This paper describes the approach of the Unibuc - NLP team in tackling the "Voight-Kampff" Generative AI Detection task from PAN at CLEF 2025. In the Subtask 1: Voight-Kampff AI Detection Sensitivity we have explored different approaches of leveraging embeddings extracted from LLMs, using either classical ML models or further fine-tuning LLMs with a classification head for the downstream binary classification task. Our experiments showed that using a majority vote ensemble of classical ML models (Light GBM, XGB, Logistic Regression and SVM) is comparable with fine-tuning for downstream task achieving around 99.4% F1 score. For the Subtask 2: Human-AI Collaborative Text Classification we explored a way of combining features at different layers extracted using transformer backbone with layer-wise projection and attentive pooling which proved modest results.
Medical image classification in resource-constrained settings must contend with limited annotations and computational capacity. We evaluate four ZACH-ViT variants with linear, Euclidean prototype, Poincare and spherical heads alongside two compact multiple-instance-learning baselines. All models are trained from scratch with fewer than 260,000 parameters across seven MedMNIST+ datasets and three external benchmarks (PCam, ISIC2017 and Shenzhen tuberculosis radiographs). The low-data protocol uses 50 training images per class and five seeds, with validation-selected checkpoints at 23 and 47 epochs; a full-data reference uses one seed. Compact-TransMIL achieves the highest ten-dataset mean test Macro-F1 at 23 epochs (53.34%) and with full data (73.93%). At 47 epochs, ZACH-Linear narrowly leads (57.76% versus 57.54% for Compact-TransMIL). Extending training improves Macro-F1 in 239 of 300 paired comparisons. On ISIC2017, hZACH-Spherical has the highest 47-epoch low-data Macro-F1 (44.09% with SD 4.09%), whereas ZACH-EuclideanProto leads with full data (51.27%). We additionally conduct 40 CPU-only training runs on ISIC2017 and Shenzhen using four CPU threads; model-only single-image inference requires approximately 4-9 ms, while image decoding and preprocessing substantially increase end-to-end latency. The results demonstrate the feasibility of CPU-only training on the tested platform and the task- and regime-dependence of compact medical imaging under strict resource constraints.
External skills provide domain knowledge and procedures without updating model parameters, but flat text makes their applicability, dependencies, and composition difficult to express. Existing skill graphs organize rules into nodes and relations mainly for retrieval and composition. They give less attention to what should change after a failure, which examples should be retested, and which edits should be reverted. We introduce GraphSkillAA, short for Graph-Skill Abductive Attribution, which connects rule use, error localization, and update validation through the same skill graph. Nodes record applicability, procedures, and exclusion conditions, while edges describe prerequisite and enhancement relations. The student records the nodes and edges used during execution. The teacher contrasts related successful and failed trajectories to identify the fields or relations that need repair, keeping changes as small as possible. Execution lapses or insufficient evidence leave the graph unchanged. A Local Gate retests affected examples, while a Big Gate checks the merged graph on the complete update pool. Rejected edits are rolled back, and model parameters remain frozen throughout. With GPT-5.6-sol, GraphSkillAA reaches 81.5%, 66.7%, and 91.2% on SearchQA, LiveMath, and DocVQA, respectively, and achieves the highest mean performance in every main setting. These results show that graph structure supports skill selection and composition while also providing explicit targets for repair, retesting, and rollback.
Rosalina Ribeiro Silva, Zichen Feng, Rikard Olajos, Tim Rolff
原文摘要
We propose an immersive virtual-reality game designed to enable multi-user, multi-language learning during co-play. We hypothesize that our method will be more engaging, leading to better rhythm, pronunciation, memorization, and speaking confidence. We implement the experience as a fantasy role-playing game by introducing vocabulary and phrases to the user as spells to learn. Players then cast more efficient spells depending on their language skills. For evaluation, we compare the game to a single-user baseline that replicates traditional language learning in a non-immersive setting. We measure short- and long-term retention of vocabulary and phrases, as well as scoring rhythm and pronunciation. Furthermore, we collect user data on engagement and mental workload.
