Papers

Logical rule induction seeks interpretable rules that transfer across propositional schemas. This requires respecting symmetries: atom naming, example order, polarity flips, and label swap. Enforcing exact symmetry by construction lets one trained inducer scale beyond its training schemas. Our central contribution is a canonical export that decodes a discrete rule from literal scores. It needs no retraining and is exactly equivariant whenever those scores respect the symmetries. We instantiate it on the Neural Rule Inducer, a disjunctive-normal-form (DNF) foundation model that natively respects only example order. We restore the remaining symmetries through architecture, inference, and training. On synthetic stress tests, accuracy on the support labels stays stable at much larger schemas, and rule fidelity on fresh inputs remains above the unmodified model. On real data, accuracy improves most on larger schemas. The exported rule is exact on synthetic full-group tests and on schema-valid real-data tests. This is a mathematical property of the export rather than of a specific model, and we validate it empirically only on the NRI. Enforcing symmetry by construction turns this small-data pretrained model into a reusable, interpretable inducer that transfers to larger schemas.

We reexamine the problem of verifying Markov chains with respect to step-bounded reachability probabilities. Prevailing approaches rely on encoding the state-transition matrix using either explicit or symbolic representations. While these approaches are effective for sparse transition dynamics, they scale less favorably in the dense regime. Our insight is to cast probabilistic model checking of Markov chains as computations over dense tensors. This methodology enables the use of off-the-shelf compiler toolchains for optimized execution of these tensor computations on hardware accelerators. We prove the soundness of the methodology of mapping probabilistic model checking to tensor computations. We implement our approach in a tool called Tessa . Empirical evaluation shows that Tessa unlocks massive speedups over state-of-theart methods on selected benchmarks from the literature.

Artificial intelligence (AI) and Modeling & Simulation (M&S) are increasingly intertwined, reflecting converging research needs across both communities, rapid technological advances such as the rise of generative AI, and the growing availability of data and computational resources. This report provides a structured overview of the intersections of AI and M&S. The relationship goes both ways: AI can support, augment, or even replace components of simulation studies, while simulations can serve as data generators, training environments, and evaluation platforms for AI. We organize this landscape along the stages of M&S from model specification and input modeling to execution, experimentation, verification and validation, and output analysis. Selected studies at each stage illustrates how techniques such as Large Language Models have reshaped simulation practices, while highlighting limitations and open challenges. This report also provides a conceptual roadmap that helps readers navigate a rapidly changing ecosystem.

Optical microscopy of particle and fiber dispersions involves interpreting subtle visual cues influenced by specimen morphology, chemical composition, magnification, and illumination conditions. We introduce an artificial intelligence (AI) distillation framework that extracts semantically rich image embeddings from microscopy images using semantic anchors. A multimodal teacher combines each image's visual embedding with three text embeddings representing illumination modality, magnification, and specimen identity and morphology. Generated by LongCLIP's extended-context text encoder, this yields a 2304-dimensional block-structured teacher vector whose component blocks remain physically interpretable throughout training and inference. A student vision transformer (ViT) with a multi-layer perceptron (MLP) decoder is trained to reconstruct this teacher vector from the image alone, minimizing a mean absolute error (L1) loss that enforces coordinate-level fidelity to the teacher's block structure. A cross-entropy term over pseudo-classes derived from HDBSCAN clustering of the teacher embedding space acts as a collapse-prevention regularizer, enforcing inter-cluster separation without requiring contrastive negative mining. At inference, the student operates on image input alone, producing compact embeddings that recover the full semantic content of the teacher vector. The framework achieves approximately 80% pseudo-class validation accuracy and 75% Recall@1 on fine-grained specimen description labels under leave-one-out nearest-neighbor retrieval. These results demonstrate that semantic anchoring enables a vision-only student to acquire richer and more interpretable representations than image-only training, with direct applicability to retrieval, classification, and exploratory analysis of heterogeneous particle and fiber dispersions.

