories of scientific-preprint writing changed in rate or direction after conversational LLMs became publicly accessible, using the release of ChatGPT in November 2022 as the observational breakpoint. We compare pre- and post-2022 trajectories across vocabulary, punctuation, readability, and syntax in abstracts from arXiv's artificial intelligence, statistics, and quantitative biology categories and from medRxiv. Using interrupted time-series regression, we find substantial post-2022 changes across multiple feature families and research domains, including increases in average word length, reduced formula-based readability, and declines in maximum dependency-tree depth and sentence-level depth variation. Individual features follow different dynamics: some lexical markers accelerate persistently, while others rise and then recede. Our findings show that isolated marker frequencies capture only part of the shift: changes in the rate, direction, and structure of linguistic trajectories provide complementary evidence of broader post-2022 change in scientific-preprint writing. More broadly, these findings motivate monitoring scientific writing as a multidimensional, evolving system rather than through a fixed set of LLM-associated markers.

Performing accurate protein-glycan docking is critical to understanding protein-glycan interactions which govern many biological processes. However, this important task is hampered by the lack of tailored datasets, benchmarks and models for protein-glycan docking. To fill this blank, we first curate a standard benchmark ProtGlycanDock for training and evaluating protein-glycan docking models under both protein and glycan generalization settings. On this benchmark, we evaluate ten models in three categories, physics-based docking methods, machine-learning-based docking models and molecular structure prediction models. To obtain more precise binding poses, we further propose two finetuning techniques to inject the knowledge of glycan conformations into a general-purpose molecular structure prediction model. These two techniques respectively augment model inputs with glycan topological features and refine glycan binding pose with a glycan energy loss. Benchmark results show the superiority of AlphaFold 3 among existing methods, and the two proposed finetuning techniques can be well combined to achieve a state-of-the-art model for protein-glycan docking. Also, we perform stratified benchmark analysis and case studies to better understand the strengths and weaknesses of current models, providing insights for future model improvements. We maintain datasets at https://huggingface.co/datasets/ProtGlycanDock/ProtGlycanDock and source codes at https://github.com/kasawa1234/ProtGlycanDock

LLM-based guardrails are effective against indirect prompt injection (IPI), but their performance depends heavily on model capability: as the guardrail model becomes smaller or weaker, detection degrades substantially. This raises a natural question: can we retain strong IPI defense while relying only on a small LLM? To understand why, we revisit what an LLM guardrail actually does. Detecting IPI requires simultaneously locating potentially injected instructions within heterogeneous tool outputs and judging whether they are authorized by the user's request. We find that the difficulty for small models arises largely from satisfying both requirements jointly, rather than from either subproblem alone. This suggests a different path to lightweight defense: explicitly decoupling localization from decision-making. In particular, localization needs no strong language-model reasoning, since injected instructions are often semantically discontinuous with their surrounding content and can be partitioned into local regions via lightweight unsupervised segmentation. Once the observation is partitioned this way, the remaining authorization check becomes a localized judgment that a small LLM can handle. Building on this insight, we design a lightweight IPI guardrail that performs attack-agnostic localization followed by block-level authorization judgment with a small language model. By separating where to look from what to decide, our approach substantially reduces the capability required of the guardrail model while preserving strong defense performance. Empirically, our guardrail achieves a 0.32% attack success rate with negligible loss in benign utility while remaining effective with substantially smaller models. These results show that strong IPI defense need not rely on strong LLMs when localization and decision-making are explicitly decoupled.