Peer Revenue Past citations are no guarantee of future returns

Papers

Fresh from arXiv, typeset properly. Highlighted words trade on the market.

All of arXiv

Friday 9 October

Chemical Physics

UGME: A python package for non-Markovian protein folding dynamics

Anthony J. Dominic III, Bertram Cham and Andrés Montoya-Castillo

Biomolecular processes, including protein folding, protein-protein association, and protein-DNA recognition, often occur on timescales beyond what individual atomistic molecular dynamics (MD) simulations can reach. To access these longer-timescale dynamics, practitioners commonly combine many parallelized, short trajectories into reduced kinetic models, such as Markov state models (MSMs), that describe how the biomolecules interconvert between a handful of biomolecular configurations (e.g., folded and unfolded states of a protein). Central to MSM construction is the assumption that, at sufficiently long time lags, these projected dynamics become memoryless, such that the future evolution of the system depends only on its current state. Yet, projecting high-dimensional biomolecular dynamics onto a small set of states necessarily discards information about unresolved degrees of freedom, including solvent reorganization, side-chain rearrangements, and intermediate configurations. Because these unresolved motions can continue to influence subsequent dynamics, the resulting dynamics retain memory that is absent from a Markovian description. Here, we present ugme, a modular, open-source Python package for constructing and validating generalized master equations (GME) models that explicitly account for these non-Markovian (memory) effects.
Read the full abstract

Materials Science

Emergence of quantum transport in SnTe films upon structural refinement

Shrinkhala Sharma, Rabindra Basnet, Thomas Parker et al. (+4 more)

Topological crystalline insulators are distinguished from the conventional Z2 topological insulators by crystalline symmetry protection of the surface states. Quantitative characterization of these states requires magneto-transport measurements from which the phase coherence length and number of independent coherent channels can be extracted. Here we report such measurements on textured thin films of SnTe grown on Bi2Te3 buffered c-plane sapphire (Al2O3) by pulsed laser ablation. A post deposition annealing protocol was implemented to control the crystallographic texture, which reveals improved crystallinity and enhanced epitaxial alignment along the [111] direction in films annealed at elevated temperatures. The higher carrier density and lower resistivity seen in films annealed at 100 oC are suggestive of Sn vacancies which act as native acceptor defects, while films annealed at 200 oC approach the 1:1 stoichiometry. The magnetoresistance of all samples at T < 10 K characterized by a cusp centered around zero-field is suggestive of weak antilocalization and interpreted in the framework of the Hikami-Larkin-Nagaoka theory in the strong spin-orbit coupling limit. At T > 20 K, the MR is small and positive, and deviates from conventional Kohler scaling. An approximate collapse of the MR curves between 50 and 250 K is obtained using an empirical temperature-dependent scaling factor, indicating that changes in the Hall-derived carrier density alone cannot account for the observed scaling behavior. These results demonstrate that post growth annealing strongly influences the interplay between stoichiometry, crystallinity, and coherent transport in SnTe thin films.
Read the full abstract

All authors Shrinkhala Sharma, Rabindra Basnet, Thomas Parker, Afolabi Akinfaderin, Alexander Samokhvalov, Rajeswari Kolagani, Ramesh C. Budhani

Sound

Speech event detector: Timestamped detection of non-verbal vocalizations and background sounds for rich speech transcription

Ayush Kumar Bar and Promila Ghosh

Automatic speech recognition produces words but drops the non-verbal content of a recording: the breaths, coughs, sighs and laughter of the speaker, and the horns, music or barking around them. We describe Speech Event Detector, a 5.8-million-parameter convolutional-Transformer model that outputs probabilities every 40 ms for 54 event classes (20 speaker vocalizations and 34 background sounds), together with post-processing that converts these probabilities into timestamped events and a procedure for merging the events into a Whisper transcript. The model is trained on material drawn from seven public datasets whose labels come in incompatible forms: single detected regions, free-text tags with timestamps, temporally strong ontology labels, symbols embedded in transcripts, isolated recordings and environmental recordings. We unify them using rule-based label mapping, recover timestamps for transcript-embedded labels through CTC forced alignment, locating 6,360 of 8,620 sounds, and combine about 168 hours of real annotated training recordings with synthetic mixtures and augmentation using real background noise. Per-class thresholds for two operating profiles are tuned on real held-out recordings, and performance is estimated using two-fold cross-validation. On whole held-out recordings from four datasets, considering the 48 classes with at least ten held-out events, the sensitive profile detects an average of 49 percent of a class's events, covering 67 percent of all 9,367 events, at a macro precision of 0.30. The balanced profile achieves a macro recall of 0.37 at a precision of 0.36, with macro event F1 scores of 0.32 and 0.33, respectively. The detector runs about 39 times faster than real time on eight CPU threads.
Read the full abstract

Computers and Society

Defective, Amplified, Contested: Folk Conceptualizations of AI Slop

Zhixin Li and Jaime Banks

AI slop has flooded social media feeds with overprocessed and derivative content–Jesus fusing with a shrimp, a humanoid orange chubby cat shoplifting, and a "hero" dog rescuing a human child after a plane crash. This might be no laughing matter, as waves of media coverage have taken a distinctly negative bent, emphasizing its low quality, ethical concerns, and threats to human creativity. Despite extensive attention from politicians and technologists, the perspectives of everyday users–the very audience most exposed to this phenomenon–remain largely unexplored. This descriptive work explored how Reddit users experience, conceptualize and contest AI slop. Inductive thematic analysis of posts and comments across six slop-related subreddits identified four themes covering constitutional defects, algorithmic overreach, user resistance, and entertainment outcomes–ranging from anxieties about eroding agency, moral norms, creativity, and existential questions to defensive distinctions between quality AI content and slop. We derive a user-driven definition: AI slop is audience-contested AI-generated content characterized by constitutional defects and enabled by platform logics. We operationalize slop through a facet-theoretical framework to inform future research into this fuzzy category.
Read the full abstract

Networking

An Availability Management Framework for Microservice-Based Safety-Critical CIoT Systems: Architecture and End-to-End Evaluation

Hassaan Siddiqui and Ferhat Khendek

Safety-critical Cellular Internet of Things (CIoT) systems increasingly rely on microservice architectures orchestrated across heterogeneous edge and cloud platforms. Although this paradigm improves modularity and scalability, the fault-tolerance mechanisms of contemporary orchestration platforms remain predominantly reactive, i.e., failed components are restarted or rescheduled only after an outage has already occurred. Such recovery can produce multi-second End to End Service Outage Time (E2E SOT), which can grow further under container restart backoff. This paper proposes an Availability Management Framework (AMF), a generic monitor-predict-recover loop that complements reactive recovery with machine learning-based anomaly detection and make-before-break switchover to a standby replica. Candidate ML models are evaluated offline under three protocols, including a change of operating regime, in order to select a model for live deployment. AMF is instantiated and evaluated on a three-tier tele-operated driving (ToD) testbed under a staged resource-exhaustion scenario. Relative to Kubernetes-only recovery, AMF completes switchover during the anomaly window, reduces E2E SOT, and avoids the tens-of-seconds outages associated with repeated in-place restarts under backoff. On this non-real-time testbed, residual E2E SOT lies inside healthy-path jitter and is therefore limited by path timing rather than by lateness of the recovery decision. The paper whows that coupling a validated ML based anomaly detector with proactive switchover can remove restart- and backoff-dominated outage from a critical path.
Read the full abstract

Human-Computer Interaction

Who Guides Whom? Negotiating Care, Religious Authority, and Technology in Bangladeshi Muslim Families

Susmita Biswas, Saiful Islam Tuhin, Faria Soroni et al. (+2 more)

Digital technologies are reshaping how families negotiate care, authority, and responsibility, yet these dynamics remain understudied in religiously situated family life. Through interviews with 38 adult members of Bangladeshi Muslim families, we examine how family members guide one another across religious, digital, and AI-mediated contexts. We find that authority did not simply follow age or family hierarchy: older members often remained sources of religious and familial guidance, while younger members increasingly provided expertise around digital risks, information credibility, and AI. Families also actively shaped digital environments and negotiated the role AI could occupy in everyday relationships. We develop the concepts of household epistemic governance, counter-curation, and domain-dependent authority to explain these shifting relations of knowledge and responsibility. Our findings suggest opportunities for family technologies that support negotiated boundaries, credible information practices, and culturally situated forms of care.
Read the full abstract

