Different ontologies with conflicting and overlapping concepts cause havoc in the design, develop, and deploy of ontology-driven applications. In this work, we propose a comprehensive ontology interoperability framework for industry-scale ontology-driven applications. The framework employs three state-of-the-art semantic techniques in different phases of the ontology engineering (OE) life cycle: ontology design patterns (ODPs) in the design phase, ontology matching and versioning (OM\&OV) in the develop phase, and ontology validation (OVA) in the deploy phase, to achieve better ontology interoperability.
Md. Robiul Islam Niloy, Abu Sayed Md. Mostafizur Rahaman
原文摘要
Large language models (LLMs) deployed in safety-critical settings must resist adversarial inputs designed to elicit harmful outputs. We present a systematic empirical evaluation of five adversarial jailbreaking attack families---direct requests, roleplay-based jailbreaking, hypothetical framing, instruction override, and obfuscation---across seven open-weight LLMs served locally: Llama-3-8B, Llama-3.1-8B, Llama-3.2-3B, Mistral-7B, Qwen2.5-7B, Gemma2-9B, and Phi-4-14B. We further evaluate four defense mechanisms---keyword blacklisting, self-reminder prompting, paraphrase-based sanitization, and perplexity-based filtering---against two high-ASR models (Mistral-7B and Llama-3.1-8B) on a 20-prompt subset. Across 250 prompts per model, a content judge counted obfuscation attacks as harmful (HARMFUL or PARTIAL responses) at a pooled rate of 62.3%, the highest of any family, although only 7.1% of responses were rated HARMFUL. Hypothetical framing was the highest family for three of seven models. Obfuscation succeeded at nonzero rates on all seven. The content judge has not yet been validated against human labels. We find that model size does not significantly predict adversarial robustness (Spearman Rho =-0.64, p=0.12, n=7). A rule-based judge suggested a large gap between the two Llama-3 generations (16.8% and 40.8%); the content judge does not (17.2% and 18.4%). Self-reminder prompting gave the largest reduction, cutting ASR from 50% to 10% on Mistral-7B and from 45% to 5% on Llama-3.1-8B (40 percentage points on each model, against matched no-defense runs, scored with the rule-based judge; these defense results are provisional because that judge has known defects). Our code and datasets are publicly available at https://github.com/Mdniloykhan/jailbreak-eval-7llms.
When language models revise a decision after another agent disagrees, neither reversal counts nor final accuracy shows whether they are fixing mistakes or abandoning correct answers. We introduce a controlled two-pass audit that holds the question, the evidence, and the model's initial answer fixed and varies only the revision context: neutral reflection, anonymous disagreement, or disagreement attributed to a domain expert. For binary decisions the accounting is exact: revision raises accuracy only if it is selective enough, reversing wrong answers more than $a/(1-a)$ times as often as correct ones, where $a$ is initial accuracy (nine times at 90%). Across five models and 1,989 evidence-grounded decisions in medicine, science, and law, no model clears this bar under expert disagreement, and accuracy falls by up to 16.1 points. The expert label, a prestige cue, adds reversals for four of the five models, yet its effect on accuracy ranges from a 5.2-point loss to a 2.4-point gain, so reversal counts are a poor proxy for harm. With identical arguments from both sources, the label still raises reversal for three proprietary models and in a pooled analysis of four newer models. Models also respond to stated source reliability. Revision pipelines should therefore be evaluated with paired, conditional metrics, per model and protocol.
Yingbin Jin, Ji'an Wang, HaoYang Chai, Hanjun Luo, Li Hu, Xinfeng Li, Huadi Zheng, PeiZhao Hu 等 10 位作者
原文摘要
Mobile GUI agents read the screen of a phone and operate apps on a user's behalf to complete multi-step tasks such as online shopping or messaging. Because the agent acts on whatever the screen shows, an attacker can embed malicious instructions into ordinary UI content and redirect its actions. We refer to this threat as environmental injection attack (EIA). Existing benchmarks measure this risk by pinning hand-crafted payloads to a fixed step of a fixed task. This approach faces two limitations. First, each payload is bound to one task in one app and does not transfer without manual redesign. Second, agents given the same task may follow different routes, so an agent that never reaches the injected page is recorded as safe without being tested. This insufficient test coverage can cause fixed-point benchmarks to underestimate the security risk of deployed mobile agents. In this work, we propose MobileProbe, a trajectory-aware red-teaming framework that evaluates mobile agents adaptively. MobileProbe discovers attacker-controllable content surfaces through automated pentesting, localizes vulnerable decision points along the agent's trajectory, and iteratively refines context-aware payloads through a surrogate-filtered evolutionary loop. We construct a benchmark of 83 long-horizon attack instances spanning 12 mobile applications. Each is annotated with a Confidentiality, Integrity, or Availability (CIA) attack goal to cover the principal categories of security threats facing mobile agents. On three representative agent architectures, MobileProbe attains higher attack success rates than human-written, zero-shot LLM, and prior automated red-teaming baselines adapted to the mobile domain. MobileProbe can serve as a pre-deployment auditing tool to systematically uncover vulnerabilities and inform security hardening of mobile agents.