Frontier AI agents increasingly succeed on realistic, long-horizon tasks and are beginning to improve the systems through which they operate. Yet existing benchmarks often restrict agents to narrow action spaces, use simplified environments, or rely on verification methods that are unreliable for evaluating arbitrary agent-generated systems. We introduce Singularity-Bench, an open-ended benchmark for autonomous AI development with six end-to-end tasks spanning pre-training, mid-training, post-training, inference, the agent harness, and cluster orchestration. Each task requires agents to implement core systems from a minimal starting point and evaluates them against mature human-engineered baselines on identical hardware. Additionally, through extensive red-teaming covering 9 of the 10 categories in the OWASP Top 10 for Agentic Applications, we expose reward-hacking vulnerabilities in the conventional Harbor framework and develop Breakwater, an extension of Harbor that blocks all attacks tested in our study. Across 11 agents built from 7 frontier models and 2 harnesses, Opus 5.5 with Claude Code matches or outperforms the human baseline on all six tasks with a Singularity-Index of 1.25, suggesting it may have the skills needed to reach the singularity. The remaining agents average only 0.58. Our vision is for Singularity-Bench to track how much of AI development agents can already automate and when they may transition from assisting to taking end-to-end responsibility for building stronger successors.
Petr Kuznetsov, Pierre Sutra, Guillermo Toyos-Marfurt
原文摘要
In this work, we study progress conditions for commutativity-aware, linearizable implementations of shared objects. Motivated by the observation that commuting operations can be executed in parallel, we introduce conflict-obstruction-freedom: a process is guaranteed to complete its operation if it runs for long enough without encountering step contention with conflicting (non-commuting) operations. This condition generalizes obstruction-freedom and wait-freedom by allowing progress as long as step contention is only induced by commuting operations. We prove that conflict-obstruction-free universal constructions are impossible to implement in the asynchronous read-write shared memory model. This result exposes a fundamental limitation of conflict-aware universal constructions: the mere invocation of conflicting operations imposes a synchronization cost. Progress requires eventual resolution of pending conflicts.
Hung Nguyen-Huu, Nguyen Anh Tuan, Ting Zhang, Yue Liu, Chengran Yang, Lwin Khin Shar, Eng Lieh Ouh, David Lo
原文摘要
Affected-version ranges in vulnerability databases decide which deployments scanners, patching pipelines, and incident responders treat as vulnerable, yet these ranges are often incomplete, so vulnerable releases are silently treated as safe. Auditing them requires executable evidence: given a proof-of-concept (PoC) exploit for a listed version, show that the same vulnerability can be triggered on an earlier, unlisted version. This cross-version PoC migration combines source-code analysis, environment repair, and execution-based verification. We introduce CrossPloit, a benchmark of 127 migration instances across 45 CVEs in the Python web ecosystem. Each instance includes a version-pinned container, a reference PoC, and a vulnerability-specific oracle; two validators check that the observed effect arises from the disclosed vulnerability. Benchmark construction confirms 149 vulnerable prior versions absent from the affected-version lists of four databases, 127 of which require migration. Across 18 agent configurations (nine models and two harnesses), the best configuration resolves 58.3% of instances, compared with 8.7% for an adapted trace-guided baseline. Attack-vector shifts are particularly difficult: no configuration resolves more than 37.9% of these instances. Higher resolve rates are associated with a smaller share of missed-API failures and a larger share of incomplete executions among the remaining failures. In one configuration, withholding the reference PoC and its environment specification reduces resolve rate from 50.4% to 15.0%, highlighting the value of the supplied reference package. CrossPloit provides an execution-based measurement of under-reported exposure and of how far its enumeration can be automated.
Siddhesh Pawar, Chen Cecilia Liu, Anna Korhonen, Isabelle Augenstein
原文摘要
As LLMs are increasingly deployed in value-sensitive contexts, aligning them with human values has emerged as a post-training objective in NLP, aimed at instilling additional values and guiding model behaviour. However, current evaluations remain narrow. Because values are interconnected, it is unclear how fine-tuning for specific values alters unrelated capabilities and behaviours. In this work, we adapt four instruction-tuned LLMs across individual Hofstede's value dimensions and evaluate them on both cross-value transfer and broader behaviours. We find that value-specific training produces unintended side effects on untrained values, risk preferences, social cognition, and safety traits. Consequently, we suggest that value alignment after post-training should incorporate broader behavioral evaluations, and we highlight the need to investigate where additional value training might best fit into the pipeline. We release all 48 value-tuned model checkpoints at https://huggingface.co/sidicity/cultural-value-models
Large language models (LLMs) are increasingly used in binary decompilation to refine the C-like pseudocode produced by traditional rule-based decompilers. While this pseudocode is useful, it is a heuristic and lossy abstraction rather than a faithful copy of the source code. It often contains decompiler errors, especially for aggressively optimized binaries where critical low-level details are obscured. We present ASAP, an Assembly-Source Aligned Pseudocode refinement framework for binary decompilation. ASAP learns source-aligned assembly representations from paired source and binary functions using joint function-level and snippet-level contrastive alignment. A Q-Former then compresses chunk-level assembly features into a fixed number of assembly tokens. These tokens and the decompiler-produced pseudocode jointly condition the decompilation LLM. During refinement, we use stochastic pseudocode masking and a relative assembly-advantage loss to discourage the model from relying solely on pseudocode. On two decompilation benchmarks across multiple compiler optimization levels, ASAP improves the average re-execution rate from 64.7% to 71.9% and the average recompilation rate from 91.6% to 96.6% compared with the strongest baseline. Our approach offers both a new perspective and a practical solution to binary decompilation.