All authors Susmita Biswas, Saiful Islam Tuhin, Faria Soroni, Sharifa Sultana, Jannatun Noor

Computer Vision

Latent World Action Model for GUI Agents

Seungjun Yu, Hojun Choi, Seojeong Park et al. (+2 more)

Pixel-space world models have recently emerged as a promising approach for GUI agents, predicting future screens prior to execution and subsequently re-perceiving them for action selection. However, this two-stage paradigm faces two key challenges in GUI environments: (L1) the high computational cost of pixel reconstruction and (L2) the need to reconstruct large and diverse appearance-level details beyond what is required for action selection. In practice, this screen-generation objective can induce an identity shortcut problem, where future-state prediction largely preserves the current state rather than capturing action-induced changes. In this work, we reformulate GUI world modeling by shifting the prediction target from full-screen reconstruction to action-induced future information in latent space. Concretely, we leverage a large multimodal model as the latent world model, whose shared representations support two dedicated heads for both action selection and predictive transition modeling. The action head directly translates these latent representations into action selection, while the transition head captures action-conditioned future information. To mitigate the identity shortcut, we further introduce an action-aware contrastive objective that encourages the transition representations to capture discriminative action-induced changes. Extensive experiments across offline and online GUI settings show that our method achieves up to 31.9% and 40.6% relative gains in success rate over pixel-space world models, respectively, while substantially reducing inference latency by up to 98.9%. Our code is available at https://github.com/sjyu001/latent2world.
Read the full abstract

All authors Seungjun Yu, Hojun Choi, Seojeong Park, Jaeyo Shin, Hyunjung Shim

Differential Geometry

Asymptotically conical shrinking ends of the $r$-mean curvature flow: rotational construction and uniqueness

Hilário Alencar, Márcio Batista and Gregório Silva Neto

For $1\leq r\leq n-1$ we study shrinking self-similar ends of the $r$-mean curvature flow, $S_r=c\langle X,N\rangle$ with $c<0$, that are asymptotic to a cone. For every rotational cone of positive slope we construct a smooth embedded rotational end and prove that its profile is the only positive solution of the profile equation with the prescribed slope at infinity. Its angular defect $xu'-u$ is strictly negative, which yields $r$-admissibility, and the profile has a complete asymptotic expansion for $u(x)/x$ in powers of $x^{-(r+1)}$, valid after arbitrary differentiation. The profiles depend real analytically on the slope and, on sufficiently distant exterior regions, foliate the image of the corresponding family. Our main result removes the symmetry assumption: two ends that converge smoothly, with multiplicity one, to the same circular cone coincide outside a compact set. The proof transforms the equation for their difference into an almost-eigenfunction inequality for a drift Laplacian on an auxiliary conical metric and applies Bernstein's asymptotic analysis on weakly conical ends. The argument extends to strictly $r$-admissible cones whose links have constant principal curvatures.
Read the full abstract

Computer Vision

CAPTOR: a vision-language foundation model for Chest-Abdomen-Pelvis Oncologic CT

Séléna Jean-Mactoux, Sophie Loizillon, Ghina Jardali et al. (+3 more)

Purpose : To develop and evaluate CAPTOR, a deep-learning vision-language foundation model for oncologic chest-abdomen-pelvis (CAP) CT interpretation. Materials & Methods : CAPTOR (Chest-Abdomen-Pelvis foundaTion model for Oncologic Radiology) is a contrastive vision-language model, trained using a pretraining strategy for oncologic CAP CT (in single- or multiple-series acquisitions), on a multicenter retrospective dataset from North America, Europe and Asia, predominantly from tumor lesion examinations. CAPTOR was compared with state-of-the-art CT foundation models on three downstream tasks: disease detection, survival prediction using TumorImagingBench and image retrieval. Classification and retrieval were both evaluated on held-out internal and public test sets, stratified by anatomical coverage and indication. Bootstrap resampling was used to estimate confidence intervals, and pairwise comparisons were performed using permutation tests. Results : CAPTOR was trained and tested on 25,512 and 10,794 CT-report pairs respectively. TumorImagingBench datasets comprised between 134 and 421 studies each. CAPTOR outperformed all baseline models, on the classification task, on the CAP subset (AUC 0.767, 95%CI: 0.734-0.793, vs. 0.695, 95%CI: 0.645-0.736 for the next-best, p<0.001) and oncologic subset (AUC 0.784, 95%CI: 0.759-0.805, vs. 0.734, 95%CI: 0.703-0.767 p<0.001), ranked first on three of the four oncologic prognostic benchmarks, and achieved superior retrieval performance on CAP and oncologic subsets. Conclusion : A foundation model trained on oncology-enriched CAP CT data outperformed relevant state-of-the-art models in its target clinical context, underscoring the importance of clinically aligned pretraining. We release our preprocessing, training and inference code : https://github.com/Guerbet-AI/CAPTOR and model weights : https://huggingface.co/Guerbet-AI/CAPTOR
Read the full abstract

All authors Séléna Jean-Mactoux, Sophie Loizillon, Ghina Jardali, Amandine Crombé, Nathalie Lassau, Alexandre Bône

Theoretical Economics

Shaping the Menu: Information Design under Screening

Panagiotis Kyriazis

How should consumers learn when a seller chooses a menu of product qualities and prices? We study information design when the seller observes the learning policy, but not consumers' private signals, before choosing its menu. Under cost-curvature conditions, we characterize optimal learning when consumer surplus receives greater weight than profit. Even with two valuations, every optimal posterior distribution combines an interval of posterior types with an efficiently served atom at the highest valuation. The induced menu features a continuum of differentiated products. Strikingly, optimal learning can also induce a lower posterior atom, separated from this interval by a support gap. Buyers in this atom either purchase an isolated low-quality product or are excluded, even when trade with every buyer creates gains. Both configurations can be optimal for the same prior and welfare weight. Served buyers with distinct posterior types are never bunched at the same quality. General finite priors can yield several separating regions and trading atoms. We also characterize the optimal joint design of learning and valuation-dependent segmentation, showing, among other results, that every posterior type with positive gains from trade must be served. Joint design can raise both consumer surplus and profit relative to buyer-optimal private learning alone. Even random assignment to segments with different learning policies and separate menus can give consumers more surplus than any single market yielding the same profit. Additional price discrimination can therefore benefit consumers, even when privacy restrictions require segmentation to reveal nothing about valuations to the seller.
Read the full abstract

Artificial Intelligence

Harness^2: Recursive Agent Harnessing for an Open World

Ruiyao Xu, Yanfei Chen, Zhongying CuiZhu et al. (+6 more)

LLM agents' capabilities depend not only on their underlying models but also on their execution harnesses. Recursive self-improvement (RSI) offers a promising approach to automatically evolve these harnesses. However, deploying RSI in open-world settings still faces two critical challenges. First, open-world tasks are typically unseen during the harness design phase and novel at test time, leaving no ground-truth verifiers to reliably guide online harness edits. Second, the RSI process remains highly task-specific and fails to generalize to novel scenarios. We introduce Harness^2, a plug-and-play, recursive agent harnessing framework that addresses these challenges through two connected levels of recursion. Task-level recursion refines the current task's harness using a contrastive Propose-Probe-Compose step that reliably guides edits without ground-truth verifiers, and can be scaled further via parallel or sequential recursion. Harnessing-level recursion then extracts experience from these edits to update a continual improvement harness, enabling the agent to better adapt to future tasks. Across three professional benchmarks and four model-harness configurations, a single step consistently improves GLM-5 and Gemini 3.5 Flash over their base harnesses, yielding gains of up to 11.5 points on WorkBuddy-Bench, 4.5 points on JobBench, and 11.8 percentage points on LAB, with further gains as the task-level recursion budget increases.
Read the full abstract

