Peer Revenue Equity in science

Papers

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

All of arXiv

Friday 28 August

Probability

A Two-regime Khintchine Inequality and an Improved Bound on the Degree-1 Fourier Weight for Linear Threshold Functions

The Khintchine inequality provides a lower bound on the expected absolute value of a weighted sum of independent Rademacher random variables. In the classical setting, when the coefficient vector has unit Euclidean norm, this lower bound is a constant, with equality attained only by a simple family of extremal configurations. A refinement due to De, Diakonikolas, and Servedio, referred to as the linear Khintchine inequality in this work, strengthens this estimate by establishing a lower bound that depends linearly on the distance of the coefficient vector from the extremal set. In this paper, we present a refined analysis of this dependence on the weight vector. Our results reveal a phase transition in the rate of improvement: when the dimension exceeds six, the lower bound undergoes an abrupt change as the weight vector deviates from the minimizer. Additionally, we improve the slope constant in linear Khintchine inequality. As a consequence, we establish an improved lower bound on the degree-1 Fourier weight for linear threshold functions $\mathbf W^{\leq1}[f]\geq0.55100929$, marking progress towards a conjecture of Benjamini, Kalai, and Schramm.
Read the full abstract

High Energy Physics (Phenomenology)

Deep-neural-network extraction of unpolarized transverse-momentum-dependent parton distributions in $b_T$ space from Drell-Yan data

We present a physics-informed deep-neural-network extraction of unpolarized transverse-momentum-dependent parton distribution functions (TMDPDFs) in impact-parameter space from Drell–Yan data. The perturbative contribution is computed with a resummed $W$ term using $\mathrm{N}^{3}\mathrm{LL}$ evolution, strict-NLO hard and operator-product-expansion matching, and smooth profile scales at small and large $b_T$. A compact feature-wise linear modulation network learns only a shared nonperturbative factor $F_{NP}(x,b_T)$; the collinear PDFs, hard factor, evolution kernel, matching coefficients, and Fourier–Bessel transform remain fixed. The primary result is a smooth light-flavor $b_T$-space TMD ensemble and its cross-section-level validation. The reported $k_T$ distributions are regularized finite-$b_T$ Hankel transforms, not independent momentum-space fits. As a separate robustness test, a smooth finite-$Y$ transition is applied to 24 additional Tevatron points extending to $q_T/Q\simeq0.30$. The nominal 329-point fit is unchanged, and the results remain stable when $F_{\rm NP}$ is held fixed while the transition profile is varied. An independent 122-bin Tevatron $\mathrm{N}^{3}\mathrm{LL}+\mathrm{NNLO}$ $W+Y$ grid provides a direct perturbative benchmark. A separate $W+Y$ candidate using the specified non-LHCb finite-$Y$ inputs is retained as an identifiability study.
Read the full abstract

Artificial Intelligence

Design the Feedback, Not Just the Reward: Environments for Open-Ended Reinforcement Learning

Reinforcement learning has advanced large language models on tasks with reliable verifiers, where checking an outcome provides a clear basis for reward. Extending this approach to open-ended tasks requires environments that can both evaluate complex, partially successful outputs and turn these judgments into effective learning guidance. We propose a reinforcement learning framework built around a training environment that integrates human domain knowledge with model reasoning. Human domain knowledge, encoded in rubrics, anchors evaluation to task intent, while model reasoning combined with hierarchical rewards distinguishes complex open-ended outputs and links diagnosed failures to likely responsible output regions. Furthermore, these diagnostic attributions guide token-level weights in the GRPO objective, thereby translating the ambiguous notion of success in open-ended tasks into concrete guidance for policy optimization. We instantiate this framework in interactive web application generation, a representative open-ended generation task. The resulting model achieves an average pass rate of 41.25% on MiniAppBench and 76.19% on ArtifactsBench. Ablation studies show the contributions of individual components, while additional out-of-distribution evaluations support transferable gains without degrading performance on other evaluated tasks.
Read the full abstract

All authors Rui Jin, Jikai Chen, Chengyue Yu, Yihan Chen, Hao Zhou, Demin Zhu, Kaichen Yang, Dong Wang, Linjian Mo, Chenyi Zhuang