Nikos Salamanos, Paula Gheorghiade, Tom Brughmans, Michele Coscia
原文摘要
Collecting data, including network data, is subject to measurement error. Network backboning is the process of statistically validating each edge in a network to estimate the probability that the edge is born of noisy measurements. The current backboning literature provides approaches in two classes: structural and statistical. The classes differ in many respects, but they share one commonality: the test used to determine whether to filter out noisy edges is data-agnostic; it does not take into account what the network is representing. However, the nodes and edges of all non-synthetic networks correspond to specific real-world entities, whose dynamics determine how the edge-generating process operates. In this paper, we propose the first in a new class of network backboning algorithms: a theory-driven backboning approach that allows us to encode known dynamics into the backboning process. The resulting backbone embeds the encoded theory and produces a network compatible with it, rather than one that obeys a data-agnostic network process, as in the structural and statistical backboning classes. We specifically focus on the case of projecting and backboning a bipartite structure. We do so by allowing users to specify an Agent-Based Model for the two types of agents (nodes) in the network and by performing Simulation-Based Inference to fit the observed bipartite structure and, as a result, reconstruct the true relationships between the two classes of agents. We test our framework both on synthetic data -- where we have perfect information about the edge-generating process -- and in a case study using archaeological data on amphorae retrieved from different archaeological sites. Our approach improves latent-edge recovery over conventional backboning methods in synthetic experiments and reveals interpretable archaeological patterns despite imperfect model fit.
Temporal link prediction is critical for understand- ing dynamic interactions in evolving networks, informing appli- cations ranging from social network analysis to recommendation systems. Existing methods typically rely on raw temporal graph data, often neglecting explicit edge-change information that can significantly improve predictive performance. To address this gap, we propose edge-gradient reinforced learning (EG-RL), a framework that explicitly encodes edge-gradient events (edge births and deaths) through dedicated node and link encoding modules. EG-RL uses a policy-driven reinforcement-learning mechanism to dynamically select and augment the most infor- mative edge snapshots, guided by critic-based feedback to opti- mize prediction accuracy. The model comprises distinct modules for node encoding, link encoding, token mixing, and channel mixing, integrated through global pooling and followed by a two-layer multilayer perceptron classifier. Extensive experiments on multiple dynamic graph datasets demonstrate substantial gains in area under the receiver operating characteristic curve and average precision. Ablation studies further confirm the necessity of adaptive edge-gradient augmentation and validate the effectiveness of the proposed encoding and reinforcement- learning mechanisms.
Generating scientific data under previously unseen equality constraints requires satisfying the prescribed constraints without destroying the physical structure learned from data. Existing inference-time approaches often enforce physical constraints by repeatedly correcting predicted endpoints, but such discrete corrections can distort learned generative structure and introduce unstable feedback. In this work, we propose Manifold Attracting Flow Matching (MAFM), which continuously evolves a persistent anchor through an ODE whose dynamics drive it analytically toward the constraint manifold. Specifically, its geometric correction preserves model-guided tangent motion while prescribing the normal motion to contract constraint violations toward the manifold. This construction separates model tracking from constraint attraction and provides theoretical guarantees for residual contraction and local feedback stabilization. Experiments across geometric and partial differential equation benchmarks demonstrate improved generation quality over variant constraint-guided baselines while maintaining low constraint residuals. MAFM provides a flexible framework for coordinating learned physical structure and explicit constraints through controllable sampling dynamics, without retraining the underlying generative model.
Modern AI models are not static. They go through multiple updates in their lifecycles. We propose to design Sequential Membership Inference (SeMI) attacks leading to tighter privacy audits by exploiting the sequence of models and injecting a target canary at a controlled insertion time. First, for empirical mean computation, we develop SeMI*, an optimal SeMI attack to identify the presence of a target inserted at a specific insertion step. We derive the power of SeMI* to show that accessing the model sequence yields more powerful MI attacks than scrutinising only the final model. SeMI* exhibits an isolation property. Its power depends on the statistics obtained right before and after insertion of the target. Leveraging this insight, we develop practical white-box (accessing model gradients) and black-box (accessing loss) SeMI attacks against models trained with (DP-)SGD. Across datasets and models trained with (DP-)SGD, our experiments show that SeMI attacks achieve higher power than snapshot-independent baselines, and yield tighter privacy audits thanks to (a) control over the insertion time and (b) observations across the model sequence. Code is available at https://github.com/Thomick/pysemi.