Testing Autonomous Driving Systems (ADS) requires realistic safety-critical scenarios, but collecting such data from real-world driving is costly and unsafe. This paper presents an automated pipeline that transforms safe driving scenes into safety-critical scenarios by combining computer vision, Large Language Models (LLMs), and Augmented Reality (AR). The system detects and tracks road users, extracts safety features including distance, velocity, motion direction, and Time-to-Collision (TTC), and assesses scene criticality. Safe scenes are modified by an LLM, which generates realistic collision-inducing objects and behaviors that are integrated into the original scene using AR. The proposed pipeline was evaluated on the nuScenes dataset, achieving 97.52% safety classification accuracy and successfully generating realistic scenarios such as pedestrian crossings, rear overtaking vehicles, and sudden-stop events. The results demonstrate an effective and flexible approach for automated generation of safety-critical scenarios to support the testing and validation of autonomous driving systems.

We read the geometry of semantic neighbourhood graphs as modal logic and give that reading a statistical form, in order to make precise the structural difference between human and machine-generated text. Texts are the worlds of a finite frame whose accessibility is the $k$-nearest-neighbour relation of a transformer embedding, and the symmetry, transitivity, Euclideanity and seriality frequencies of the two subcorpora are shown to be degrees of validation of the modal axioms $\mathsf{B}$, $\mathsf{4}$, $\mathsf{5}$, $\mathsf{D}$. Each degree is at once the proportion of instances of a rule that the subframe licenses in Negri's labelled calculus $\mathsf{G3.K}$ and a plug-in estimate of a population probability. A prompt-balanced comparison finds consistently higher artificial degrees for $\mathsf{4}$ and $\mathsf{5}$. We add a degree of groundedness and of situatedness, and recast the licensing reading in Cuconato's one-sided sequent-style tableaux, where each degree becomes a rate of set membership.

Attribution methods for genomic sequence models are commonly evaluated by how much of a known motif they recover, or by how a prediction degrades as evidence is deleted. Neither score is interpretable without the value it would take by chance, and neither is routinely reported against one. We show that this omission is not a matter of precision but of validity. The uniform chance level for contiguous motif overlap is \(L/(N-L+1)\); across 268 transcription factors in UniBind it ranges from 0.0118 to 0.0427, a 3.6-fold spread determined by motif length and window size alone. For two factors the bootstrap intervals of the chance levels themselves do not overlap, so their raw recovery rates are not comparable quantities. Correcting for this dissolves a published three-way classification of five factors: a factor reported as a resolution failure attains the second-highest corrected value, ahead of one of the two positive controls, and two reported as complete failures fall at or below chance. We further show that perturbation-based evaluation can fail its own precondition: for one factor a fully masked input still scores above the decision boundary, and the curve is not monotone in the number of masked positions, so the area under it is not a measure of faithfulness. We provide chance levels in closed form, a chance-corrected score, and two screens that run before any attribution is computed.

Open-ended strategic wargames are high-stakes LM-based social simulations: they model adversaries, institutions, escalation, plan brittleness, doctrine, and crisis response. Language models (LMs) are attractive because they can play agents, generate scenario branches, adjudicate ambiguous actions, and summarize lessons, but the same affordances make open-ended roles dangerous: model language determines both what an actor attempts and what becomes simulated reality. This position paper argues that no LM-enabled wargame should inform planning, doctrine, policy, or crisis response without an auditable safety case, and that the proper use of open-ended wargames today is to stress-test decision-influencing LM agents. We identify five failure modes: decision laundering, adjudication opacity, role collapse, escalation-through-adjudication, and failure of strategic imagination. Ordinary benchmarks cannot establish safety for these settings. Wargames can expose failures as stress tests; they are not themselves safety cases for consequential use.

Large language models (LLMs) prompt a recurring question in research computing: should classical methods like Naive Bayes (NB) be retired? We benchmark Complement Naive Bayes against zero-shot and few-shot LLMs spanning four model families and a 37x range in scale (27B to a 1T-parameter mixture-of-experts) across text classification tasks. LLMs dominate only in zero-data regimes (98.0% vs 88.2% on Amazon Polarity sentiment), and even that win is contamination-prone: on a low-contamination sentiment task NB beats the zero-shot LLM (81.7% vs 73.0%). However, once labeled data is available (e.g., AG News), NB reaches 89.1% accuracy, statistically indistinguishable from the zero-shot 27B LLM (89.0%) and better than the 397B frontier model (84.8%), at thousands of samples/sec on a commodity CPU. Fine-tuned DistilBERT reaches 90.6% but at far lower throughput than NB at batch size 1 (Table 2). Our measured GPU throughput analysis shows small-LLM batched inference is 40-486x slower than NB CPU inference (the multiplier depends strongly on the host CPU), exposing a structural gap bounded by memory bandwidth, with roughly two orders of magnitude lower energy per sample. For resource-constrained HPC practitioners performing text classification with labeled data, NB remains the optimal choice. We show the decision line is task-dependent (NB reaches LLM parity around $N \sim 10^4$ labels for topic classification, while zero-data sentiment favors the LLM at all N tested) and provide a Kubernetes Helm operator that automates model selection using configurable thresholds and verifiable Prometheus metrics.

