Science & Mathematics

Modeling cascading failure across interdependent critical infrastructure requires either a composable physical simulator per domain or a topological abstraction that discards quantity awareness. Simulators are costly, data-hungry, and require detailed calibration per domain; topological abstractions stay data-parsimonious while missing quantity-based behaviors that govern real dynamics. We present, instead, a knowledge-driven propagation architecture in which heterogeneous service dependencies are identified by categories: a general logical propagation resolves multi-dependency and redundancy structures purely from how nodes are categorized, while quantity-aware categories plug in dedicated mechanisms - here, a max-min fair-share flow module for water distribution. Both share one discrete functionality scale and a monotonic update algorithm guaranteeing convergence in finitely many rounds. We assess the flow module by comparing its results with those we get from preexisting pressure-driven simulation with WNTR (Water Network Tool for Resilience) on eight realistic networks across four failure families (clustered damage, capacity-targeted attack, source failure, tank loss). Under capacity-targeted attack and source failure a topological-reachability baseline misses 46% of the truly critical junctions, where the flow module misses 10% and 8% at an average critical-class precision of 0.73. Supplementary materials, code and results are available at: https://github.com/cascade-platform-org/CASCADE

Facing imbalanced binary classification, we aim at maximizing the recall while limiting the proportion of false detections in the minority class. For doing that, the present work studies the effect of several rebalancing strategies, from the uniform subsampling of the observations from the minority class, to a more involved local data augmentation scheme. To be more specific, we establish that only rebalancing class proportions cannot modify the classification performance, whereas local data augmentation is proved to have an impact on the final results. The local augmentation scheme is introduced by means of a two-step procedure. The first step randomly subsamples anchor points among the training observations. These anchor points are used in the second step, for drawing new augmentation data following an ''augmentation kernel'' to be specified. This enables us to decouple the impact of the anchor points subsampling from the one of drawing augmentation data accounting (or not) for the local geometry of the support. The experiments on synthetic data illustrate the practical performance of several strategies for subsampling anchor points combined with either a Gaussian or a diffusion-based augmentation kernel. Compared to SMOTE-like data augmentation procedures, a better performance is achieved when subsampling in the close neighborhood to the classification boundary combined with a Gaussian augmentation kernel. Furthermore in specific settings where the geometric structure of the support matters, the experimental results show that using a diffusion-based augmentation kernel outperforms the Gaussian one.

Deep research agents (DRA) are commonly used for open-ended synthesis tasks where closed-world QA benchmarks are poorly suited to characterize relevant aspects of performance. We study the self-consistency of one commercial deep research system under repeated identical prompting, representing each report with the set of named entities it mentions, categorised by a custom schema, and the set of citations it retrieves. Across repeated runs of a single open-ended cybersecurity related prompt, we quantify pairwise overlap to characterize how stable the retrieved and synthesized output structure is under repeated identical prompts. In ten runs, the pairwise entity Jaccard results in a median of 0.12 and the pairwise citation Jaccard is below 0.02, with discovery curves that do not visibly saturate. The magnitude and structure of (dis)agreement between entity sets varies by entity class, which suggests that self-consistency in a DRA is better understood as a function of the underlying entity space than as a stable property of the system.

The regulation of emerging technologies is often conceived as a strategy of catching up, because legislators arrive too late. Yet gathering signals, constructing scenarios and adapting rules to an anticipated future are fundamental tasks for those entrusted with governance, including the governance of rules themselves. The literature on so-called future law has already shown that anticipation means building institutional capabilities rather than predicting outcomes precisely (anticipatory governance). Taking this premise as its starting point, this article proposes to understand regulation as an exploration of the transformations that may occur. The Ruliological Framework for Anticipatory Regulation distinguishes rules, transformations and trajectories, assessing alternatives and possible developments. Within this approach, computational irreducibility is adopted conditionally, not as a demonstrated property of legal systems. The Quantum Europe Strategy provides a case study and further grounds for reflection.