Predicting object interaction outcomes from initial conditions is a central problem in data-driven physical reasoning. Existing methods often use rigid-body-level states or predict the evolution through autoregressive rollouts. Coarse representations cannot represent local deformation, while rollout errors may compound over long prediction horizons. Instead, we introduce PIM, a Physical Intuition Model that directly predicts a physical system's equilibrium state from its initial mesh geometry, kinematics, and physical attributes. PIM combines local k-nearest-neighbor (KNN) attention to encode intra-object geometry, anchor-based cross-object attention to model interactions, and a dual-branch decoder that predicts object-level translations and per-vertex residual displacements. Across MuJoCo scenarios involving both rigid and deformable interactions, PIM achieves lower vertex and centroid errors than the evaluated baselines. By predicting complete equilibrium states in a single forward pass, PIM substantially reduces latency relative to autoregressive learned models and MuJoCo rollouts under the evaluated protocols, enabling low-latency inference when only the final outcome is required.
Speedrun, a practice of completing a game as quickly as possible, has fostered vibrant communities driven by creativity, competition, and mastery of game mechanics and motor skills. However, this contest also attracts malicious actors as financial incentives come into play. As media and software manipulation techniques advance - such as spliced footage, modified game software and live stream with staged setups - forged speedruns have become increasingly difficult to detect. Volunteer-driven communities invest significant effort to verify submissions, yet the process remains slow, inconsistent, and reliant on informal expertise. In high-profile cases, fraudulent runs have gone undetected for years, allowing perpetrators to gain fame and financial benefits through monetised viewership, sponsorships, donations, and community bounties. To address this gap, we propose Tracer, Tamper Recognition via Analysis of Continuity and Events in game Runs, a modular framework for identifying artefacts of manipulation in speedrun submissions. Tracer provides structured guidelines across audiovisual, physical, and cyberspace dimensions, systematically documenting dispersed in-game knowledge and previously reported fraudulent cases to enhance verification efficiency.
Planning for a heterogeneous robot team from a scene image and an instruction involves two kinds of decisions. Some must be judged from the scene: which goals are already met, how the task splits, and which robot does what. Others follow from these and can be computed: prerequisite actions, handoffs between robots, and timing. We argue that each kind should go to the component suited to it: a small vision-language model learns the first, and a deterministic solver computes the second. In NSDL (Neuro-Symbolic Division of Labor), the trained model writes its judgments as a plan sketch, which lists the goals that still need action and the intended steps in order, each with a robot. The solver expands the sketch into an executable joint plan. The model calls two tools, a detector for evidence and the solver, and is trained with the solver in the loop. We evaluate on the public VIKI-L2 benchmark and on LOCUS-Bench, a new benchmark that grades the two kinds of difficulty separately. NSDL outperforms end-to-end planners and formalize-then-plan baselines on both benchmarks and with three backbones from 3B to 8B; the 7B model reaches 97.0% out-of-distribution accuracy on VIKI-L2. Ordering and scheduling errors nearly disappear once the solver writes the structure; the remaining failures are misjudged states, and most goals the model misses are ones it never lists. Code, model weights, and LOCUS-Bench are publicly available at https://github.com/Xincheng-Tan/NSDL.
Benny George Kenkireth, Gopalan Sajith, Sreyas Sasidharan
原文摘要
The multi-word-representation number $μ(G)$ is the minimum number of word-representable graphs whose union is $G$. We investigate $μ(H)$ for graph products $H$ obtained from $G_1$ and $G_2$ via six fundamental products: lexicographic, Cartesian, rooted, corona, tensor, and strong. We prove $μ(H) = \max\{μ(G_1), μ(G_2)\}$ for Cartesian and rooted products. For the corona product, we show $\max\{μ(G_1), μ(G_2)\} \le μ(H) \le \max\{μ(G_1), μ(G_2)\} + 1$, and show that the lower bound is tight when $μ(G_1) > μ(G_2)$ or $G_2$ admits a covering by $μ(G_2)$ word-representable graphs, one of which is a comparability graph. For the lexicographic product, we show $\max\{μ(G_1), μ(G_2)\} \le μ(H) \le μ(G_1) + μ(G_2)$, and show that the lower bound is tight when $\mathrm{cov}_{\mathrm{comp}}(G_2) \le \max\{μ(G_1), μ(G_2)\}$. We provide logarithmic bounds for tensor and strong products. We prove $G^{[k]}$ is word-representable if and only if $G$ is a comparability graph. We establish bounds $μ(G^{[k]}) \le \mathrm{cov}_{\mathrm{comp}}(G)$ and $μ(G^{[k]}) \le k$ for non-comparability word-representable graphs. Using lexicographic powers, we obtain the sublinear bound $τ(n) \le n^{\log_8 6+ε}$ for the extremal function $τ(n)$. Finally, we address the Word-representable Bipartition (WB) problem, proving a negative answer for $n \geq 433$: showing that for every such $n$, there exists a graph of order $n$ that cannot be vertex-partitioned into two word-representable induced subgraphs.