We introduce AutoFyn, an agent harness inspired by the Expert Iteration algorithm, adapting a frozen model across many rounds by updating persistent state from verified reward signals rather than model weights. Each round begins from a fresh model session, and durable information is reintroduced only through explicit interfaces such as persistent memory files, reports, and repository state. Within a round, an orchestrator explores, plans and builds many alternative approaches with specialized agents, while a task-grounded verifier verifies the work and supplies an objective reward for measuring progress. This reward is distilled back into the persistent state, which updates the effective policy for the next round. In this technical report, we formalize this loop and describe its persistent state and verification interfaces. We then demonstrate its use in three domains, namely olympiad mathematics, data science, and cybersecurity. On the six fresh problems of the 2026 International Mathematical Olympiad, every model with room to improve scores higher under AutoFyn than in its provider's own coding agent. AutoFyn also built the top-ranked agent on the Spider 2.0 dbt benchmark, and has produced $16$ maintainer-confirmed vulnerability advisories in Next.js, MetaMask, pnpm, Warp, LiteLLM, Langflow, and Open WebUI.

Compound drought-to-extreme-precipitation (CDEP) events are recognized in climate science as a growing driver of extreme impact, but whether this recognition carries over into real-world early warning and post-event documentation is unknown, so a meteorologically real CDEP event may pass with neither advance warning nor any later record. Here we present CDEP Agent, an auditable LLM-agent framework that tests this mismatch directly by linking CDEP candidates detected from meteorological reanalysis to real-world hazard and impact evidence across sources with different spatial scales, temporal resolutions, and reporting conventions. Using California as a case study, we identify 408 candidate CDEP events from ERA5 observations during 2021-2025 and evaluate each against the U.S. Drought Monitor, NOAA Storm Events, and public webpages along five dimensions: antecedent drought, extreme rainfall, local impact, hazard-impact attribution, and explicit drought-to-rainfall linkage. Only 34.3% of candidates are corroborated on both hazard components, and just 1.5% are ever explicitly linked to their antecedent drought, indicating that most meteorologically detected CDEP events go undocumented and their compound nature almost never enters the record at all. Our framework gives climate scientists a way to test physical event definitions against what actually gets documented, and gives social scientists, economists, and disaster-response agencies a provenance-linked evidence base for compound events that current warning and reporting systems largely fail to capture.

Designing high-performance tactical wireless networks under realistic operational constraints gives rise to challenging combinatorial optimization problems, where the evaluation of candidate solutions relies on detailed physical and traffic-aware models. Although classical metaheuristics such as Tabu Search offer effective mechanisms for exploring large search spaces, their computational cost remains high because numerous candidate moves must be evaluated at every iteration. In this paper, we propose a data-driven framework that improves the efficiency of Tabu Search by learning to guide its move selection process. Rather than altering the neighborhood structure, our approach exploits the information contained in the search trajectories generated during the optimization process. At each iteration, we record both improving and non-improving edge-based transformations together with a set of descriptive features capturing the structural, geometric, and performance characteristics of the network. This information is used to train a Graph Neural Network (GNN) that predicts the impact of candidate moves on the objective function. The trained model is then integrated into the Tabu Search algorithm to rank candidate transformations according to their predicted quality, thereby reducing the number of costly objective evaluations while maintaining an effective exploration of the search space. Experimental results on synthetic benchmark instances demonstrate that the proposed learning-assisted Tabu Search notably reduces computation time while consistently producing higher-quality solutions than the standard algorithm. These findings highlight the potential of combining machine learning with metaheuristics by leveraging the implicit knowledge embedded in search trajectories, paving the way for more efficient solution methods for large-scale network design problems.