Ugo Insalaco, Jean-Emmanuel Deschaud, Pierre Dellenbach
原文摘要
Cameras and LiDARs (Light Detection and Ranging) provide complementary observations of the same scene: cameras capture rich appearance cues while LiDAR sensors provide accurate geometric measurements. In this work, we address the problem of fine image-to-point cloud registration from a single image and a roughly estimated camera pose. On robotic platforms, residual reprojection errors of several pixels are commonly observed even after calibration due to temporal desynchronization, mechanical vibrations, or inaccuracies in point cloud reconstruction. Such errors can lead to decimeter-level geometric discrepancies at moderate distances and significantly affect multiple downstream tasks, such as dense cross-modal supervision, texture mapping, and semantic label transfer. To address this challenge, we introduce ReFineR: an offline framework to finely register individual images to pre-reconstructed point clouds. ReFineR builds upon differentiable point splatting rasterization and a smooth Gradient Orientation Measure (GOM) to iteratively refine the camera pose through analysis-by-synthesis optimization. We evaluated the method on a curated dataset of manually calibrated images collected from heterogeneous data sources. Furthermore, we show that an automatic point cloud rescale factor can be estimated prior to optimization, enabling robust refinement across a wide range of scenarios using a single scale-invariant formulation. The project page can be found at https://refiner-project.github.io/
Social scene datasets label people at two levels: what each person is doing and what the group is there for. Training on both levels at once is often assumed to help each of them. We test this assumption for vision-language models (VLMs) on JRDB-Social, treating person actions as the details and group intent as the whole. We fine-tune Qwen3.5 4B under eight training arms with three seeds, repeat the core arms on Qwen3.5 9B and Qwen3-VL 4B, and judge every arm with a verdict rule fixed before testing. Writing each member's actions before the intent raises intent macro-F1 from 30.1 to 31.0, and writing the intent first reaches 31.2, while action F1 stays within 0.2 points of an action-only model. Under the selected recipe, no arm trained with intent improves actions. The gain requires correct, image-grounded actions: empty action lists or actions borrowed from another group with the same intent lower intent F1. Forced-prefix tests, occlusion, member masking, linear probes and component grafting show that the intent follows the written actions, relies more on people and partly resides in the vision tower. Supplying actions as input instead of as targets hurts both levels.
Parameswaran Kamalaruban, Viktor Drobnyi, Maeve Madigan, Julia Rozanova, David Sutton, Stuart Burrell
原文摘要
Sequential financial transaction data are increasingly being used by financial institutions to train payment foundation models at scale, producing rich contextual embeddings that support applications such as fraud prevention, credit risk assessment, and offer personalization. However, such models are not designed for flexible instruction-following reasoning and require bespoke downstream models for each application. In contrast, large language models (LLMs) provide a natural-language interface for flexible transaction reasoning, yet naively serializing transaction histories into text representations is inefficient and costly. Multimodal fusion architectures bridge these paradigms, injecting transaction embeddings directly into flexible language reasoning pipelines. In this work, we present a systematic study of how transaction embeddings, pretrained at the scale of a global payments network, interact with LLM-based reasoning through embedding injection, transaction-language alignment, and instruction tuning. To do so, we train the Multimodal Instruction Network for Transactions (MINT), which combines a transaction encoder pretrained on billions of real-world payments with multimodal instruction tuning on millions of transaction question-answering examples. We evaluate predictive and extractive question answering, zero-shot generalization to unseen question types, and inference efficiency. Our results reveal a consistent trade-off: compact transaction embeddings improve predictive QA accuracy while substantially reducing input tokens, latency, and memory relative to a text-serialized baseline, whereas serialized histories remain highly competitive for extractive QA. Overall, our empirical findings provide a set of actionable insights and practical guidance for building multimodal transaction-aware language systems at network scale.