Maximilian Maier, Tsz Kwan Emmy Wong, Jiahao Eric Yang, Falk Lieder, Vanessa Cheung
原文摘要
People increasingly ask AI chatbots for advice, yet large language models (LLMs) are sycophantic, overconfident, and can amplify human cognitive biases. In this article, we investigate how people perceive advice from humans versus LLMs and whether AI advice reshapes incentive-compatible donation decisions. Using Carrier, a new experimental platform, we designed an experiment where participants made a donation decision after discussing it with an AI chatbot advisor, another human, or receiving no advice. Participants judged LLM advisors to be significantly more rational, trustworthy, convincing, and sycophantic than human advisors. Analysis of chatlogs showed that LLMs were more engaged than human advisors: they shared more facts and more often stated their opinion and justified their recommendations. Despite this, donation decisions did not vary across conditions.
Lhea Beumer, Sander de Haan, Niclas Pokel, Pau Vilimelis Aceituno, Benjamin F. Grewe
原文摘要
Predictive coding (PC) trains neural networks through local energy-minimization dynamics without depending on a global error signal that backpropagation (BP) requires. Although PC is a promising alternative algorithm for efficient learning, the computational cost of inference dynamics currently outweighs the resulting learning benefits in practice. In this work, we take inspiration from optimization theory to extract more learning benefit from each inference phase. We introduce Natural PC, which augments standard PC with a lateral circuit, enabling exact and efficient natural gradient learning locally in space and time. Our lateral circuit learns the activity correlations that determine the Fisher information matrix and implicitly computes its inverse, avoiding expensive matrix inversion. Across all tested image reconstruction benchmarks, Natural PC achieves state of the art in terms of sample and compute efficiency compared to existing PC approaches. Most notably, Natural PC scales to ImageNet and matches the performance of BP trained with AdamW and various second-order optimizers at an order of magnitude lower computational cost while also using two orders of magnitude fewer training samples. Our results demonstrate that local natural gradient learning makes PC competitive with BP in both sample efficiency and computational cost in reconstruction tasks.
Shelly Schwartz, Alan Ropp, Michael B. Sneider, Nur Yildirim, Cree M. Gaskin, Chirag Agarwal
原文摘要
AI-generated radiology reports are commonly evaluated against radiologist-authored references, limiting the applicability of standard reference-based metrics when no reference is available. We introduce RadRef, a reference-free metric that scores report claims against image-derived beliefs over a clinical finding vocabulary, weighted by relevance to the clinical indication. By evaluating findings beyond the report's stated claims, RadRef assesses omissions as well as unsupported assertions. On ReXVal, RadRef achieves $73.1\%$ agreement with radiologist rankings among non-tied same-radiograph report pairs, with $61.8\%$ pair coverage and $64.3\%$ tie-inclusive agreement. Its non-tied agreement is close to GREEN ($73.9\%$) and RadFact ($72.6\%$), with its strongest relative results on omissions. We further evaluate $100$ model-generated and $100$ radiologist-authored reports from $100$ MIMIC-CXR studies, independently annotated by two board-certified radiologists. On same-image omission rankings, RadRef attains Kendall's $τ_b = 0.295$, compared with $0.386$ between the readers. In separate paired comparisons across generated reports, no evaluated comparator significantly outperforms RadRef on omissions. The same image-grounded representation provides a false-finding readout, RadRef-FF, with $τ_b = 0.32$ on ReXVal and $0.29$ in the reader study. Controlled perturbations confirm that RadRef depends on the radiograph, report, and clinical indication. RadRef enables image-grounded evaluation before a radiologist-authored reference exists, supporting quality monitoring of report-generation systems in clinical settings, and complements reference-based metrics when one is available.