Large language models (LLMs) are increasingly used to generate peer reviews, prompting examination of their capacity for critical evaluation. This study evaluates two multimodal LLMs, Qwen2.5-VL-72B and Pixtral-Large-124B, as reviewers across 165 submissions to the 2026 International Conference on Learning Representations, a venue that postdates both models' training cutoffs. Manuscripts were presented to both models with author identities blinded, replaced with high-prestige affiliations, or replaced with low-prestige affiliations, and in either text-only or text-with-figure format. Additionally, 145 verifiably detectable errors were inserted into 55 manuscripts to assess error identification under natural and verification-oriented prompts. Across all manuscript groups, including rejected submissions, LLM scores ranged from 7.0 to 8.1, whereas human mean scores ranged from 3.4 to 6.8. The models detected 12.1% of the verified errors under natural prompting, and a one-sentence verification instruction increased detection to 22.2%; however, 78% of the errors remained undetected. Providing figures reduced error detection while increasing review scores. No visual error was reliably verified against its corresponding figure, and half of the text-only reviews described figures that were not provided. Author identity did not influence either review scores or error detection. LLM editorial decisions exactly matched those produced by simple score averaging.

We study predictive pretraining for scientific document representation using the discourse structure of papers. We propose SciJEPA, a citation-free framework that learns through asymmetric within-document prediction: title and abstract representations are used to predict method representations, and method representations are used to predict conclusion representations. Experiments on RELISH, high-influence citation, SciDocs, and cite prediction show that plain predictive training is viable but weaker than a controlled contrastive baseline using the same section pairs. Adding Sliced Isotropic Gaussian Regularization (SIGReg) substantially improves performance and narrows this gap. The effect of regularization is task-dependent: moderate SIGReg helps fine-grained ranking, while stronger regularization can weaken local alignment. We further show that different encoding branches support different retrieval regimes. These results position within-document predictive learning as a promising citation-free complement for scientific document representation, provided that embedding geometry is carefully controlled.

AI systems whose outputs inform real decisions, and increasingly consequential ones, require something that current documentation practice does not provide: a structured, inspectable representation of the knowledge they need to ground, contextualize, and reason about those decisions, ideally reviewed and signed off by a domain expert. Established documentation artefacts already capture important aspects of an AI system. Model cards describe how a system behaves, data cards describe what it was trained on, and system cards describe the risks of a deployed system. None of them addresses the layer between inputs and outputs, more precisely, the concepts a system holds, the relationships it models, and the patterns of reasoning it applies. For pattern-recognition tasks this gap is tolerable. For agentic AI, where systems act on their conclusions, it is the step that most often separates a promising proof of concept from an operational solution an organisation can rely on. This paper introduces the Knowledge Card, a structured artefact that captures validated knowledge about a single bounded concept in a form that experts can review, organisations can audit, and AI systems can reason over. For one concept, such as a specific failure mode, a compliance obligation, or a process decision, a Knowledge Card records the entities and relationships involved, the reasoning that connects them, the conditions under which that reasoning no longer holds, and the provenance of every claim, all grounded in a formal domain ontology and signed off by a domain expert. Initial prototype cards have been built in the energy and pharmaceutical domains. The schema is released as a public draft for community engagement.

In this research-to-practice paper we present a survey that can be used to assess students' AI knowledge. As the use of artificial intelligence (AI), including generative artificial intelligence (GenAI), has proliferated, so has the need to educate students about the topic. A range of AI literacy frameworks have been proposed, outlining the essential knowledge that students should have. Alongside, different ways of assessing AI knowledge have been developed. As yet, there is a lack of assessment instruments capable of evaluating multiple forms of student knowledge, including technical concepts, practical applications, and ethical concerns about AI use. In this article, we present a study implementing a comprehensive instrument to assess AI knowledge. The instrument combines measures from multiple scales to capture a range of literacy features and actual knowledge. We implemented the instrument in a higher education setting to assess its viability and usefulness and found that the instrument exhibited useful diagnostic capabilities and was able to identify common misconceptions among students. Although students performed well overall, there was a significant misunderstanding of how AI, especially GenAI systems, work. It also identified a lack of higher-level knowledge. The instrument is publicly available for use by others. We foresee its usefulness as a diagnostic that goes beyond understanding students' attitudes and perceptions of AI and GenAI use and tests multiple aspects of students' knowledge and conceptual understanding. This can enable the development of targeted instruction.