Ziying Song, Shengkai Zhang, Caiyan Jia, Lei Yang, Lin Liu, Yuchen Liu, Ziyang Liu, Zirui Li 等 13 位作者
原文摘要
Joint-Embedding Predictive Architectures (JEPAs) enable future prediction in latent space without pixel reconstruction. However, latent predictive consistency alone does not establish the relative driving quality of candidate plans. We present ActWorld-JEPA, a joint-embedding formulation for candidate-based driving. Where Plan-Conditioned JEPA (PC-JEPA) uses global and step-wise trajectory intentions to guide recurrent latent prediction, then combines the predicted future with the current scene through gated cross-attention; Relational Residual Reranking ($R^3$) models agreement and discrepancy between plan and future representations under planning supervision, yielding bounded, gated residual corrections to the base planning scores. This formulation connects latent prediction to candidate selection using only observed context and proposed trajectories at inference. Benchmark-specific implementations report 94.1 PDMS on NAVSIM v1 navtest, with the final selector calibrated using aggregate NAVTEST feedback, and 91.6 and 42.7 EPDMS on NAVSIM v2 navtest and navhard, respectively. The nuScenes evaluation reports a mean $L_2$ error of 1.20 m and a predicted collision rate of 0.76% over 1--6 s. This work provides an effective path from JEPA's predictive representations to driving decisions informed by anticipated consequences.
Dongzhihan Wang, Yi Du, Chen Wang, Xiuxian Li, Liang Xu
原文摘要
Vision-Language Navigation (VLN) requires long-horizon instruction-consistent planning and short-horizon path planning grounded in the currently observed geometry. Anticipating how future observations evolve along the instructed route provides predictive guidance beyond the current field of view, motivating video-action models (VAMs) that learn representations of future visual states. However, many existing VAMs couple future visual prediction with action generation during inference, despite their distinct temporal requirements: instruction-conditioned route guidance can be reused across multiple control steps, whereas local trajectories must continuously adapt to fresh observations. To bridge this gap, we introduce HAVANA, a dual-system VAM that decouples predictive reasoning from reactive local planning. A slow system produces a compact language-conditioned latent plan from observation history, while a fast system combines the latest plan with the current observation to generate local trajectories. Their asynchronous execution enables rapid adaptation without repeated future prediction, and no future video is generated during inference. Experiments on OmniNavBench demonstrate state-of-the-art navigation performance over prior baselines, while zero-shot real-world deployments further show robust generalization to unseen environments.
In this paper, we study classes of Boolean functions that are testable with $O(ψ+1/ε)$ queries, where $ψ$ depends on the parameters of the class (e.g., the number of terms, the number of relevant variables, etc.), is independent of (or sublinear in) $1/ε$, and is independent of the total number of variables $n$. In particular, when $ε$ is sufficiently small so that $ε=O(1/ψ)$, the query complexity is $O(1/ε)$, matching the known tight lower bound $Ω(1/ε)$. In particular, if the structural parameters of the class are fixed and $ε=1/ω_n(1)$ (e.g., $1/\log n$), then the asymptotic query complexity is $O(1/ε)$. This result was previously known for classes of terms of size at most $k$ and exclusive OR functions of at most $k$ variables. In this paper, we extend this list to include the classes: $k$-junta, functions with Fourier degree at most $d$, $s$-sparse polynomials of degree at most $d$, and $s$-sparse polynomials. Additionally, we show that for any class $C$ of Boolean functions that depend on at most $k$ variables, if $C$ is properly exactly learnable, then it is testable with $O(1/ε)$ queries for $ε<1/ψ$, where $ψ$ depends on $k$ and independent of the total number of variables $n$.
Many benchmarks for rare events score a prediction in three steps: pool the prediction and the target over spatial blocks, threshold each pooled field, and score the resulting contingency table. We show that at a fixed rare threshold the max-pooled Critical Success Index (CSI) cannot separate two different things: how well a system places events, and how well its amplitudes are calibrated. Sharp observations push many blocks above the threshold, while the attenuated predictions of squared-error regression leave those same blocks below it. A monotone calibration map is fitted on held-out data and applied separately to each system. The map needs no retraining and cannot change the order of any two pixels, so whatever part of a difference it reproduces cannot be evidence of better spatial ranking. We report how much of a reported difference the map removes. On SEVIR, two released checkpoints of one architecture differ by -29.5% in extreme-threshold CSI before the control and by +5.3% after it. Across 450 pairwise contrasts among 6 systems, how far apart two systems sit in pooled frequency bias predicts how far their CSI difference moves under the control (r = 0.896 when the map is fitted after pooling), and 64 contrasts reverse sign. At CasCast's published extreme-event operating point, calibration alone removes 88%-106% of the cascade's gain over its backbone across three controls. Calibration also reveals advantages hidden by a better-calibrated baseline. On geostationary infrared imagery, a larger share of the score can be explained by calibration as events get rarer. The problem therefore needs two things at the same time: a threshold that is fixed in advance, and a model whose output is poorly calibrated at that threshold. We recommend reporting pooled frequency bias and a symmetric held-out FreeKnob Audit alongside rare-event pool-and-threshold scores.