This paper, Part III of a series, treats $K$-user multiple-access channels (MACs). Its starting point is a pair of fundamental lemmas, a random-coding lemma of Feinstein type and a converse lemma of Verdú--Han type, which are the $K$-user extensions of the two lemmas underlying Han's information spectrum approach to the general two-user MAC. From them we derive the capacity region of the general $K$-user MAC with cost constraint and show that it contains the polymatroid capacity region of the $K$-user Gaussian MAC. We then prove, by an elementary information spectrum calculation that adapts a diffuse--exceptional subspace decomposition used in recent second-order converses for the Gaussian MAC, that for every subset $\cS$ of users and every independent input satisfying the power constraints, the conditional information spectral-sup rate of the $K$-user Gaussian MAC does not exceed $\frac12\log(1+P_\cS/N)$, where $P_\cS$ is the total power of the users in $\cS$, with a probability bound that does not depend on the input. Consequently, the strong converse for the $K$-user Gaussian MAC, in the sense that the error probability tends to one for every rate tuple outside the capacity region, follows directly from the converse lemma, without the wringing technique and without hypothesis-testing meta-converses. This strong converse was first established by Fong and Tan, whose per-blocklength bounds yield it, through wringing and quantization. It is the key ingredient in the converse parts of the coding theorems for mixed Gaussian MACs in Part I, where the $\varepsilon$-strong converse would not suffice.
Theory of Mind (ToM), the ability to infer mental states in others, is pivotal for human social cognition. Existing evaluations of ToM in large language models (LLMs) mainly focus on static belief attribution and false-belief reasoning, leaving a critical aspect of social reasoning underexplored: can LLMs decide whether perceptual evidence is sufficient to justify another person's belief revision? To address this gap, we present MuSeToM, a human-validated, text-based benchmark for evaluating epistemic belief revision in ToM reasoning. MuSeToM contains 2,415 instances adapted from BigToM and OpenToM, with textual cues spanning six sensory categories, and tests whether models can determine when a protagonist should revise a belief and infer the resulting updated belief state. Using MuSeToM, we systematically evaluate state-of-the-art LLMs under direct and chain-of-thought prompting, revealing a pronounced dissonance: while models often detect that a belief may change, they struggle to identify the updated belief and frequently over-update on perceptually salient but epistemically insufficient cues.
Wei Feng, Meiling Chen, Boxiao Chen, Lu Liu, Bin Zi, Libin Ye, Xingyu Lu, Lei Xu 等 9 位作者
原文摘要
Efficient batch execution of agent tasks is critical to LLM training and development workflows, including benchmark evaluation, synthetic-data generation, and RL rollouts. Since these tasks alternate between LLM inference and environment interaction, we present coordinated optimizations to reduce batch makespan, the time required to complete all tasks in a batch. For hybrid LLMs, a workload-based resource model guides GPU-memory allocation between key-value (KV) cache and recurrent state and provides a reference sample concurrency to exploit parallelism while limiting recomputation. Trajectory-aware routing and scheduling promote cache reuse while balancing worker load. Under constrained environment capacity, we decouple sample admission from environment execution and introduce Chain-Advancing Action Scheduling (CAAS), which prioritizes long chains using observed progress to shorten the batch tail. Controlled replay combines actual inference with recorded environment responses and durations for ablations of individual optimizations. Live evaluations with Ling-3.0-Flash and Ling-3.1-Flash on the BFCL, SWE-bench Multilingual, WideSearch, and MCP-Atlas benchmarks demonstrate makespan reductions of 10.4% to 65.0%, while evaluation scores remain aligned. Furthermore, these methods have been successfully applied to training and evaluation workflows for the Ling-3.x model series, accelerating the model-development cycle.
Benny George Kenkireth, Gopalan Sajith, Sreyas Sasidharan
原文摘要
Word-representable graphs, characterized by the existence of a semi-transitive orientation, form a well-studied class of graphs. Comparability graphs form another well-studied class and constitute a subclass of word-representable graphs. Both classes are hereditary and admit characterizations in terms of minimal forbidden induced subgraphs. While the minimal forbidden induced subgraphs for comparability graphs are completely characterized, the corresponding characterization for word-representable graphs remains open. In this paper, we precisely determine which minimal non-comparability graphs are also minimal non-word-representable graphs by classifying minimal non-comparability graphs according to whether they are word-representable. As a consequence, we provide a complete description of minimal non-word-representable graphs containing an all-adjacent vertex. We also address an open problem posed by Kenkireth et al.\ concerning the cover number of word-representable graphs by comparability graphs. We demonstrate the existence of word-representable graphs on $n$ vertices whose cover number by comparability graphs is $Ω(\log n)$, which establishes that the universal $O(\log n)$ upper bound is asymptotically tight for the class of word-representable graphs. For triangle-free circle graphs, we establish that the cover number by comparability graphs is at most $3$ and demonstrate that this bound is tight. More generally, we show that for any circle graph $G$ with clique number $ω(G)$, the cover number by comparability graphs is bounded by $O(\log ω(G))$. Finally, we identify four subclasses of word-representable graphs for which the cover number by comparability graphs of every graph in these classes is at most $2$.