Patent prior-art retrieval is a recall-oriented search task over long and highly structured technical documents. Dense retrieval improves semantic matching, but single-vector representations may compress multiple technical components, functions, and constraints into a single embedding. We propose Sparse Coverage, an unsupervised semantic retrieval framework that maps local span embeddings to a sparse vocabulary of embedding-space centers. The centers are selected with a coverage-oriented k-center objective, and spans activate nearby centers to produce sparse representations compatible with inverted-index retrieval. Experiments on CLEF-IP 2013 show that Sparse Coverage matches or exceeds the document-level recall of strong dense patent encoders in several configurations, while remaining competitive for passage-level retrieval. By combining local semantic evidence with sparse inverted-index search, Sparse Coverage provides an effective first-stage retrieval approach for patent search.

This study proposes a ring-based SpatialTransformer to learn how building uses at different distances from a railway station interact to generate pedestrian flow. Concentric ring buffers at 100-meter intervals up to 800 meters were defined around 100 randomly selected stations in Tokyo, treating each ring as a spatial token. Self-Attention was applied to learn inter-zone interactions directly from data, without prior structural assumptions. GPS-derived walking trip counts served as the target variable and Geographically Weighted Regression as the baseline. Across 30 independent trials, the SpatialTransformer consistently outperformed GWR in predictive accuracy. SHAP analysis revealed that mid-to-outer distance zone features dominate pedestrian flow prediction, while features from the 0-100m zone contributed little. The attention matrix showed that each distance zone attends most strongly to spatially distant zones, demonstrating that pedestrian flow is regulated by structural interactions across the entire catchment area rather than by any single zone in isolation. These findings challenge the compact city assumption that station-proximate development maximizes pedestrian flow, and suggest that land use distribution across the full walkable catchment area deserves greater consideration in urban planning practice.

Large language models for code generation often produce incorrect solutions without reliable indicators of failure. We study whether uncertainty estimation methods developed for natural language transfer to code generation, and whether such signals can improve code generation via selective self-correction. We evaluate five uncertainty methods: mean token entropy, verbalized confidence, $P(\text{True})$, entropy ensembles, and semantic entropy probes, across three small code LLMs on HumanEval and BigCodeBench. We find that multi-sample $P(\text{True})$ achieves the strongest correlation with correctness, while all the other methods, including semantic entropy probes, yield only weak correlation. We then use these uncertainty signals to drive three self-correction policies: adaptive decoding, uncertainty-based regeneration, and verification-based regeneration. Our results reveal a stronger negative finding than anticipated: uncertainty-based self-correction fails to reliably improve Pass@1, degrading accuracy in 5 of 6 configurations across both benchmarks ($-3$pp to $-10$pp), and adaptive decoding degrades accuracy in 4 of 6 configurations. Only verification-based self-correction reliably improves Pass@1, with gains of $+6$ to $+26$ percentage points on HumanEval and $+8$ to $+20$ percentage points on BigCodeBench, scaling inversely with baseline strength. These findings replicate consistently across both benchmarks and suggest that cheap uncertainty estimators are insufficient on their own to improve code correctness, and that their practical value lies in serving as gating signals for costlier execution-based correction loops rather than as standalone substitutes for verification.

We introduce pico-type, a byte-level multi-head content classifier with approximately 1.5 million parameters that simultaneously predicts seven content properties from raw UTF-8 bytes in a single forward pass. Operating directly at the byte level – no tokenizer, no subword vocabulary, no pretrained embeddings – pico-type classifies coarse type (12 classes), modality (8), subtype (24), code language (62), text language (30), file MIME type (90), and risk flags (6-label multi-label: API keys, JWTs, passwords, emails, phone numbers, SSH keys). The architecture combines a learned byte embedding, three convolutional blocks with growing receptive fields, two bidirectional attention layers with rotary position encodings, and a statistical pooling layer feeding seven Matryoshka-style classification heads. Four tiered variants (tiny/small/base/pro) share the same trunk with sliced representations from 16 to 576 dimensions, yielding ONNX exports under 210 KB and CPU inference under 10 ms. Trained on a mixture of synthetic templates and real-world data (8709 GitHub code samples, 5000 Wikipedia articles), pico-type achieves 60.3 percent code language accuracy on The Heap benchmark (24 languages) and 98.2 percent text language accuracy on Wikipedia (30 languages) – improvements of +57 and +79 percentage points respectively over the synthetic-only baseline. Format-based heads (coarse, modality, subtype, file_mime, risk) maintain 100 percent accuracy on synthetic benchmarks. The model, code, and pretrained weights are released under Apache 2.0.