Chenkai Hu, Yingqian Zhang, Ming Xie, Yi Fang, Chengxiang Jin, Chao Li
原文摘要
Multi-agent reinforcement learning has shown great promise for traffic signal control. However, most existing methods rely on two key assumptions: (i) agents are assumed to be homogeneous with unified observation/action spaces; (ii) full and reliable lane-level traffic observations are available to all agents. In practice, urban road networks exhibit substantial structural heterogeneity (e.g., lane configurations and phase designs), while sensor failures and incomplete detector coverage often result in missing incoming lane measurements. Such issues can severely impair coordination and destabilize learning. To address these challenges, we propose HALight, a communication-enhanced heterogeneous multi-agent reinforcement learning framework for traffic signal control under lane-level missing observations. HALight supports intersection-specific observation and action modeling to accommodate heterogeneous intersections, and incorporates a lightweight one-hop neighborhood communication module to enrich spatial context when lane-level observations are partially unavailable. Moreover, HALight adopts a sequential policy update scheme to improve coordination among heterogeneous agents. Experimental results on two real-world road networks demonstrate that HALight outperforms representative homogeneous and heterogeneous baselines in terms of traffic efficiency and convergence behavior. The code and datasets used in this work are available at https://github.com/Huchenkaii/HALight.
We present ZFLean, a Lean 4 library for doing core mathematics inside a model of ZFC with the ergonomics expected of typed Mathlib developments. Building on Mathlib's ZFC model, we contribute a relational calculus for sets with rewriting hints and small predictable tactics, canonical set-theoretic constructions -- Booleans, naturals, integers, sums/option -- and bridges between ZFC objects and Lean's native types enabling mixed set-level/typed proofs. The layer reduces boilerplate for extensional reasoning while remaining compatible with vanilla Mathlib. We discuss library organization and usage patterns that lower the friction of set-theoretic formalization in a dependently typed assistant. We demonstrate typical use of the framework with a case study exercising our constructions and relational calculus through a proof of an isomorphism theorem on curried functions.
In dexterous in-hand manipulation, learning-based methods achieve strong task performance with increasing generalization to unseen objects. A complementary advantage of model-based methods, central to this work, is online adaptation to new task specifications without policy retraining, given suitable system models. However, due to the complexity of physical contacts and limited sensing modalities, existing model-based methods still struggle with modeling errors, even when using advanced contact models or optimization schemes, which limit their real-world applications. To advance the effectiveness and robustness of model-based contact-rich in-hand manipulation, this paper proposes a novel bi-level framework with tactile incorporation to mitigate these limitations. Specifically, at the high level, reference finger motions and contacts are jointly generated via contact-implicit model predictive control. The high-level module enables online reactive planning and disturbance recovery. At the low level, these references are concurrently tracked through a hand force-motion model and tactile feedback. The low-level module compensates for modeling errors and enhances manipulation robustness. The complementarity between the two levels is manifested in two dimensions: 1) subsystem decomposition, with a slow optimization-based planning layer and a fast tactile-reactive control layer; and 2) functional decomposition, combining task-level optimality with contact-level stability. Extensive experiments demonstrate that our approach outperforms existing model-based methods in accuracy and robustness, while supporting online replanning during execution. Our method successfully completes a diverse set of challenging tasks in real-world environments, even under appreciable external disturbances and modeling uncertainties.
Ahmad Halimi Razlighi, Maximilian H. V. Tillmann, Edgar Beck, Bho Matthiesen, Armin Dekorsy
原文摘要
Cooperative multi-task semantic communication (CMT-SemCom) improves task execution performance by leveraging shared representations. However, as we demonstrated in [1], cooperative multi-tasking can be either constructive or destructive, depending on the semantic relationships among tasks. To ensure constructive cooperation, we propose a semantic-aware task clustering method for CMT-SemCom. We have formulated a sequential multi-stage optimization problem in which semantically aligned tasks are clustered once after a short initial training phase, and then end-to-end (E2E) joint training is conducted exclusively within the discovered groups. Specifically, the problem decomposes into two stages: (i) a semantic clustering problem leveraging hierarchical density-based spatial clustering, and (ii) an intra-cluster E2E CMT-SemCom learning problem. Simulation results demonstrate that the proposed framework effectively mitigates destructive cooperation and negative transfer, yielding accuracy gains compared to unclustered multi-tasking and individual training baselines.