All authors Ruiyao Xu, Yanfei Chen, Zhongying CuiZhu, Bhavana Dalvi Mishra, Yifei Ming, Han Yu, Rujun Han, Chen-Yu Lee, Tomas Pfister

Differential Geometry

An Abresch-Rosenberg type tensor and a gap theorem for CMC hypersurfaces in $\mathbb{S}^n\times\mathbb{R}$ and $\mathbb{H}^n\times\mathbb{R}$

Hilário Alencar, Márcio Batista and Harold Rosenberg

We construct a symmetric trace-free tensor for constant-mean-curvature hypersurfaces in $\mathbb{S}^n\times\mathbb{R}$ and $\mathbb{H}^n\times\mathbb{R}$ that extends the Abresch-Rosenberg tensor. Within its natural class, it is characterized by the linearity of its Codazzi defect, and its complete connected zero set consists exactly of the rotational CMC spheres. An exact integrated Simons identity yields explicit gap theorems, valid for every $h>1$ in $\mathbb{H}^n\times\mathbb{R}$ and every $h>(n\int_0^{\pi/2}\sin^{n-1}r\,dr)^{-1}$ in $\mathbb{S}^n\times\mathbb{R}$. We compare the pinching thresholds with the pointwise algebraic level that implies convexity. In $\mathbb{H}^n\times\mathbb{R}$ the explicit pinching conditions imply strict convexity, so there the method gives an independent proof with explicit constants. In $\mathbb{S}^n\times\mathbb{R}$ the pinching region exceeds the convexity level for $n\in\{3,4\}$ and every admissible $h$, and for $n\geq5$ whenever $h^2\geq n/(6(n+2))$; there the gap theorem does not require convexity to be known in advance. We also determine second-order large-$h$ asymptotics of the thresholds.
Read the full abstract

Medical Physics

Separating active irradiation from passive outcome learning in radiotherapy: a finite-model computational benchmark

Yuntao Wang

Apparent gains from informative irradiation may reflect treatment reoptimization or better use of outcomes from static treatment. We tested whether changing irradiation adds value beyond optimized static treatment, passive outcome learning, and data-dependent stopping. Three synthetic three-dimensional geometries supplied 51 parent fluences. Four known response scenarios generated Bernoulli risks from normalized organ-dose contrasts at uncalibrated slopes of 0, 4, 16, and 64; one unknown scenario applied to all 2,816 hypothetical uses. Loss included acquisition and deployment. We compared fixed passive and locally selected fixed active sampling, passive optimal stopping, and a restricted one-checkpoint, one-switch active policy with a shared terminal menu. Fixed active sampling improved on fixed passive sampling in three of nine nonzero-slope conditions, but never on passive stopping. One-switch adaptation added at most 4.00625e-5 expected model events avoided. With the policies held fixed, two of four increments above the numerical reporting threshold reversed sign under alternative integration; all four became negative at some allowed prior weighting. Adding three generated probes changed none of 72 paired policy values, and none was used on a positive-probability path. The tested probes did not demonstrate stable incremental value beyond optimized passive stopping. Treatment opportunities, stopping permissions, acquisition costs, and input sensitivity must be separated when attributing benefits to active irradiation. These are restricted finite-model results, not evidence about unrestricted active learning or clinical safety and effectiveness.
Read the full abstract

High Energy Physics (Experiment)

Energy-energy correlators within $Z$-tagged jets in proton-proton collisions at $\sqrt{s}$ = 13 TeV

LHCb collaboration: R. Aaij, M. Abdelfatah, A.S.W. Abdelmotteleb et al. (+1187 more)

Energy-energy correlators probe perturbative and nonperturbative QCD regimes of jet formation through momentum-weighted angular correlations between hadrons. First measurements in light-quark-enriched jets are presented for inclusive, like-sign, and opposite-sign charged-hadron pairs, using proton-proton collisions at LHCb. The results support universality of the confinement scale for light-quark-initiated jets across collision systems and the reported transverse momentum scale of the jet. In the charge-dependent measurements, opposite-sign correlations are enhanced as angular separation decreases down to the confinement-scale region, a behavior that is well described by simulations.
Read the full abstract