Photoplethysmography (PPG) is widely used in consumer wearables because of its low cost and ease of acquisition. However, unlike electrocardiography (ECG), PPG measures peripheral pulse dynamics rather than cardiac electrical activity, limiting its ability to predict cardiac conditions that rely on ECG-specific morphological cues. Existing methods attempt to bridge this gap by reconstructing ECG signals from PPG signals. However, this inverse mapping is inherently ill-posed, and faithful waveform reconstruction does not necessarily translate into improved downstream performance. To address this challenge, we propose P2E-VQ, a retrieval-augmented framework that replaces ECG waveform reconstruction with ECG-linked representation retrieval. Specifically, P2E-VQ converts PPG patches into discrete tokens and retrieves ECG-linked information from a memory bank constructed exclusively from the training data. This process augments PPG representations while requiring only PPG signals during inference. Extensive experiments on five public datasets covering six downstream tasks, including clinical endpoint prediction and affective state recognition, demonstrate that P2E-VQ consistently outperforms pretrained baselines under a unified frozen-feature linear-probing protocol.

Federated Learning (FL) is a collaborative paradigm that enables multiple devices to train a global model while preserving local data privacy. A major challenge in FL is the non-Independent and Identically Distributed (non-IID) nature of data across devices, which hinders training efficiency and slows convergence. To tackle this, we propose Federated Impurity Weighting (FedImp), a novel algorithm that quantifies each device contribution based on the informational content of its local data. These contributions are normalized to compute distinct aggregation weights for the global model update. Extensive experiments on EMNIST and CIFAR-10 datasets show that FedImp significantly improves convergence speed, reducing communication rounds by up to 64.4%, 27.8%, and 66.7% on EMNIST, and 44.2%, 44%, and 25.6% on CIFAR-10 compared to FedAvg, FedProx, and FedAdp, respectively. Under highly imbalanced data distributions, FedImp outperforms all baselines and achieves the highest accuracy. Overall, FedImp offers an effective solution to enhance FL efficiency in non-IID settings.

Prediction uncertainty is a widely adopted metric for quantifying model confidence, with downstream applications spanning model explanation, data selection, and prediction rollback. Despite its demonstrated utility, the potential of uncertainty quantification to enhance code generation in large language models (LLMs) remains largely underexplored, raising a critical question: to what extent can uncertainty serve as an effective signal for improving LLM-based code generation? To answer this question, we study uncertainty-aware rollback decoding, an inference-time strategy that uses uncertainty signals to identify unreliable generation regions and roll back to earlier valid prefixes without retraining the model. We evaluate this framework on seven code LLMs, five code generation benchmarks, and eight token-level uncertainty signals under a unified decoding setup. Our results show that the complete rollback framework improves over equal-budget restart across the evaluated benchmarks and model settings, with gains of up to 0.26 in pass@1 and 0.35 in AvgTestPassRate on functional code generation benchmarks, and an absolute improvement of up to 6.4% in Patch-Aligned Safe Rate on Dsec-Python. Among the evaluated signals, information-theoretic measures such as token entropy and negative log-likelihood show the most favorable overall trend, frequently achieving the best or near-best results on standard benchmarks. A component-controlled ablation further shows that feedback-guided rollback provides the main improvement, while uncertainty localization provides an additional gain when checking, budget, rollback, and branch decay are held fixed.