We present SNO-PI, the top-performing submission on SNOMED International's SNOMED CT Entity Linking Benchmark as of 18 October 2026. SNO-PI achieves a macro character IoU of 0.5387, outperforming the organisers' reference system (0.4582; based on the 2024 challenge winner's approach) by 18% relative. Linking clinical mentions to SNOMED CT is a zero- and few-shot problem over a very large label space: the 272 annotated MIMIC-IV discharge summaries cover only 6,595 of the roughly 219,000 eligible concepts, and a third of those occur only once. This motivates a design built on transfer from pre-trained models and human-curated knowledge. SNO-PI combines high-recall span proposal, SapBERT candidate retrieval and listwise ensemble reranking with three BERT-family encoders, each further pre-trained with masked language modelling on unlabelled MIMIC-IV notes and then on SNOMED CT synonyms and hierarchy before task fine-tuning. We find that curated knowledge helps most on concepts never seen in training, yet these concepts remain the main source of error.

Rubric-based reinforcement learning (RL) trains policies for open-ended generation with natural-language criteria as rewards, but these rewards can become ineffective as training progresses. We identify two failure modes. Under non-discriminative rewards, scores collapse to uniformly low or high values due to insufficient exploration or a mismatch between task difficulty and policy capability. Under spurious rewards, higher scores do not correspond to better responses. Both reflect a training configuration whose task sampling, exploration, and evaluation stay fixed as the policy evolves. We introduce HarnessRL, a training-time harness that co-evolves with the policy through three interfaces for task sampling, rollout guidance, and rubric criteria. HarnessRL maintains a training-time memory from which a harness evolver retrieves within-task history and cross-task evidence. Using this evidence, the evolver attributes each reward failure to the interface whose update can address it. It then runs a proposer and just-in-time critic loop in which the critic accepts a candidate update only after measuring its effect on the current frozen policy, returning rejected candidates for revision. We also curate Rubric-Mix-10k, a five-domain training and evaluation suite for open-ended RL across medical, science, writing, role-play, and instruction-following tasks. On Rubric-Mix-10k, HarnessRL outperforms the strongest baselines by 2.8 to 3.1 points across both 4B and 9B model scales and nearly halves the training steps to its best checkpoint. Code and data are available at https://github.com/Tongyi-ConvAI/Qwen-Character/tree/main/HarnessRL.

Sudden cardiac death mostly originates from post-infarct ventricular tachycardia. This reentry is sustained by propagation around scar tissue, but its risk remains difficult to predict from infarct imaging alone. While in silico induction studies can reveal these mechanisms, exploring many substrates and pacing locations remains computationally expensive. We propose a graph attention classifier for early prediction of sustained ventricular reentry from simulated cardiac electrophysiology. Each simulation is represented as a graph with early nodal voltage samples and edge attributes encoding geometry, stimulation frequency, tissue labels, amplitude, and diastolic interval. These descriptors are computed only from the first two paced activations, before sustained reentry is observed. On a balanced independent test set of 1000 simulations, the classifier achieves 94.6% accuracy, 95.4% sensitivity, 93.8% specificity, 94.6% F1-score, 0.987 ROC-AUC, and 0.985 PR-AUC. Ablation experiments show that diastolic interval is the most critical descriptor. Qualitative attention inspection suggests that different heads focus on border-zone, stimulation location, and early propagation path. These results support graph attention as a fast early-prediction approach for large-scale in silico ventricular reentry studies.

While VideoLLMs have made significant progress in understanding short video benchmarks, they often struggle to summarize long-form ones. This is mainly due to performance degradation caused by long-context prompts, the inherent difficulty of preserving temporal continuity and a lack of semantically rich concepts in event descriptions. To tackle these issues, this paper introduces the first recursive Video Language Summarization Model (RVLSM) designed for video processing and tailored to long-form video-to-text summarization. RVLSM recursively decomposes videos into semantically coherent segments and delegates chunk-level reasoning to VideoLLM sub-models and external tools. It employs (1) Semantic Temporal Decomposition (STD), which preserves temporal coherence and continuity through a scene-aware chunking strategy that dynamically partitions videos according to semantic boundaries, and (2) Adaptive Multimodal Tool Calling (AMTC), which ensures semantic richness of the summary via a routing mechanism that selectively invokes auxiliary modules such as action recognition, optical character recognition, automatic speech recognition, and object detection. Experiments on the VideoXum and MixTyVSum benchmarks show that RVLSM consistently improves over zero-shot VideoLLM baselines across multiple model settings and performance metrics, and is competitive with a state-of-the-art fine-tuned model for video-to-text summarization. Importantly, the named entities in the RVLSM's summaries align more closely with those detected in the ground-truth summaries than those detected by baseline and state-of-the-art methods, confirming the saliency of the event descriptions it generates.