All authors LHCb collaboration: R. Aaij, M. Abdelfatah, A.S.W. Abdelmotteleb, C. Abellan Beteta, F. Abudinén, T. Ackernley, A.A. Adefisoye, B. Adeva, M. Adinolfi, P. Adlarson, C. Agapopoulou, C.A. Aidala, S. Akar, K. Akiba, H. Al Saleh, P. Albicocco, J. Albrecht, R. Aleksiejunas, F. Alessio, P. Alvarez Cartelle, A.A. Alves Jr, S. Amato, J.L. Amey, Y. Amhis, Z. Amos, L. An, L. Anderlini, P. Andreola, M. Andreotti, S. Andres Estrada, A. Anelli, D. Ao, C. Arata, F. Archilli, Z. Areg, M. Argenton, S. Arguedas Cuendis, L. Arnone, M. Artuso, E. Aslanides, R. Ataíde Da Silva, M. Atzeni, B. Audurier, J.A. Authier, D. Bacher, I. Bachiller Perea, S. Bachmann, M. Bachmayer, J.J. Back, M. Bai, Z.B. Bai, V. Balagura, A. Balboni, W. Baldini, Z. Baldwin, L. Balzani, H. Bao, J. Baptista de Souza Leite, C. Barbero Pretel, M. Barbetti, I.R. Barbosa, W. Barker, R.J. Barlow, M. Barnyakov, S. Baron, S. Barsuk, W. Barter, J. Bartz, S. Bashir, B. Batsukh, P.B. Battista, A. Bavarchee, A. Bay, A. Beck, M. Becker, F. Bedeschi, I.B. Bediaga, N.A. Behling, S. Belin, A. Bellavista, I. Belyaev, G. Bencivenni, E. Ben-Haim, J.L.M. Berkey, R. Bernet, A. Bertolin, L. Bertsch, F. Betti, J. Bex, O. Bezshyyko, S. Bhattacharya, M.S. Bieker, N.V. Biesuz, A. Biolchini, M. Birch, F.C.R. Bishop, A. Bitadze, A. Bizzeti, T. Blake, F. Blanc, J.E. Blank, S. Blusk, J.A. Boelhauve, O. Boente Garcia, T. Boettcher, A. Bohare, C. Bolognani, R.B. Bonacci, A. Bordelius, F. Borgato, S. Borghi, M. Borsato, J.T. Borsuk, E. Bottalico, S.A. Bouchiba, M. Bovill, T.J.V. Bowcock, A. Boyer, C. Bozzi, J.D. Brandenburg, A. Brea Rodriguez, N. Breer, C. Breitfeld, J. Brodzicka, J. Brown, E. Buchanan, M. Burgos Marcos, C. Burr, E. Butera, C. Buti, J.S. Butter, W. Byczynski, S. Cadeddu, H. Cai, Y. Cai, Y. Cai, A. Caillet, R. Calabrese, L. Calefice, M. Calvi, M. Calvo Gomez, P. Camargo Magalhaes, J.I. Cambon Bouzas, P. Campana, A. Campomagnani, A.C. Campos, A.F. Campoverde Quezada, Y. Cao, S. Capelli, M. Caporale, L. Capriotti, R. Caravaca-Mora, A. Carbone, L. Carcedo Salgado, R. Cardinale, A. Cardini, P. Carniti, L. Carus, R. Caspary, G. Casse, M. Cattaneo, G. Cavallero, V. Cavallini, S. Celani, I. Celestino, S. Cesare, A.J. Chadwick, M. Charles, Ph. Charpentier, E. Chatzianagnostou, R. Cheaib, M. Chefdeville, C. Chen, J. Chen, S. Chen, Z. Chen, A. Chen Hu, M. Cherif, S. Chernyshenko, X. Chiotopoulos, G. Chizhik, V. Chobanova, A. Christakakis, M. Chrzaszcz, Y. Chu, V. Chulikov, P. Ciambrone, X. Cid Vidal, P. Cifra, P.E.L. Clarke, M. Clemencic, H.V. Cliff, J. Closier, C. Cocha Toapaxi, V. Coco, A. Codovini, C.S. Codovini, J. Cogan, E. Cogneras, L. Cojocariu, S. Collaviti, P. Collins, T. Colombo, M. Colonna, A. Comerma-Montells, L. Congedo, J. Connaughton, A. Contu, N. Cooke, A. Corallo, G. Cordova, C. Coronel, I. Corredoira, A. Correia, G. Corti, G.C. Costantino, C. Cotirlan, J. Cottee Meldrum, B. Couturier, D.C. Craik, N. Crepet, M. Cruz Torres, M. Cubero Campos, E. Curras Rivera, R. Currie, C.L. Da Silva, X. Dai, J. Dalseno, C. D'Ambrosio, G. Darze, A. Davidson, O. De Aguiar Francisco, C. De Angelis, F. De Benedetti, J. de Boer, K. De Bruyn, S. De Capua, M. De Cian, U. De Freitas Carneiro Da Graca, F. De Gregorio, E. De Lucia, J.M. De Miranda, L. De Paula, A. De Robertis, E. De Santis, M. De Serio, P. De Simone, F. De Vellis, J.A. de Vries, F. Debernardis, D. Decamp, S. Dekkers, L. Del Buono, B. Delaney, B. Demaire-Lepape, J. Deng, O. Deschamps, F. Dettori, B. Dey, P. Di Nezza, S. Ding, Y. Ding, L. Dittmann, J.F. Diverchy, A.D. Docheva, A. Doheny, C. Dong, F. Dordei, J. Dorta Moreno, A.C. dos Reis, J. Dos Santos Oliveira, A.D. Dowling, L. Dreyfus, W. Duan, P. Duda, L. Dufour, V. Duk, P. Durante, M.M. Duras, J.M. Durham, K. Duwe, A. Dziurda, S. Easo, E. Eckstein, U. Egede, S. Eisenhardt, E. Ejopu, L. Eklund, M. Elashri, D. Elizondo Blanco, J. Ellbracht, S. Ely, A. Ene, T. Evans, F. Fabiano, S. Faghih, L.N. Falcao, B. Fang, R. Fantechi, L. Fantini, M. Faria, K. Farmer, F. Fassin, D. Fazzini, L. Felkowski, C. Feng, M. Feng, A. Fernandez Casani, M. Fernandez Gomez, B. Fernandez Rodino, J. Fernandez-John, A.D. Fernez, F. Ferrari, F. Ferreira Rodrigues, R.A. Fini, R. Fiorenza, M. Fiorini, M. Firlej, D.S. Fitzgerald, C. Fitzpatrick, T. Fiutowski, F. Fleuret, A. Fomin, M. Fontana, M. Fontes Vaz, L.A. Foreman, R. Forty, D. Foulds-Holt, V. Franco Lima, M. Franco Sevilla, M. Frank, E. Franzoso, G. Frau, C. Frei, D.A. Friday, J. Fu, Y. Fu, Q. Führing, T. Fulghesu, G. Galati, M.D. Galati, A. Gallas Torreira, D. Galli, S. Gambetta, M. Gandelman, P. Gandini, B. Ganie, H. Gao, R. Gao, T.Q. Gao, Y. Gao, Y. Gao, Y. Gao, L.M. Garcia Martin, P. Garcia Moreno, J. García Pardiñas, P. Gardner, L. Garrido, C. Gaspar, A. Gavrikov, J. George, E. Gersabeck, M. Gersabeck, T. Gershon, S. Ghizzo, Z. Ghorbanimoghaddam, F.I. Giasemis, V. Gibson, H.K. Giemza, A.L. Gilman, M. Giovannetti, A. Gioventù, L. Girardey, M.A. Giza, F.C. Glaser, V.V. Gligorov, C. Göbel, L. Golinka-Bezshyyko, E. Golobardes, A. Golutvin, S. Gomez Fernandez, A.G. Gomez Mongui, W. Gomulka, F. Goncalves Abrantes, I. Gonçales Vaz, M. Goncerz, G. Gong, S. Gong, J.A. Gooding, C. Gotti, E. Govorkova, J.P. Grabowski, L.A. Granado Cardoso, R. Grande Quartieri, E. Graugés, E. Graverini, L. Grazette, G. Graziani, A.T. Grecu, N.A. Grieser, L. Grillo, C. Gu, M. Guarise, L. Guerry, A.-K. Guseinov, Y. Guz, T. Gys, K. Habermann, T. Hadavizadeh, C. Hadjivasiliou, G. Haefeli, C. Haen, S. Haken, G. Hallett, P.M. Hamilton, Q. Han, S. Han, X. Han, S. Hansmann-Menzemer, N. Harnew, T.J. Harris, L. Hartman, M. Hartmann, S. Hashmi, J. He, N. Heatley, A. Hedes, F. Hemmer, C. Henderson, R. Henderson, R.D.L. Henderson, A.M. Hennequin, K. Hennessy, A. Henrot, J. Herd, P. Herrero Gascon, J. Heuel, A. Heyn, A. Hicheur, G. Hijano Mendizabal, J. Horswill, R. Hou, Y. Hou, D.C. Houston, N. Howarth, W. Hu, X. Hu, W. Hulsbergen, R.J. Hunter, D. Hutchcroft, M. Idzik, P. Ilten, A. Iohner, S. Jacevicius, H. Jage, S.J. Jaimes Elles, S. Jakobsen, T. Jakoubek, E. Jans, A. Jawahery, C. Jayaweera, A. Jelavic, V. Jevtic, Z. Jia, E. Jiang, X. Jiang, Y. Jiang, Y.J. Jiang, E. Jimenez Moya, N. Jindal, M. John, A. John Rubesh Rajan, D. Johnson, C.R. Jones, S. Joshi, B. Jost, J. Juan Castella, N. Jurik, I. Juszczak, K. Kalecinska, D. Kaminaris, S. Kandybei, M. Kane, Y. Kang, C. Kar, A. Kauniskangas, J.W. Kautz, M.K. Kazanecki, F. Keizer, M. Kenzie, T. Ketel, B. Khanji, S. Kholodenko, V. Kholoimov, G. Khreich, F. Kiraz, T. Kirn, V.S. Kirsebom, N. Kleijne, A. Kleimenova, D. Klekots, K. Klimaszewski, M.R. Kmiec, T. Knospe, R. Kolb, S. Koliiev, L. Kolk, A. Konoplyannikov, P. Kooij, P. Kopciewicz, P. Koppenburg, A. Korchin, I. Kostiuk, O. Kot, S. Kotriakhova, E. Kowalczyk, O. Kravcov, M. Kreps, W. Krupa, W. Krzemien, O. Kshyvanskyi, S. Kubis, M. Kucharczyk, A. Kupsc, A. Kurzina, V. Kushnir, B. Kutsenko, J. Kvapil, I. Kyryllin, D. Lacarrere, P. Laguarta Gonzalez, A. Lai, A. Lampis, D. Lancierini, C. Landesa Gomez, G. Lanfranchi, C. Langenbruch, T. Latham, F. Lazzari, C. Lazzeroni, R. Le Gac, H. Lee, R. Lefèvre, M. Lehuraux, C. Lemettais, E. Lemos Cid, O. Leroy, T. Lesiak, E.D. Lesser, B. Leverington, A. Li, C. Li, C. Li, H. Li, J. Li, K. Li, L. Li, L. Li, P. Li, P.-R. Li, Q. Li, T. Li, T. Li, W. Li, Y. Li, Y. Li, Y. Li, Z. Li, Z. Lian, Q. Liang, X. Liang, Z. Liang, S. Libralon, A. Lightbody, J. Lin, S. Lin, T. Lin, R. Lindner, H. Linton, R. Litvinov, D. Liu, F.L. Liu, G. Liu, K. Liu, S. Liu, W. Liu, X. Liu, Y. Liu, Y. Liu, Y.L. Liu, G. Loachamin Ordonez, I. Lobo, A. Lobo Salvia, A. Loi, T. Long, F.C.L. Lopes, J.H. Lopes, A. Lopez Huertas, C. Lopez Iribarnegaray, Q. Lu, C. Lucarelli, D. Lucchesi, M. Lucio Martinez, Y. Luo, A. Lupato, M. Lupberger, E. Luppi, K. Lynch, J. Lyu, S. Lyu, X.