Although self-supervised 3D anomaly detection assumes that acquiring high-precision point clouds is computationally expensive, in real manufacturing scenarios it is often feasible to collect a limited number of anomalous samples. Therefore, we study open-set supervised 3D anomaly detection, where the model is trained with only normal samples and a small number of known anomalous samples, aiming to identify unknown anomalies at test time. We present Open-Industry, a high-quality industrial dataset containing 15 categories, each with five real anomaly types collected from production lines. We first adapt general open-set anomaly detection methods to accommodate 3D point cloud inputs better. Building upon this, we propose Open3D-AD, a point-cloud-oriented approach that leverages normal samples, simulated anomalies, and partially observed real anomalies to model the probability density distributions of normal and anomalous data. Then, we introduce a simple Correspondence Distributions Subsampling to reduce the overlap between normal and non-normal distributions, enabling stronger dual distributions modeling. Based on these contributions, we establish a comprehensive benchmark and evaluate the proposed method extensively on Open-Industry as well as established datasets including Real3D-AD and Anomaly-ShapeNet. Benchmark results and ablation studies demonstrate the effectiveness of Open3D-AD and further reveal the potential of open-set supervised 3D anomaly detection.
Byte-level BPE fragments Chinese text (for Qwen3.8 on Chinese Wikipedia, 3.84 bytes per token on average; 64% of tokens are single characters) and hides spelling from the model. We ask whether an already-trained 27B subword model can be converted directly into a pure byte-level model (one byte, one token), without a patching structure or a local encoder, while keeping most of its ability. Starting from Qwen3.8-27B (48 Gated DeltaNet layers + 16 full-attention layers), we shrink the vocabulary from 248,320 to 384 entries (the 256 byte symbols already in the BPE vocabulary plus special ids), leave every other weight untouched, and continue training in a two-stage curriculum (encoding first, then reading) on 4.77 GB of supervised bytes: all dialogue/behavioral supervision is generated on-policy by the original model itself (character dictionary, word dictionary, general dialogue, coding sessions), and knowledge reading uses encyclopedia text (54% of supervised bytes). Training takes 178.6 h on 8xH200. On the same GPU and serving stack, the byte model's gap to the original falls within the paired bootstrap interval on MATH-500, HumanEval+, MMLU-Pro and non-thinking GSM8K, and it loses 4-6 points on C-Eval, thinking-mode GSM8K and non-thinking IFEval. Without thinking it leads on every character-level task (CUTE 75.2 -> 84.0, EXECUTE-zho 81.6 -> 93.9, our Chinese character suite 54.4 -> 64.0), but lookup knowledge about Chinese characters such as pronunciation and stroke count drops, and thinking does not recover it. The cost is throughput: per-step speed is identical, but each step yields 3-5x less output; Chinese characters per second is about 1/4 of the original, and prefill at 64K bytes is 6.5x slower. We also document four byte-model failure modes (greedy runaway, sampling breaking UTF-8, ineffective penalty terms on a small vocabulary, repetition) and their fixes.
My Thi Diem Phan, Tuyen Trung Truong, Hoai Phuong Ha, Dat Thanh Nguyen
原文摘要
After the 2021-2022 energy crisis, day-ahead prices in Nordic bidding zones no longer move together. We ask whether this divergence also changes the information carried by prices. The answer matters for integrating variable renewable energy (VRE), because it can help identify where wind and solar forecasts are worth producing, publishing, and using. We measured the co-movement of each Norwegian and Swedish zone with Germany-Luxembourg (DE-LU) in 2018-2025. We then used a single model, selected on 2024 data, to forecast 2025 prices at 11:00 on the previous day and tested which inputs improved accuracy in each zone. The zones moved alike until 2019, but co-movement diverged in Sweden in 2020 and in Norway in 2021, and the split has persisted. In 2024, four southern zones co-moved strongly with DE-LU (intraday correlations of 0.54-0.71), while four northern zones co-moved weakly (0.25-0.30). Yet price information did not follow this split. Previous-day neighbouring prices reduced mean absolute error by at most 2.2% in any zone, equally in coupled and isolated zones. Removing the transmission system operators' wind and solar forecasts increased mean absolute error by at most 6% in Norway, 8-16% in Sweden, and 31% in DE-LU, and the ordering held in 2024. Within Norway, the value of these forecasts showed no detectable dependence on reservoir filling. The divergence in co-movement was therefore not matched by a divergence in price information, which differed mainly between national generation systems. Thus, a zone's coupling to the continent does not indicate the value of VRE forecasts there.