Iñaki Dellibarda Varela, R. Sendra-Arranz, Pablo Romero-Sorozabal, J. M. Valverde-García, Annemarie F. Laudanski, Álvaro Gutiérrez, Eduardo Rocon, Manuel Cebrian
原文摘要
Large language model (LLM)-based multi-agent systems (LLM-MAS) are kept out of safety-critical domains by emergent failures that are hard to detect and to attribute. Current evaluation relies on a centralised evaluator, human or LLM judge, that must master the domain and hold the whole interaction history in one context. We propose that the agents executing a task can themselves attribute its failures. We introduce POIROT (Phased Oversight via Introspection, Review and Opinion Tallying), a modular protocol in which each agent casts a private vote within a hazard space derived from the system architecture, and the votes are aggregated into a single attribution. Before the vote, phases are added that progressively increase the interaction between agents, from independent self-assessment to sharing the self-assessments and peer interrogation. We build the protocol phase by phase, measure the contribution of each phase through an ablation, and compare every configuration with a single-LLM evaluator. We evaluate one proprietary and two open-weight LLMs on the Who&When benchmark and on BLAME (Benchmark for Localising Agent Malfunctions Effectively), a new benchmark for fault attribution in clinical rehabilitation and algorithmic trading systems. Agent-based attribution outperforms the single-LLM evaluator, but the benefit comes from independent self-assessment, not from interaction: sharing the assessments makes agents converge on the same, often wrong, culprit, and interrogation does not recover the loss. Without sharing and interrogation, the protocol outperforms the evaluator in all nine model-benchmark configurations (up to +50 percentage points). POIROT and BLAME are released as open source.
Photonic neural network accelerators have emerged as a promising solution for reducing the power consumption of deep neural network (DNN) inference by executing matrix multiplication on analog photonic hardware. However, their inference accuracy is limited by low-bit quantization required for photonic processing and by analog photonic noise. This paper proposes an approach for improving inference accuracy by exploiting the high similarity among activation vectors in DNNs. Specifically, our approach constructs a representative vector for each group of similar activation vectors, computes its inner product digitally with high precision, and offloads the residual inner products to the photonic accelerator. Since the activation vectors within each group are highly similar, the residual vectors have much smaller norms, thereby reducing the effects of quantization errors and analog photonic noise. Furthermore, our approach preserves the photonic inner-product computation pipeline, allowing seamless integration into existing and future photonic accelerators without hardware modification. Experimental results on six representative DNN models demonstrate that the proposed approach can improve inference accuracy across diverse network architectures and application domains.
We study how the sign operator interacts with gradient noise to shape the learning dynamics and scaling behavior of sign-based optimizers. In power-law kernel regression with diagonal features, we establish a functional scaling law (FSL) for SignSGD under general learning rate schedules. The FSL describes the learning dynamics through two processes: signal learning and noise forgetting. The key mechanism is noise-induced preconditioning: gradient noise smooths the sign operator, inducing an effective preconditioned SGD. Noise-curvature alignment enables this preconditioning to transform the effective Hessian from $\mathbf{H}$ to $\mathbf{H}^{1/2}$. The resulting dynamics exhibit faster signal learning but slower noise forgetting than SGD. This tradeoff yields a regime-dependent comparison with SGD: (i) SignSGD achieves improved data efficiency in the hard-task regime; (ii) both optimizers achieve the same data efficiency in the easy-task regime, but SignSGD maintains this efficiency at larger batch sizes, requiring fewer optimization steps for a fixed data budget. Power-law kernel regression experiments support the proposed mechanism and predicted scaling behavior. LLM pre-training experiments further suggest that the SignSGD FSL better captures the batch-size dependence of loss dynamics for sign-based optimizers than the original SGD FSL.