-R. Lyu, H. Ma, S. Maccolini, F. Machefert, F. Maciuc, B. Mack, I. Mackay, L.M. Mackey, L.R. Madhan Mohan, M.J. Madurai, D. Magdalinski, J.J. Malczewski, S. Malde, L. Malentacca, G. Manca, C. Mancuso, R. Manera Escalero, A. Mangalasseri, F.M. Manganella, R. Mangrulkar, D. Manuzzi, S. Mao, D. Marangotto, J.F. Marchand, R. Marchevski, U. Marconi, L. Mareso, E. Mariani, S. Mariani, C. Marin Benito, J. Marks, A.M. Marshall, L. Martel, G. Martelli, G. Martellotti, L. Martinazzoli, M. Martinelli, C. Martinez, A. Martinez Armas, D. Martinez Gomez, D. Martinez Santos, F. Martinez Vidal, A. Martorell i Granollers, A. Massafferri, R. Matev, A. Mathad, C. Matteuzzi, K.R. Mattioli, L. Matzner, A. Mauri, E. Maurice, J. Mauricio, P. Mayencourt, J. Mazorra de Cos, M. Mazurek, D. Mazzanti Tarancon, M. McCann, N.T. McHugh, A. McNab, R. McNulty, B. Meadows, S.E.R. Medaer, D. Melnychuk, D. Mendoza Granada, P. Menendez Valdes Perez, F.M. Meng, M. Merk, A. Merli, L. Meyer Garcia, D. Miao, H. Miao, S. Mico, M. Mikhasenko, D.A. Milanes, A. Minotti, E. Minucci, B. Mitreska, D.S. Mitzel, R. Mocanu, A. Modak, L. Moeser, R.D. Moise, E.F. Molina Cardenas, T. Mombächer, M. Monk, T. Monnard, S. Monteil, A. Morcillo Gomez, G. Morello, M.J. Morello, M.P. Morgenthaler, A. Moro, J. Moron, W. Morren, A.B. Morris, A.G. Morris, R. Mountain, Z. Mu, N. Muangkod, E. Muhammad, F. Muheim, M. Mulder, K. Müller, F. Muñoz-Rojas, V. Mytrochenko, P. Naik, T. Nakada, R. Nandakumar, G. Napoletano, I. Nasteva, M. Needham, N. Neri, S. Neubert, N. Neufeld, J. Nicolini, D. Nicotra, E.M. Niel, L. Nisi, Q. Niu, B.K. Njoki, P. Nogarolli, P. Nogga, J. Nombela Royo, C. Normand, A. Novo Cal, J. Novoa Fernandez, G. Nowak, H.N. Nur, A. Oblakowska-Mucha, T. Oeser, O. Okhrimenko, R. Oldeman, N. Oldman, F. Oliva, E. Olivart Pino, M. Olocco, R.H. O'Neil, J.S. Ordonez Soto, D. Osthues, J.M. Otalora Goicochea, A. Oyanguren, O. Ozcelik, F. Paciolla, A. Padee, K.O. Padeken, B. Pagare, T. Pajero, A. Palano, L. Palini, L. Palombini, M. Palutan, C. Pan, X. Pan, S. Panebianco, S. Paniskaki, L. Paolucci, A. Papanestis, M. Pappagallo, L.L. Pappalardo, C. Pappenheimer, C. Parkes, D. Parmar, G. Passaleva, D. Passaro, A. Pastore, M. Patel, J. Patoc, C. Patrignani, A. Paul, C.J. Pawley, A. Pellegrino, J. Peng, X. Peng, M. Pepe Altarelli, S. Perazzini, H. Pereira Da Costa, M. Pereira Martinez, A. Pereiro Castro, C. Perez, A. Perez Casas, P. Perret, A. Perrevoort, A. Perro, M.J. Peters, A. Petkovic, K. Petridis, A. Petrolini, S. Pezzulo, J.P. Pfaller, H. Pham, L. Pica, E. Picatoste Olloqui, M. Piccini, L. Piccolo, F. Piernas Diaz, B. Pietrzyk, R.N. Pilato, D. Pinci, F. Pisani, M. Pizzichemi, V.M. Placinta, M. Plo Casasus, T. Poeschl, F. Polci, M. Poli Lener, A. Poluektov, I. Polyakov, E. Polycarpo, S. Ponce, D. Popov, K. Popp, K. Prasanth, C. Prouve, D. Provenzano, V. Pugatch, A. Puicercus Gomez, G. Punzi, J.R. Pybus, Q. Qian, W. Qian, N. Qin, R. Quagliani, R.I. Rabadan Trejo, B. Rachwal, R. Racz, J.H. Rademacker, M. Rama, M. Ramírez García, V. Ramos De Oliveira, M. Ramos Pernas, G. Ramsey, M.S. Rangel, G. Raven, M. Rebollo De Miguel, F. Redi, J. Reich, F. Reiss, Z. Ren, P.K. Resmi, M. Ribalda Galvez, R. Ribatti, G. Ricart, D. Riccardi, S. Ricciardi, K. Richardson, M. Richardson-Slipper, F. Riehn, K. Rinnert, P. Robbe, G. Robertson, E. Rodrigues, A. Rodriguez Alvarez, E. Rodriguez Fernandez, J.A. Rodriguez Lopez, E. Rodriguez Rodriguez, J. Roensch, A. Rogovskiy, D.L. Rolf, P. Roloff, A. Romano, V. Romanovskiy, A. Romero Vidal, G. Romolini, F. Ronchetti, T. Rong, W. Rose, M. Rotondo, M.S. Rudolph, G. Ruggiero, M. Ruiz Diaz, J. Ruiz Vidal, J. Ruz Armendariz, J.J. Saavedra-Arias, J.J. Saborido Silva, D. Sahoo, N. Sahoo, B. Saitta, M. Salomoni, I. Sanderswood, R. Santacesaria, C. Santamarina Rios, M. Santimaria, L. Santoro, E. Santovetti, A. Saputi, A. Sarnatskiy, G. Sarpis, M. Sarpis, C. Satriano, A. Satta, M. Saur, H. Sazak, F. Sborzacchi, A. Scarabotto, S. Schael, S. Scherl, M. Schiller, H. Schindler, M. Schmelling, B. Schmidt, N. Schmidt, S. Schmitt, H. Schmitz, O. Schneider, A. Schopper, N. Schulte, H. Schumacher, M.H. Schune, G. Schwering, B. Sciascia, A. Sciuccati, G. Scriven, I. Segal, S. Sellam, M. Senghi Soares, A. Sergi, N. Serra, L. Sestini, B. Sevilla Sanjuan, Y. Shang, D.M. Shangase, R.S. Sharma, L. Shchutska, T. Shears, S. Shelton, J. Shen, Z. Shen, S. Sheng, B. Shi, J. Shi, Q. Shi, W.S. Shi, E. Shmanin, R. Silva Coutinho, G. Simi, S. Simone, M. Singha, I. Siral, N. Skidmore, T. Skwarnicki, M.W. Slater, E. Smith, M. Smith, L. Soares Lavra, M.D. Sokoloff, F.J.P. Soler, A. Solomin, K. Solovieva, N.S. Sommerfeld, R. Song, Y. Song, Y. Song, Y.S. Song, F.L. Souza De Almeida, G. Souza De Castro, B. Souza De Paula, K.M. Sowa, E. Spadaro Norella, E. Spedicato, J.G. Speer, P. Spradlin, F. Stagni, M. Stahl, S. Stahl, S. Stanislaus, M. Stefaniak, O. Steinkamp, F. Suljik, J. Sun, L. Sun, M. Sun, D. Sundfeld, P. Svihra, V. Svintozelskyi, J. Swallow, K. Swientek, F. Swystun, A. Szabelski, T. Szumlak, Y. Tan, Y. Tang, Y.T. Tang, M.D. Tat, J.A. Teijeiro Jimenez, F. Terzuoli, F. Teubert, E. Thomas, D.J.D. Thompson, A.R. Thomson-Strong, R. Thornton, H. Tilquin, V. Tisserand, S. T'Jampens, M. Tobin, T.T. Todorov, L. Tomassetti, G. Tonani, X. Tong, T. Tork, L. Torlai, L. Toscano, D.Y. Tou, G. Tuci, N. Tuning, L.H. Uecker, A. Ukleja, A. Upadhyay, B. Urbach, A. Usachov, U. Uwer, V. Vagnoni, A. Vaitkevicius, A. Valassi, V. Valcarce Cadenas, G. Valenti, N. Valls Canudas, J. van Eldik, H. Van Hecke, E. van Herwijnen, C.B. Van Hulse, R. Van Laak, M. van Veghel, P. Varrella, R. Vazquez Gomez, P. Vazquez Regueiro, C. Vázquez Sierra, S. Vecchi, J. Velilla Serna, J.J. Velthuis, M. Veltri, A. Venkateswaran, M. Verdoglia, M. Vesterinen, W. Vetens, D. Vico Benet, P. Vidrier Villalba, M. Vieites Diaz, X. Vilasis-Cardona, E. Vilella Figueras, A. Villa, P. Vincent, B. Vivacqua, F.C. Volle, D. vom Bruch, K. Vos, C. Vrahas, J. Wagner, J. Walsh, N. Walter, E.J. Walton, G. Wan, A. Wang, B. Wang, C. Wang, C. Wang, C. Wang, G. Wang, H. Wang, J. Wang, J. Wang, J. Wang, J. Wang, M. Wang, N.W. Wang, X. Wang, X. Wang, X. Wang, X.W. Wang, Y. Wang, Y. Wang, Y. Wang, Y.H. Wang, Z. Wang, Z. Wang, J.A. Ward, M. Waterlaat, N.K. Watson, D. Websdale, Y. Wei, Z. Weida, J. Wendel, B.D.C. Westhenry, A.S. White, C. White, M. Whitehead, E. Whiter, A.R. Wiederhold, D. Wiedner, M.A. Wiegertjes, C. Wild, G. Wilkinson, M.K. Wilkinson, M. Williams, M.J. Williams, M.R.J. Williams, R. Williams, S. Williams, Z. Williams, F.F. Wilson, M. Winn, W. Wislicki, M. Witek, L. Witola, T. Wolf, E. Wood, G. Wormser, S.A. Wotton, H. Wu, J. Wu, T. Wu, X. Wu, Y. Wu, Z. Wu, K. Wyllie, S. Xian, Z. Xiang, Y. Xie, T.X. Xing, A. Xu, L. Xu, M. Xu, R. Xu, Z. Xu, Z. Xu, Z. Xu, Z. Xu, S. Yadav, K. Yang, X. Yang, Y. Yang, Y. Yang, Z. Yang, Z. Yang, H. Yeung, H. Yin, X. Yin, C.Y. Yu, J. Yu, K. Yu, X. Yuan, Y Yuan, S. Zalambani, J.A. Zamora Saa, F. Zangari, M. Zavertyaev, M. Zdybal, F. Zenesini, C. Zeng, M. Zeng, S.H Zeng, C. Zhang, C. Zhang, D. Zhang, J. Zhang, L. Zhang, Q.Z. Zhang, R. Zhang, S. Zhang, S.L. Zhang, Y. Zhang, Z. Zhang, J. Zhao, M. Zhao, Y. Zhao, A. Zhelezov, S.Z. Zheng, X.Z. Zheng, Y. Zheng, T. Zhou, X. Zhou, V. Zhovkovska, L.Z. Zhu, X. Zhu, X. Zhu, Y. Zhu, V. Zhukov, J. Zhuo Gao, T. Zies, D. Zuliani, X. Zuo