Multimodal representation learning combines complementary signals to capture shared semantic information. For dynamic scenes, these representations need to remain consistent across views while preserving distinctions between temporal states. In visual speech representation learning, view variation changes the visibility and reliability of articulation cues, making it difficult to learn such representations from synchronized audio and video. To address this challenge, we propose MuViRL, a self-supervised framework for multimodal view-invariant visual speech representation learning via calibrated view aggregation. The framework combines synchronized 3D Gaussian view synthesis with multimodal self-supervision to learn consistent representations from real and synthesized data. A multimodal decision model learns relative view utility from teacher prediction losses, and its calibrated scores guide the aggregation of complementary information. Aggregation-to-Single Distillation transfers the aggregated predictions to a single branch. Real-synthetic alignment matches synchronized windows across views while preserving their temporal relations. Experiments on real and synthesized multiview data demonstrate that MuViRL improves multimodal representation consistency and visual speech recognition performance across views.
Recent works have proposed incorporating heavy-tailed (HT) noise into diffusion- and flow-based generative models, with the goals of better recovering the tails of target distributions and improving generative diversity. This motivation is intuitive: if the data are heavy-tailed, HT noise may appear better matched than light-tailed (LT) Gaussian noise. However, replacing Gaussian noise by HT noise also changes the underlying estimation problem. In this paper, we revisit this paradigm through a combined theoretical and empirical study, establishing sampling-error bounds for two representative diffusion models driven by HT and LT noise. We show that HT diffusion models can achieve exponentially fast initialization, but accumulated training errors may offset this advantage. Experiments on synthetic and real-world datasets support our theoretical findings. Our results call into question a growing design trend in generative modeling and challenge the use of HT noise to improve rare-region exploration.
In a nonergodic channel such as a slowly fading link, the channel law is drawn once and held fixed for the whole codeword, and the relevant performance measure is the $\varepsilon$-capacity. We study the $\varepsilon$-capacity regions of two-user mixed multiple-access channels (MACs) with finite alphabets, additive cost constraints and an arbitrary mixing measure. The direct parts are derived from a single information spectrum formula; the converse parts use, in addition, Fano's inequality and Ahlswede's strong converse applied to each component. Without knowledge of the realized component, we prove a single-letter inner bound for every $0\le\varepsilon<1$, in which one input serves all components; for $\varepsilon=0$ it is tight, and the capacity region is given by the essential infima of the component mutual informations. We prove a single-letter outer bound, the outage region, in which each component may use its own input; it holds with or without channel state information (CSI) at either terminal, and it is the $\varepsilon$-capacity region when the encoders know the component and the parameter space is countable. Both bounds coincide, in all four CSI scenarios, for well-ordered mixed MACs, which include degradedly ordered families and mixtures with a common optimal input, such as mixed additive MACs. Without CSI the outer bound can be strictly loose, and the no-CSI $\varepsilon$-capacity region for $0<\varepsilon<1$ remains open in general. Finally, we extend Ahlswede's strong converse to $K$ users with cost constraints and show that most of these results extend to $K$-user mixed MACs.
We introduce CABOT, a computational model for autonomous mobile robots in which the discrete plane the robots operate on is not a passive stage but an active computing medium. In the classical Look-Compute-Move (LCM) model every increase in computational capability is obtained by strengthening the robots themselves, with lights, memory or wider visibility, at a cost paid once per robot. CABOT strengthens the environment instead: each cell of the grid holds a state from a finite set Q, the states evolve synchronously under a uniform local rule that also senses whether a robot stands on the cell, that is, the floor is a cellular automaton, and the Look operation returns the states of the visible cells in addition to the positions of the visible robots. The robots are as weak as we can make them: oblivious, silent, lightless, myopic with visibility range 1, without multiplicity detection and unable to write to the floor. The number of field states |Q|, rather than any robot capability, becomes the measure of the resources required. On an inert floor (|Q| = 1) such robots can neither form a line, nor gather from scattered positions, nor explore a grid. We show that a handful of states changes this. Line Formation from connected configurations is solvable with four states, and field-silently, so that the floor also reaches a fixed point, with five. Gathering from arbitrary, possibly disconnected, configurations is solvable with three states and Line Formation from arbitrary configurations with six; we prove that neither can be solved field-silently with any number of states. A single robot explores every finite grid forever with four states and every torus with two, which is optimal. Further rule sets, validated by simulation only, equalize the spacing of robots on a row, fill a pattern painted on the floor and guide robots through a maze, and indicate the range of the model.