Achieving high sign-off performance in advanced ICs increasingly depends on layout-stage optimization under realistic physical conditions. While recent efforts have moved toward synthesis-layout fusion, most existing works still focus on bringing estimated physical information earlier into synthesis, whose optimization does not fully correlate with real layout-stage behavior. Meanwhile, layout-stage optimization remains largely restricted to a fixed logic structure, relying mainly on buffering and sizing. This fundamentally constrains the optimization space under realistic physical conditions. To address this, we present E-Layout, an ML-driven framework for post-placement layout restructuring. E-Layout jointly optimizes logic transformation and physical implementation across multiple subcircuits under design-level timing feedback. This enables the selection of the most timing-beneficial set of disjoint restructured subcircuits to form the optimized circuit. Specifically, it first constructs equivalence-preserving logic alternatives via e-graph equality saturation, then couples logic selection with physical realization through a graph-based retrieval-augmented generation (RAG) model, and finally refines these local logic-physical decisions with reinforcement learning under design-level timing feedback. Evaluated on 9 diverse designs, it achieves on average 6% TNS improvement and 2% WNS improvement, while even reducing area by 1%, over an industry-leading commercial physical design flow.

We study whether geometric relationships between conversation turns, memory texts, and queries can predict answer correctness before generation. We evaluate four memory implementations with dense or hybrid retrieval and one observation-summary comparator. We embed these texts in a shared space. A logistic classifier uses 25 features describing query proximity, memory-conversation correspondence, and conversation structure. On LongMemEval-small, pooled AUROC is 0.640 over 2,367 question-system pairs, below a question-difficulty baseline of 0.655. Performance varies by question type: geometry reaches 0.841 on single-session-assistant questions, but system identity alone reaches 0.826. After controlling for difficulty and identity, geometry adds no statistically established pooled improvement (delta = +0.002, 95% CI [-0.010, 0.014]). A fixed-store context-selection experiment finds higher accuracy with proximity-ranked memories than with random subsets, consistent with ordinary retrieval relevance. On LoCoMo, nearest-turn similarity has modest predictive value for single-hop questions and near-chance estimates for several other categories. Training uses correctness labels; prediction uses pre-answer features and no additional LLM generation call. The results characterize associations between memory representations and correctness, rather than establish a reliable deployment diagnostic. In particular, embedding proximity does not verify fact coverage or distinguish memory-construction, retrieval, and generation errors.

We develop R2-GP, an R2D2-type prior for Gaussian process (GP) generalised additive models (GAMs) that calibrates explained variance ($R^2$) for Matérn covariance kernels. The function variance for standard Matérn kernels depends jointly on the effect magnitude and length-scale. As such, smooth and rough components with identical amplitudes contribute very different amounts of variation. With independent priors on component amplitudes, this variation results in variance overestimation, especially in higher dimensions. We overcome this by taking direct control over each component's expected contribution to prior explained variance. We do this by rescaling each component based on its length scale and observed inputs, and modelling each rescaled variance as a share of the function variance induced by a global $R^2$ prior. This allows R2-GP to scale well to many components. To also scale to large sample sizes, we extend R2-GP to Hilbert space approximation methods, a framework for kernels exploiting their spectral decompositions for fast and efficient inference. Our simulations and real-world case studies demonstrate superior generalisation and parameter recovery over independent variance priors.