Materials Science

Anisotropic pressure response of thermal conductivity in crystalline polytetrafluoroethylene

Maissa Adi, Jamal Abou Haibeh and Samuel Huberman

Compression can change thermal conductivity differently along polymer chains and across them by altering phonon velocities and lifetimes. We examine this directional response in orthorhombic Form III polytetrafluoroethylene (PTFE) at calculated pressures of 1.503, 3.452, and 5.056 GPa, using first-principles harmonic and cubic force constants within the single-mode relaxation-time approximation (SMRTA). At 100, 200, and 300 K, the chain-axis conductivity decreases at every pressure step. Between 3.452 and 5.056 GPa, both transverse components recover: at 300 K they increase by 23.7% and 13.7%, while the longitudinal conductivity decreases by 14.5%. An exact factorization separates the conductivity into a heat-capacity–velocity factor and a weighted effective lifetime. The transverse recovery follows from larger heat-capacity-weighted squared velocities, consistent with stiffer interchain restoring forces. Along the chains, shorter effective lifetimes dominate, and both conductivity-weighted Normal and Umklapp rates increase at 300 K. Comparing the same cubic force-constant blocks at all three pressures, we find that interchain coupling is already substantial at 1.503 GPa and increases under compression.
Read the full abstract

Machine Learning

Learnable Low-Rank Polynomial Sketch for Effective Linear Attention

Zongyue Qin, Haoran Deng, Jason Cong et al. (+1 more)

Softmax attention in Transformers suffers from quadratic complexity in sequence length, making it impractical for long-context applications. Linear attention alleviates this issue by replacing the exponential kernel with alternative functions that enable linear-time computation. Among existing linear attention approaches, recent studies have shown that polynomial kernels are particularly effective, as they exhibit sparse and spiky behavior similar to softmax attention, which emphasizes large dot products while suppressing irrelevant interactions. However, the exact computation of high-degree polynomial kernels is infeasible for high-dimensional representations. As a result, prior work relies on approximate polynomial kernels. This introduces a non-negligible approximation error. In this paper, we show that, from the perspective of polynomial kernel approximation, existing linear attention methods are still suboptimal. We propose Learnable Low-Rank Polynomial Sketch (LLoPS), a principled and flexible framework for approximating polynomial kernels with linear attention. Our method learns a low-rank polynomial sketch that provably achieves a strictly smaller approximation error than existing approaches. Experiment results show that LLoPS achieves the highest performance across extensive benchmarks, comparing with various linear attention baselines.
Read the full abstract

All authors Zongyue Qin, Haoran Deng, Jason Cong, Yizhou Sun

Geometric Topology

Tight surgeries on torus knots

John B. Etnyre, Hyunki Min, Bülent Tosun et al. (+1 more)

We give a complete classification of tight contact structures on manifolds obtained from any surgery along any torus knot, which provides the first such classification on an infinite family of knots. The main tools are the classification of non-loose torus knots, bordered contact invariants via their immersed curve interpretations, and non-thickenable tori in knot complements. As a byproduct of this classification, we also give the first classification of contact structures on some surface bundles over the circle with fibers of genus larger than $1$. The symplectic fillability of these contact structures is also studied and leads to, among other things, the disproof of a conjecture of Ghiggini and Van Horn-Morris on Stein fillability of some contact structures on a Brieskorn sphere.
Read the full abstract

All authors John B. Etnyre, Hyunki Min, Bülent Tosun, Konstantinos Varvarezos

Combinatorics

The Path Partition Conjecture and 3-(quasi-)Transitive Digraphs

Shamik Ghosh and Jing Huang

A digraph $D = (V,E)$ is 3-transitive (respectively, 3-quasi-transitive) if whenever $(u,v,w,z)$ is a directed path of length 3, $(u,z) \in E$ (respectively, $(u,z) \in E$ or $(z,u) \in E$). These digraphs extend the concepts of transitive and quasi-transitive digraphs, respectively. For 3-transitive and 3-quasi-transitive digraphs, only the strong case has been characterized. In this paper, we introduce the notion of semi-acyclic digraphs and use it to describe completely the structure of 3-transitive and 3-quasi-transitive digraphs. As an application of this type of structure description, we prove that the path partition conjecture holds for the class of extended semi-acyclic digraphs, which generalizes two previously known results on the conjecture.
Read the full abstract