Anna V. Guglielmi, Linda Senigagliesi, Marco Baldi, Stefano Tomasin
原文摘要
We consider a physical layer authentication scenario in which a legitimate user (Alice) communicates with an authen- ticator (Bob) through a reconfigurable intelligent surface (RIS), controlled by Bob. An attacker (Trudy) aims to impersonate Alice by suitably precoding her messages and still transmitting them through the RIS. This two-part study analyzes this attack-defense interaction between Bob and Trudy under perfect (Part I) and partial (Part II) channel knowledge by all the agents. Part I formulates the joint RIS selection and precoder design under perfect channel knowledge as a finite zero-sum game, establishing exact channel subspace conditions for attack indistinguishability alongside equilibrium mixed strategies. Part II extends the defense to incomplete channel information, where each agent observes only its own channel, by formulating a finite-action Bayesian game and characterizing the resulting Bayesian Nash equilibrium. Simulations in geometric and 3GPP TR 38.901 UMi-NLOS channels reveal that authentication performance is governed by available spatial degrees of freedom, channel correlation, and operating signal-to-noise ratio. The proposed equilibrium strategies effectively balance communication rate and security, outperforming conventional benchmark policies.
David Hartman, Jaroslav Hlinka, Dominik Klepl, Anna Pidnebesna, Petra Vidnerová
原文摘要
Symmetry in complex networks is a structural property with broad implications for network function, dynamics, and robustness. While classical approaches rely on exact automorphisms and thus remain highly sensitive to small perturbations, recent work of Liu has introduced approximate formulations that relax the strict invariance condition and recast symmetry detection as an optimization problem over permutations. We investigate how node-level similarity information can be used to bias the search toward structurally plausible vertex correspondences, with network centralities serving as one interpretable instance. Building on this perspective, we adapt simulated annealing, genetic algorithms, tabu search, particle swarm optimization, differential evolution and the quadratic symmetry approximator to this problem, add similarity guidance to each, and introduce Bilinear Approximate Symmetry Inference (BASIN), which learns a soft assignment matrix via bilinear embeddings. Using a unified evaluation pipeline, we compare their guided and unguided variants, BASIN and a random baseline on four synthetic graph families, structural brain connectomes, and three empirical networks with known automorphism groups. Node-similarity guidance is decisive where a planted orbit exists that the unguided search misses, and algorithm-dependent elsewhere, on synthetic and empirical networks alike. Overall, tabu search achieves the strongest performance; its guided variant (SGTS) has the lowest median score among the guided variants in all 44 synthetic conditions, while BASIN is the better of the continuous relaxations on rewired lateral graphs. The effect of node-similarity guidance is therefore regime-dependent rather than uniformly positive.
Mourchid Adegbindin, Brenda Dietrich, Jamol Pender
原文摘要
Mixed-integer programming (MIP) solver runtimes vary substantially across random seeds. Parallel replication exploits this variability but can waste computational resources when applied indiscriminately. We investigate whether information from an ordinary solve can guide this allocation decision. We formulate replication under a finite deadline, establish an exact relationship between runtime dispersion and the value of independent searches, and analyze delayed replication and imperfect allocation decisions. Using 101,615 seeded SCIP runs, we examine the association between root-LP dual degeneracy and replication value. We then introduce a non-invasive root-gated policy that preserves the original solver trajectory and launches three additional seeded processes when a root dual-degeneracy indicator reaches a fixed threshold. In a prospective two-hour experiment on 130 paired instances, root-gated racing proves optimality on 77 instances (59.23%), versus 73 (56.15%) for both Single and unconditional Race-4, while reducing active worker-process occupancy by 43.00% relative to unconditional racing. These findings demonstrate the promise of selective parallel resource allocation without solver-source modifications.
Simplified swarm optimisation (SSO) updates each variable by one draw that copies the global or personal best, keeps the current value or redraws uniformly, with probabilities $c_g$, $c_p$, $c_w$ and $c_r$. As a finite Markov chain under frozen memory, its coordinate kernel $c_w I + (1 - c_w)Π$ has an explicit stationary law and exact convergence rate $c_w^t$. The elitist chain has a geometric hitting-time bound, from a non-optimal start a fixed $c_r$ needs at least $\exp(nc_r/2)$ expected evaluations to reach a unique optimum, and a coupling under changing memories gives $O(n\log n)$ and $O(n^2)$ expected generations on OneMax and LeadingOnes for every population size with redraw share of order $1/n$ and no inertia, and within a $\log n$ factor for fixed inertia. With a frozen memory, inertia cannot speed up first entry into any set from stationary positions, though it can from others. Where the three reference values agree, only $c_r$ acts; where they disagree, copy branches recombine the memories, forcing a redraw only where no active branch holds the target value. Branch logging and checkpoint replay show inertia slowed the tested OneMax and LeadingOnes runs and suggest memory-mediated gains on wider gaps. On Jump functions SSO needed 3.8--8.9 times fewer evaluations than the (1+1) evolutionary algorithm with rate $1/n$, but no consistent advantage over rate-adjusted mutation or a $(μ+1)$ genetic algorithm with uniform crossover was established. The analytical redraw rule improved six of eight benchmark--setting pairs and redundancy allocation, but was slower than $c_r = 0.1$ on Jump$_2$.