The development of 0.1$^{\circ}$ global weather forecasting models based on machine learning (ML) is constrained by the limited availability of high-resolution data, as decades of reanalysis are only available at 0.25$^{\circ}$ resolution. While existing approaches fine-tune 0.25$^{\circ}$ forecast models on limited 0.1$^{\circ}$ samples, we show that this transfer is hindered by the irreversible information loss inherent in coarse-resolution forecasting. Therefore, we propose BaguanHR, a framework that shifts the focus from transferring models to transferring data. We first show that super-resolution (SR) has lower conditional entropy and input amplification than forecasting, making it a more robust vehicle for resolution transfer. By leveraging this advantage through variable-wise SR, we synthesize extensive 0.1$^{\circ}$ data from ERA5. BaguanHR's performance on the synthetic-plus-real dataset exceeds both ML-based methods and IFS-HRES, achieving superior performance across over 85% of the lead times within 72 hours. Furthermore, our findings highlight a power-law scaling effect, as a twofold increase in data reduces RMSE by 4.6% for 72-hour forecasting and 4.9% for 120-hour forecasting. Our results demonstrate that scaling high resolution ML-based forecasting is primarily a data bottleneck, and that variable-wise super-resolution provides a simple yet general solution to unlock long coarse-resolution reanalyses for high-resolution training.

Agent evaluation relies almost entirely on outcome metrics such as success rate, which capture whether an agent succeeds but not how consistently it behaves. We argue that behavioral consistency across tasks is a distinct and measurable property, and we introduce the Behavioral Consistency Metric (BCM) to quantify it. BCM trains a model to predict task success from behavioral features of agent execution traces, derives a per-trajectory feature-attribution vector, and measures the mean pairwise similarity of these vectors within an agent system. Across roughly 9,000 trajectories from six language model agents on software engineering tasks, our central finding is that cross-task and within-task consistency are distinct axes that can diverge: some systems are locally reproducible, behaving similarly on repeated attempts at one task, yet globally fragmented, with no stable strategy across different tasks, while others are consistent at both scales. Prior work measures only same-task reproducibility and so cannot observe this separation. We further find that consistency is not reducible to success rate, since systems with comparable success can differ sharply in consistency, and that the frontier-versus-open-source consistency gap persists under a within-task control that holds task difficulty constant. We position BCM as a process-level reliability signal that complements outcome metrics, and we are explicit about the conditions under which it is meaningful.

Coverless image steganography (CIS) synthesizes a stego image rather than modifying an existing cover image, enabling authorized recipients to reconstruct the original secret image from the stego. Existing diffusion-based CIS methods can generate natural-looking stego images but preserve substantial visual similarity to the secret image. This resemblance risks exposing structural and semantic cues, giving rise to security vulnerabilities that cannot be evaluated solely via recovery fidelity. Achieving substantial visual dissimilarity between the secret and stego images without compromising stego quality and recovery fidelity remains challenging. To address this issue, we propose InvCISD, an invertible diffusion framework that couples the latent representations of the secret and an irrelevant reference image with an invertible network called LIMNet. We first train LIMNet in diffusion latent space, followed by end-to-end fine-tuning of the entire network, i.e., LIMNet integrated diffusion inversion and generation modules. Experiments demonstrate that the proposed method substantially reduces secret-stego visual similarity, improves stego quality, and retains satisfactory secret reconstruction quality. Our further investigation shows that all evaluated methods are highly detectable by the CIS-oriented steganalysis model, indicating that resistance against targeted steganalysis constitutes a critical direction for future CIS research.

Heterogeneous model fusion seeks to combine models that differ in tasks, initializations, architectures, or scales. We study an underexplored cross-scale setting: improving a small recipient language model with a stronger donor despite substantial architectural mismatch. We ask whether useful capabilities can be transferred without explicit neuron-wise semantic alignment. Building on the observation that truncating a large model to a smaller architecture and injecting it with a tiny mixing weight can already improve the recipient, we propose Activation-Prune-Merge (APM), an activation-guided framework for cross-scale fusion. APM constructs task-conditioned activation maps on the donor, selects salient layers, hidden dimensions, attention heads, and MLP neurons to prune it to the recipient architecture, and injects the resulting donor slice into the original recipient using a micro interpolation coefficient. This formulation treats the donor as a source of concentrated functional components rather than requiring precise structural transplantation. Across 16 benchmarks spanning reasoning, mathematics, code generation, instruction following, and classification, APM improves the overall average accuracy from 55.5% to 60.6% over the original 3B recipient. RTE accuracy increases from 64.3% to 82.3%, QNLI from 52.3% to 65.7%, and BoolQ from 70.8% to 79.2%. Analyses of injection ratios and sequential multi-stage fusion further suggest that activation-guided extraction improves the quality of the transferable donor slice while preserving the small-ratio fusion regime. These results provide evidence that cross-scale heterogeneous fusion can succeed without explicit semantic alignment when the donor contribution is sufficiently concentrated and carefully selected.