Francesco Ragusa, Rosario Leonardi, Michele Mazzamuto, Daniele Di Mauro, Camillo Quattrocchi, Alessandro Passanisi, Irene D'Ambra, Antonino Furnari 等 9 位作者
原文摘要
Understanding human behavior from complementary egocentric (ego) and exocentric (exo) points of view enables the development of systems that can support workers in industrial environments and enhance their safety. However, progress in this area is hindered by the lack of datasets capturing both views in realistic industrial scenarios. To address this gap, we propose ENIGMA-360, a new ego-exo dataset acquired in a real industrial scenario. The dataset is composed of 180 egocentric and 180 exocentric procedural videos temporally synchronized offering complementary information of the same scene. The 360 videos have been labeled with temporal and spatial annotations, enabling the study of different aspects of human behavior in industrial domain. We provide baseline experiments for 4 foundational tasks for human behavior understanding: 1) Temporal Action Segmentation, 2) Keystep Recognition, 3) Egocentric Human-Object Interaction Detection, and 4) Video-Grounded Cross-View VQA, showing the limits of state-of-the-art approaches on this challenging scenario. These results highlight the need for new models capable of robust ego-exo understanding in real-world environments. We publicly release the dataset and its annotations at https://fpv-iplab.github.io/ENIGMA-360/.
Reconstructing optically thick media from photographs requires modeling multiple scattering and its response to changes in extinction. Frozen-source tomography discards this derivative; differentiable path tracing estimates it through expensive, noisy sampling. We instead predict the multiple-scattering field with a neural operator and differentiate through it during inverse rendering, while computing transmittance and single scattering directly. A further challenge arises during optimization: an operator that predicts the field accurately on held-out media can still produce poor derivatives at the density states visited by the optimizer. We address this by fine-tuning on states generated during inversions driven by the operator, using Monte Carlo scattering-field supervision. On held-out optically thick clouds, our method improves relighting over frozen-source tomography at equal time and outperforms differentiable path tracing given substantially more time. Training on inversion trajectories also improves derivative alignment at visited states and reduces optimization drift.
To a time series can be also given the structure of directed graph, opening the doors to the usage of graph neural networks (GNNs) in this context. In this paper we develop a new pipeline involving novel geometric deep learning models that merge the capabilities of ordinary TCNs with the ones of GNNs to solve an anomaly detection problem on electrocardiogram signals.
John Ayotunde, Qinghua Xu, Guancheng Wang, Lionel C. Briand
原文摘要
Safety monitoring is essential for Cyber-Physical Systems (CPSs) such as Unmanned Aerial Vehicles (UAVs). However, unsafe events are rare in CPS operations, creating class imbalance that degrades data-driven safety predictors. Behavioral uncertainty, reflected in erratic control signals or rapid heading changes, is often correlated with safety outcomes: uncertain behaviors occur more often during unsafe operation. Yet safety monitoring rarely exploits it. We propose U-Balance, a supervised approach that leverages behavioral uncertainty to rebalance datasets before training a safety predictor. U-Balance first trains a GatedMLP-based uncertainty predictor that represents telemetry windows using distributional kinematic features and outputs an uncertainty score. Its uncertainty-guided label rebalancing (uLNR) mechanism probabilistically relabels $\textit{safe}$ windows with unusually high uncertainty as $\textit{unsafe}$, enriching the minority class without synthesizing data. A safety predictor is then trained on the rebalanced dataset. We evaluate U-Balance on a UAV benchmark and ForEnt, a quadruped-robot entrapment dataset. On UAV, U-Balance achieves F1/AUROC/AUPRC of $0.801/0.972/0.776$, outperforming the strongest safety-prediction baselines by 14.3/2.2/6.1 percentage points ($\textit{pp}$), respectively. While the strongest baseline varies across metrics (MLP for F1 and AUPRC; CNN for AUROC), U-Balance leads on all three. On ForEnt, it achieves $0.522/0.806/0.323$, outperforming the strongest baselines by 4.4/9.2/1.4 $\textit{pp}$. Using the same Bi-LSTM safety predictor, uLNR outperforms the strongest alternative rebalancing methods by 17.0/4.1/8.2 $\textit{pp}$ on UAV and 2.3/6.5/1.9 on ForEnt. uLNR also outperforms Early and Late Fusion on both datasets. To our knowledge, this work is the first to use behavioral uncertainty for dataset rebalancing in CPS safety monitoring.