Benjamin Bourel, Jean-Christophe Lombardo, Alexis Joly, Mathias Chouet, Thomas Paillot, Antoine Affouard, Vanessa Hequet, Herv{é} Go{ë}au 等 9 位作者
原文摘要
1. Monitoring plant biodiversity requires identifying large numbers of plants across space and time. Although artificial intelligence (AI) can now identify plant species from images with high accuracy, most existing systems rely on users manually submitting images. As ecological monitoring increasingly generates large image collections from cameras, drones, robots and online platforms, there is a growing need for plant identification services that can be directly integrated into automated data-processing workflows. 2. Here, we present Pl@ntNet's machine-to-machine AI plant identification service. This service uses an application programming interface (API) to provide programmatic access to Pl@ntNet's plant recognition models and the associated taxonomic processing. This service enables the implementation of automated identification workflows without human intervention and can thus directly link large image collections and automated observation systems to an AI-based plant identification system. 3. We establish that this service, which covers 85,000 species from around the world, (1) has a species prediction accuracy of 0.89, (2) process images up to 666 times faster than workflows requiring human intervention, (3) and has processed over 84 million identification requests since its launch, including more than 40 million in 2025 alone. We demonstrate how it has already enabled the implementation of a new generation of ecological workflows: identifying plants from images of insects to reconstruct the plant-pollinator interactions, analysing geolocated images from social media to study plant distribution, detecting invasive plants using images taken at the roadside, and identifying weeds using agricultural robots. Finally, we illustrate its application using a reproducible example: identifying plants from insect photos on iNaturalist. In approximately 30 min on a standard laptop, 2,424 images were processed to generate a dataset of 708 plant-pollinator interactions, with a weighted precision of 0.98 for flower detection and an accuracy of 0.88 for genus plant identification. 4. In conclusion, this machine-to-machine species identification is already operating on a considerable scale and meets a genuine need. All of this paves the way for a new paradigm in biodiversity monitoring, in which automated sensors and image collections can be directly connected to AI-based identification services.
We present a suite of 50 unpublished research-level theoretical physics problems with verifiable answers. Without AI assistance, a typical problem in our collection could be solved by a world expert in about a day or less but would take considerably longer for experts in adjacent specialties. We test the ability of off-the-shelf AI agents to tackle these problems autonomously, without human hints. When given access to tools available in a realistic setting, including the ability to search the literature and write scripts for numerical and symbolic work, the best-performing agent, GPT-6-Astra harnessed in Codex, obtains the final answers with a near-perfect accuracy of 98% at an average cost under USD 3 per problem. On average, agents show only small changes in accuracy when denied web search and access to background research papers, suggesting significant memorization of the literature. These results suggest that state-of-the-art agents can accurately perform concrete calculations required across theoretical physics research. We comment on the broader implications of these findings and discuss possible explanations for the current lack of agent-led breakthroughs in theoretical physics despite their outstanding problem-solving abilities.
Xiaokai Li, Ziyu Zhou, Qibin Cheng, Yangcheng Wu, An Zeng, Ying Fan
原文摘要
As large language models increasingly generate mathematical proofs, whether they develop a proving style distinct from that of human mathematicians remains an open question. This question is particularly pressing as mathematicians increasingly voice concerns that proofs generated by large language models are difficult to interpret. Here we ask three large language models to independently reprove all 465 propositions of Euclid's \textit{Elements} and compare their citation choices against the human-proof baseline through Jaccard similarity and complex network analysis. We find that AI-generated proofs differ from human proofs, and this difference is stable and reproducible: mean pairwise Jaccard similarity among AI runs (~0.68) is nearly twice that between AI and human proofs (~0.37), and cross-run classification consistency is also high. By examining citation structure in detail, we identify the concrete signature of this difference: approximately 40% of AI citations point to foundational elements (definitions, common notions, and postulates), compared with only 13\% in human proofs; even among theorem-to-theorem citations, AI tends to cite more distant results. Roughly half of AI's extra citations trace to the upstream dependencies of the human proof. Together, these findings suggest that the AI proving style is a tendency to construct arguments on its own from foundational elements rather than directly citing later, more established results, which we term "building from the ground up". Our findings offer a new perspective on how AI systems organise and invoke mathematical knowledge, with implications for understanding the interpretability of AI-generated proofs.