As large language models serve ever more requests, cumulative inference cost is growing relative to the one-time cost of training. In typical serving, prompt prefill runs in parallel and is compute-bound, whereas autoregressive decode is sequential and memory-traffic-bound. Conventional width or depth scaling raises both costs together, since every added layer is evaluated in both phases and enlarges the weights read at each decode step. We instead ask whether additional learned computation can be allocated to continuation prediction while preserving prompt-wide primary computation and a single KV cache. We realize this with the Decode-Branch Transformer. Its primary path alone processes the prompt and writes the KV cache; the decode branch is omitted during prefill and activated only from the final prompt position onward, adding continuation computation without writing state or affecting the primary path. The paths share attention, MLP, and output matrices, using separate token embeddings with lightweight coupling. Grouped decode reuses loaded weight tiles and the primary KV cache across both paths, so the added arithmetic does not proportionally increase dominant memory traffic or decode latency. Across matched-token comparisons, Decode-Branch achieves lower validation loss across architectures and data settings. In MoE models, the primary and branch expert fan-outs become independent knobs for trading prompt cost, decode cost, and predictive quality. We study two expert-allocation regimes, holding prefill or decode computation fixed, and expose a prefill-decode-quality trade-off enabled by phase-specific expert allocation.

When the context window is under pressure, LLM systems compact prior context to continue ongoing tasks. We study Side Constraints (SCs), user-issued instructions such as "do not delete any emails until I confirm," that constrain an LLM's behavior across subsequent turns but can be silently dropped during compaction. In an inbox experiment, an agent deleted emails without approval in 30/100 sessions after compaction, compared with none before compaction. To systematically evaluate SC preservation, we introduce CONPACT, an evaluation suite measuring semantic retention and downstream compliance across three scenarios: multi-turn chat, multi-turn tool use, and long-horizon research. The evaluated compactors retain only 17% of injected SCs on average. Loss occurs across compactors and settings, with retention varying by compactor, context length, SC framing, and injection location. We propose an SC extractor that maintains constraints separately from the task summary and appends them to the compacted context, achieving over 90% retention across all three scenarios without modifying the compactor or LLM. Code and data are available at https://github.com/ZhiqiEliWang/compaction-integrity.

Deploying LLM agents into industrial recommender operations exposes a three-way tension we frame as the autonomy-determinism-efficiency trilemma: general autonomy (interpreting operator intent, generating glue code zero-shot), industrial determinism (schema-conforming feature extraction, non-crashing A/B, zero compliance-path hallucination), and end-to-end efficiency. Any two can be maximized against the third. We present RecSys Factory, an LLM-agent platform deployed for 78 days across three heterogeneous Tencent recommender business lines. The design principle is autonomy at decision points, not over pipelines, made concrete through three deconstructions that each discharge one vertex of the trilemma. Runtime is deconstructed into three host-emitted event sources (Claude Code Stop hooks, corporate-IM webhooks, workflow scheduler APIs): the platform carries no long-running daemon during the wait phase and consumes zero CPU during the 94% of wall-clock spent waiting on Spark or GPU jobs. Capability is deconstructed into a 29-file skill ecosystem (8,971 lines of SKILL.md) whose per-skill pitfall tables mechanically compile into a 400-entry PitfallStore, confining autonomy to bounded typed decision surfaces inside pre-committed pipelines. Deployment spans three business lines with disjoint label semantics, A/B layer topologies, and operator personas; an onboarding-time compression is observed on two of the three and is reported as a case-study observation, not a generalization claim, and not measured against a controlled pre-platform baseline. The human is retained at the diagnostic-versus-execution boundary via a human-in-the-loop card protocol, deployed as an audit-trail primitive (schema-validated, idempotent, replayable) and reported from an 8-day 16-run pilot. Across the 78-day window the platform recorded 1,624 CLI-tool dispatches at a 78.6% aggregate success rate.