Large language model (LLM) agents now trade real capital using news as evidence, yet they cannot reliably tell fabricated headlines from real news. Because they read the same feeds under near-identical policies, a single fabricated post can push them into the same trade at the same moment. We present a systematic attack benchmark for information injection against LLM trading agents. Unlike prompt injection, which targets the model's instructions, information injection targets the evidence the agent trusts. The benchmark spans five injection vectors and seven event categories, evaluated over ~10,000 decisions on 15 U.S. stocks. Five findings emerge. (i) Prompt injection achieves 14% attack success rate (ASR), but fabricated news reaches 62% and two 30-word templates reach 100% on all 15 stocks, inverting the usual LLM-security intuition. (ii) For fact-poisoning templates, surface form matters more than content, since removing credibility cues such as a named source or a "BREAKING" tag drops ASR from 100% to 0% while the fake claim remains identical. (iii) Success depends on the agent's state, with a fake indicator panel scoring 91% against an agent about to hold but 12% against one about to buy. (iv) Attacks generalize across model vendors and remain effective when blended, unlabeled, into real news feeds. (v) Most attacks evade standard defenses, since model scaling stops shallow vectors but not fact poisoning, peer debate fails, and only hierarchical aggregation reduces ASR.

Statistical privacy audits are essential for verifying the theoretical privacy guarantees of differentially private (DP) mechanisms. Auditors typically use Membership Inference Attacks (MIAs) to estimate a lower bound on the true privacy leakage of DP mechanisms at a specified confidence level, which should not exceed the theoretical privacy upper bound guaranteed by DP. However, we found that many auditing implementations in prior work introduce a selection bias when running MIAs, which leads to statistically invalid and overestimated lower bounds. Because most audits are not tight, this bias has thus far gone unnoticed: the estimated lower bounds remained well below the theoretical upper bounds. Nevertheless, accurately evaluating this lower bound is critical, as a flawed statistical assessment could lead to false alarms on correct mechanisms or an incorrect evaluation of the guarantee's tightness. This work employs tools from the statistical literature to correct this bias. We present an overview of the different correction methods and assess their respective advantages. Furthermore, we perform a comprehensive re-analysis of results returned by prior audit implementations and show that without statistical correction, these audits violate the theoretical upper bounds up to four times more often than their claimed confidence level implies. With correction applied, the resulting lower bounds are statistically valid, and violations occur at no more than the nominal rate.

Hidden randomized rounding couples the privacy loss of successive integer releases. We study a rational-feedback mechanism under predictable signed-unit participation and full observation of every physical output. A preparation's lifetime connects three results. First, preserving complete carries until an observed echo synchronizes the worlds gives exact finite-law point compilation, with support-width exponent $3C-2$ at every fixed budget $C\ge2$. Second, releasing commitments earlier gives a monotone hierarchy whose exact endpoint, strict gaps and numerical brackets quantify calibration loss. At one fixed discrete Gaussian, a loss allowance of 0.09 requires all four commitments. Third, legal regeneration yields an unrestricted KL rate with a square-root testing remainder and a complete asymptotic frontier under joint physical-time and participation constraints. At fixed rational parameters, finite-window occupation programs approximate that frontier to error $2^{-b}$ in polynomial bit time in $b$. Verified countable-transcript semantics and independently checked finite certificates support these analyses. A matched static workload illustrates the resulting design tradeoff: a structured finite integer matrix improves error and fresh randomness, while the scalar feedback instance retains fewer noise-state and source-table bits.

Fine-tuning large language models (LLMs) on downstream tasks often weakens their previously aligned safety behavior, creating an inherent trade-off between task adaptation and safety preservation. Existing approaches attempt to mitigate this issue by identifying safety-relevant neurons, layers, or update directions in the original parameter space. However, safety-related information is often distributed across many directions in this space, causing downstream updates to overwrite parameters that encode safety behavior. We propose WSR-Tune, a Weight Space Reparameterization-based fine-tuning framework that preserves safety during downstream adaptation. Our approach constructs a safety-conditioned basis and reparameterizes the weight matrices such that safety-relevant information becomes concentrated in a small subset of directions. We then identify safety-critical directions and freeze them in the reparameterized space, while updating the remaining complementary directions for downstream adaptation. By explicitly structuring the parameter space to disentangle safety and task information, our method reduces interference between objectives and enables simultaneous preservation of safety and improvement in downstream performance. Experimental results show that our method achieves a more favorable trade-off between downstream performance and safety retention, demonstrating its effectiveness for reliable LLM fine-tuning.