Software Engineering

Cortex: Measurement-Gated Evolution and Pruning of Coding-Agent Skills and Rules

Samuele Giampieri

Coding agents such as Claude Code also follow text files that developers write for them: a project file, skills the agent may choose to open, and rules that load automatically when the agent reads matching files. We call these files the agent's harness. Developers add to it whenever they correct the agent, but one success proves little: the same agent can pass a task once and fail it the next time. We present Cortex, a tool that tests changes to the harness instead of trusting them. It saves fixes from the developer's sessions as replayable tasks, finds the corrections that recur, writes one skill or rule at a time, and keeps it only if repeated runs with and without it show a net gain and no task made noticeably worse. Existing text is tested the same way before it is deleted, and Cortex neither adds nor deletes text the agent never used. We tested Cortex with Claude Haiku 4.5 on a repository built for the purpose, with conventions the model tends to break, one area with nothing to learn, and a script playing the developer. On held-out tasks, the harnesses it built made the agent follow the conventions far more often than no skills or rules did, left the comparison area unchanged, and added no text read on every turn. Rules, which load automatically, did most of the work; skills helped only when offered with the files a task touches, and those offered on every turn were rarely opened. Cortex kept none of the useless or harmful suggestions we planted, but the agent never opened the useless ones or the harmful skills, and it reached no verdict on the harmful rules, so the test shows that Cortex keeps no unused text, not that it detects damage. Under the stronger Claude Sonnet 5 a single run's harness added little, and on a real open-source project, where the agent was rarely corrected, Cortex found nothing to learn. We release the tool, the lab and its data.
Read the full abstract

Cryptography and Security

Physical Side-Channel Attacks on Edge LLMs: Extracting Output Tokens and User Prompts

Najeh Alhalawani, Gunes Acar, Zhuoran Liu et al. (+1 more)

The deployment of large language models (LLMs) on edge devices is largely motivated by privacy concerns over sharing sensitive user data with cloud servers. In this paper, we demonstrate that keeping inference local does not eliminate privacy risks: an adversary with physical access to the device may exploit hardware leakage to recover sensitive information without compromising or modifying the software system. In particular, we present an electromagnetic (EM) side-channel attack targeting LLM inference on edge devices. Targeting LLaMA-3.2-1B-Instruct running on a Raspberry Pi 5, we demonstrate that EM traces collected during inference can leak both generated tokens and user prompts. We propose three attacks under different threat assumptions. First, a profiling attack that identifies generated tokens among 50 medical and privacy-related keywords with 97.2% accuracy on profiled words. Second, a reconstruction attack recovers the model's hidden representations and uses them to predict tokens from the full 128k-token vocabulary, achieving 94.3% accuracy on profiled classes. For unseen words, the correct token is ranked within the first 100 positions of the 128k-token vocabulary in 67% of cases, with a median rank of 54. Finally, we present a hidden-state inversion attack that uses the recovered hidden state to reconstruct user prompts semantically. Our proposed attacks require only passive EM side-channel measurements and a limited number of traces during model training. These results demonstrate that local execution alone may not guarantee the confidentiality of generated tokens and user inputs.
Read the full abstract

All authors Najeh Alhalawani, Gunes Acar, Zhuoran Liu, Lejla Batina

Machine Learning

Embedding Graph Kernel Entropy (EGKE): Fine-grained Hallucination Detection for LLMs from Embeddings

Dilrangi Sankalpana, Pragatheeswaran Vipulanandan, Uthayasanker Thayasivam et al. (+1 more)

Hallucination detection is critical for ensuring the reliability of Large Language Models (LLMs) in real-world deployments. While recent works, such as Semantic Entropy (SE) and Kernel Language Entropy (KLE), identify hallucinations by measuring semantic disagreement across multiple generated responses using Natural Language Inference (NLI) models, these methods depend on separately loaded NLI classifiers to estimate pairwise semantic similarity, introducing significant computational overhead. We propose Embedding Graph Kernel Entropy (EGKE), a variant of Kernel Language Entropy (KLE) that replaces its NLI-derived semantic graph with a Radial Basis Function (RBF) kernel applied to dense sentence embeddings generated by a BGE embedding model, while retaining KLE's graph Laplacian, heat kernel, and von Neumann entropy formulation unchanged. By computing similarities directly in a continuous embedding space, the proposed method provides a symmetric and computationally efficient alternative to NLI-based semantic comparison. We evaluate our approach across 30 experimental setups (covering multiple question-answer datasets and open-weight LLM families) using both AUROC and AUARC, and demonstrate that this RBF-based method achieves competitive hallucination detection performance against state-of-the-art methods.
Read the full abstract

All authors Dilrangi Sankalpana, Pragatheeswaran Vipulanandan, Uthayasanker Thayasivam, Kamal Premaratne

Data Structures and Algorithms

Many Scores Force Many Reads: An Optimality Proof for Score of Parentheses

Andrew Au

LeetCode Problem 856, Score of Parentheses, admits a one-pass algorithm using $O(n)$ time and $O(1)$ auxiliary registers. Its correctness follows from a local interpretation: every occurrence of () contributes a power of two determined by its nesting depth. The space bound is immediately optimal in the usual programming-challenge model, but the linear-time bound is less obvious. In particular, a direct adversary that forces an algorithm to inspect every character is difficult to maintain under the promise that the input is balanced. We instead prove a counting lower bound. For every $k$, we construct $2^k$ balanced parenthesis strings of the same length $4k+2$, all having distinct scores. The construction encodes each bit using one of two operations, changing a current score $v$ to either $2v+1$ or $2v+2$. Consequently, any deterministic character-query decision tree computing the score must have depth at least $k$. A two-character padding argument extends the result to every even input length, proving an $\Omega(n)$ worst-case time lower bound. The one-pass algorithm is therefore asymptotically optimal in the unit-cost RAM model.
Read the full abstract

Computation and Language

Modeling the Arrival Time of Information-Bearing Events in Natural Language

Leyi Lao, Yulin Ou, Yang Xu et al. (+1 more)

Understanding how information is distributed over time is fundamental to modeling natural language. While existing empirical studies primarily rely on post-hoc analysis of information-theoretic metrics, such as surprisal, little is known about the generation processes under the surface. We propose a novel approach for addressing this gap using the tool of neural temporal point processes (TPPs) by modeling the key information-bearing events in text sequences with learnable event intensity functions. Our model takes the history of event intervals and the contextual semantic representations from LLMs as input, and predicts the future arrival times for two types of events: new-word and high-surprisal-token. The experimental results reveal an interesting distinction in event intensities: LLMs tend to produce new words at significantly higher rates than humans, while high-surprisal tokens do not exhibit the difference. A decaying pattern of intensity over time is found in new-word events for both humans and LLMs, but not in high-surprisal-token events. Our findings revisit and extend previous studies that use simpler methods; provide supplementary evidence for theories about constrained context representation inherent in human language production; and find a new angle to view human and LLM generation differences.
Read the full abstract

All authors Leyi Lao, Yulin Ou, Yang Xu, Hao An

High Energy Physics (Phenomenology)

Factorisation theorem and pole cancellation in the resolved $Q_1^q$-$Q_{7\gamma}$ contribution to $\bar{B}\to X_s\gamma$ at NLO

Riccardo Bartocci and Tobias Hurth

We study the subleading resolved $Q_1^q$-$Q_{7\gamma}$ contribution to the inclusive decay $\bar{B}\to X_s\gamma$ at next-to-leading order (NLO) in $\alpha_s$. We present a formulation of the factorisation theorem which is valid beyond the leading order (LO). On this basis, we explicitly demonstrate the cancellation of all $1/\varepsilon$ singularities between the hard, hard-collinear, anti-hard-collinear and soft contributions. For this purpose, a central result of this work is the computation of the $u$-integrated massless two-loop anti-hard-collinear jet function, which determines the structure of the $1/\varepsilon$ singularities. In addition, we compute the hard function at the required order, combine our results with the hard-collinear jet function known from the literature, and employ the shape function whose renormalisation was determined in our previous work. There, we showed, for inclusive as well as for exclusive modes, that certain contributions to these multi-light-cone soft functions are physically irrelevant and can be consistently omitted; the explicit cancellation of the corresponding divergences provides a non-trivial confirmation of this statement. Overall, our findings constitute a non-trivial consistency check of the factorisation framework for this subleading resolved contribution at NLO and provide the basis for a full phenomenological analysis.
Read the full abstract

Machine Learning

Agreement versus Softmax under Spatial Disambiguation

Michał Zakrzewski, Mariia Shpir, Taras Rumezhak et al. (+1 more)

ScanRefer Multiple queries require spatial disambiguation: several instances of the target class are present, and the referencer must pick one. When per-query logits were never logged, a common substitute is cross-epoch agreement. On the full ScanRefer validation set (9,508 queries; GeoClassifier with ground-truth proposals), object-level softmax is similarly miscalibrated on Unique and Multiple (ECE 0.068 vs. 0.102; the scene-bootstrap gap interval includes zero). Fixed-final agreement on the same checkpoints reports ECE 0.041 vs. 0.272 (gap +0.231). Modal-vote agreement is 0.045 vs. 0.318. Most Multiple errors are same-class wrong-instance (45.6%); geometric miss (H2) is 59.6%, and wrong-category (H1) is 16.7%. A non-monotone H1 pattern does not appear under the default fixed-final protocol, and we retract it. Logit-free thresholds rank spatially hard queries poorly. If only agreement was logged, calibrate abstention on Multiple and report wrong-instance rates with H1.
Read the full abstract

All authors Michał Zakrzewski, Mariia Shpir, Taras Rumezhak, Volodymyr Karpiv

Artificial Intelligence

Should SLMs be in the Driver's Seat? A Study of Local Models as Interpreters in Cloud-Written Agent Programs

Boris Radovič, Marco Canini and Veljko Pejović

The most accurate agents executing user-provided tasks are driven by large, cloud-hosted models, which solve them by reading the private data of the user. Agents powered by models small enough to run on edge devices keep that data local, but yield far worse results. A third option is to have cloud models write a program that devices execute: the data stays local, but as the program is written before any data is seen, it fails whenever a step depends on interpreting content that appears only at run time. We explore an intermediate architecture, in which the cloud-written program defers bounded questions over user data to a small local model through an ordinary function call. On AppWorld this improves the success rate of a code-only agent in all six combinations of backbone model and test partition we evaluate, by up to 14.9pp, recovering a substantial part of the gap to a cloud-hosted ReAct loop while keeping user data on the device.
Read the full abstract

Information Theory

Sequential Change Point Detection for PSK Modulation Switching

Gopal Kataria, Davin Sequeira and Arti Yardi

Transmitters in adaptive communication systems may switch their modulation scheme at an unknown time, and a receiver must detect this switch quickly without raising frequent false alarms. Existing modulation classification methods label a finite block of received data and do not provide stopping rules with controlled false-alarm and delay performance. We study the sequential detection of a switch from one phase shift keying (PSK) scheme to another over an additive white Gaussian noise channel. We build cumulative-sum (CUSUM) and moving-sum (MOSUM) detectors from the log-likelihood ratio and its low-SNR approximation. For a switch from quadrature PSK to binary PSK, we derive the moment-generating function and probability density function of the approximate statistic. We use them to relate the CUSUM threshold to the average run length to false alarm and the worst-case detection delay. For a switch between any two PSK orders, we approximate the vector of overlapping MOSUM statistics as multivariate Gaussian and obtain bounds on both measures. Monte Carlo simulations show that our derived expressions for ARL2FA and WADD closely match the empirical results across thresholds, validating the accuracy of the proposed analytical approximations and theoretical bounds.
Read the full abstract

Computer Vision

Token-Budgeted Candidate Retrieval for 3D Scene-Grounded VLMs: Category Names vs. Pairwise Geometry

Michał Zakrzewski, Mariia Shpir and Taras Rumezhak

Under a fixed context budget, which objects and relations should a 3D scene-grounded VLM see first? On ScanNet indoor referring benchmarks (Dai et al., 2017), matching the query to object category names outperforms pairwise geometry scoring at every reported K. The gaps are largest at high recall. Analytic pairwise cues beat per-object cue scoring and query-agnostic kNN graphs, including on paraphrases. MiniLM and BGE-small still score higher on those splits. We evaluate token-budgeted candidate retrieval with axis-aligned boxes, a token-ratio metric, training-free baselines, two small encoders that match q to each object's category name, and QCRR (Query-Conditioned Relation Retrieval), a deterministic pairwise-geometry scorer. Across ScanRefer gt_val plus a custom 20% scan-id split of Nr3D and Sr3D (35,504 queries; not official ReferIt3D val), additive noun matching reaches R@10 0.864 on ScanRefer against QCRR's 0.626; MiniLM and BGE-small reach 0.911 to 0.913 on ScanRefer and 0.941 to 0.947 on Nr3D. A corrected 3DGraphLLM harness (native Acc@0.25 0.67 on the full graph, n=300 ScanRefer queries) shows that context selection changes downstream accuracy under a matched object budget (Random 0.44, QCRR 0.56, MiniLM 0.67, oracle 0.70). Semantic retrieval beats hand-designed geometry on that harness. We will release the protocol, retriever implementations, and per-query outputs so new selectors can be swapped in without rewriting the analysis.
Read the full abstract

Statistics Applications

Changing Geographies of Violence in North and West Africa

Steven M. Radil, Olivier J. Walther and Mohammed Sanda

This paper analyses spatial trends in armed conflicts in North and West Africa. It shows that all types of violence saw a significant increase in 2025. However, these violent incidents are distributed very unevenly across the region: while conflict is at its lowest in North Africa, the vast majority of violent incidents and victims are concentrated in five West African countries: Nigeria, Cameroon, Mali, Burkina Faso, and Niger. Coastal countries are experiencing a very mixed picture: jihadism is not advancing as a front, from the Sahel toward the Gulf of Guinea, and most of the violence stems from local disputes, ethnic conflicts, and banditry. Second, the spatial indicator developed by the OECD/SWAC to track the evolution of this violence confirms that the conflict landscape is expanding at its margins and intensifying at its center. Even more concerning is that peripheral areas are increasingly being drawn directly into well-established conflict zones, rather than going through a gradual, low-intensity transition phase. The geography of expansion is shifting from a diffuse, low-intensity spread toward peripheral areas to a rapid outward expansion from dense, well-established conflict hotspots. The local indicator used in the report suggests that violence continues to spread to new locations at the margins of established conflict zones. At the heart of these zones, tensions are intensifying relative to their historical baseline levels, and a growing number of regions have experienced clashes for a decade or more. Projections indicate that the intensity of violence is expected to increase in 2026, reaching more than 15,000 incidents and more than 32,500 deaths. Much of the increase is explained by the growing number of conflicts in Nigeria. In the Sahel, fatalities should increase in Mali and decrease in Burkina Faso.
Read the full abstract

Machine Learning

Refined Grid Sampling for Efficient Implicit Neural Representations of Large-Scale Scientific Data

Guang Zhao, Xihaier Luo, Sean McCue et al. (+5 more)

Implicit neural representations (INRs) offer a powerful framework for modeling high-dimensional physical fields, but their training remains inefficient because evaluating dense coordinate grids is computationally expensive. Existing coordinate-pruning strategies rely on local error functions that can oversample correlated points and incur substantial overhead from full-grid error evaluation. We introduce Refined Grid Sampling (RGS), which combines stratified grid sampling, error-based refinement, and dynamic grid-resolution adjustment. This design reduces spatial redundancy without evaluating errors over the full grid. On high-resolution Navier–Stokes fields, RGS achieves higher reconstruction accuracy within a fixed wall-clock budget and reaches the accuracy of full-grid training up to $6\times$ faster than the evaluated baselines.
Read the full abstract

All authors Guang Zhao, Xihaier Luo, Sean McCue, Luke P. Van Roekel, Seungjun Lee, Yihui Ren, Shinjae Yoo, Wei Xu