Symplectic Geometry

Obstruction to Symplectic Embeddings between Toric Domains

We study symplectic embeddings between four-dimensional toric domains using embedded contact homology. Extending Hutchings' criterion for embeddings between convex toric domains, we obtain obstructions for embeddings from convex toric domains into concave toric domains, from semi-weakly convex toric domains into convex toric domains, and from concave toric domains into concave toric domains. The obstructions are formulated in terms of factorizations of convex, concave, and semi-weakly convex generators together with combinatorial constraints relating their ECH indices and actions. As applications, we recover the sharp obstruction for symplectic embeddings of a polydisk into a union of cylinders, and obtain sharp results for embeddings of certain quadrilateral toric domains into balls and ellipsoids. We also show that these obstructions are in some cases strictly stronger than those coming only from ECH capacities.
Read the full abstract

Geometric Topology

Instantons, indefinite 4-manifolds, and Dehn surgery

We prove that there exist hyperbolic integer homology spheres with arbitrarily large Dehn surgery number. Previously, no integer homology sphere was known to have a surgery number larger than $2$. Our approach uses Froyshov's invariant $q_3$ of integer homology spheres, which is defined in terms of mod 2 instanton homology. We show that if $W: Y \to Y'$ is a cobordism between integer homology spheres with no $2$-torsion in its first homology, then $-b^+(W) \le q_3(Y') - q_3(Y) \le b^-(W)$. We also extend both $q_3$ and the inequality to rational homology spheres.
Read the full abstract

Theoretical Economics

Strategic Disclosure and the Design of Attention

An institution that cannot compel disclosure can still allocate exposure. We study two opposed advocates who select verifiable observations and a receiver who averages the selected record without correcting for selection. The institution commits exposure before the state and evidence are known; each withheld item's assigned weight expires. With commonly known pools and strictly opposed monotone preferences, all finite-game Nash equilibria produce one action and common cutoff. In large pools with a vanishing anchor, directional tail moments determine the implementing exposure share. Under common evidence laws, equal exposure implements every state. The institutional restriction matters: full within-camp reallocation induces extreme selection and, on a fixed support, eliminates state responsiveness. In a finite-state common-law benchmark, small reallocation separates disclosure cutoffs and changes implementing shares according to favorable-tail frequencies. Variation in these frequencies across states creates a positive local welfare cost even after exposure is optimized. Ex ante results bound the cost of commitment and identify information sufficient for limiting implementation. Separately, under linear payoffs, uniform item weights and continuous evidence laws on a compact state space, all Bayesian equilibria converge to the baseline outcome when advocates observe only their own samples and learn the state. Finite-pool results characterize a precision cost under common laws, balanced exposure, uniform weights and equal limiting camp sizes, and first corrections to exact risk-minimizing shares under additional regularity.
Read the full abstract

Computation and Language

LandingAgent: A Reference-Annotated Dataset and Agentic Generation Framework for Landing Pages

Landing pages are goal-oriented web interfaces that must communicate a target-specific value proposition while organizing information flow, visual hierarchy, and calls to action (CTA). Although large language models can generate plausible webpage code from natural-language prompts, direct generation often yields generic templates and unsupported persuasive claims. We study target-grounded, reference-guided landing-page generation, where a system must create an executable page for a new target by adapting reusable patterns from real pages without copying them. We introduce LandingBench, a reference-profile dataset that abstracts real landing pages into section sequences, layout patterns, tone descriptors, visual emphasis, and CTA structure. Building on LandingBench, we propose LandingAgent, a three-phase agentic framework that profiles the target, constructs a reference-guided wireframe, and refines the page through critique-guided polishing. We evaluate LandingAgent against direct prompting on faithfulness, conciseness, readability, aesthetics, and structural diversity. Experiments show improved target grounding, presentation quality, and layout diversity. Code is available at https://github.com/IAURAI/LandingAgent.
Read the full abstract

All authors Injun Baek, HyeongSeok Lee, Yearim Kim, Junhoo Lee, Nojun Kwak

Quantum Physics

High-Throughput Normalized Min-Sum Belief Propagation Decoding for Quantum LDPC Codes with Near-Memory Processing

Real-time quantum error correction requires classical decoders to process growing syndrome workloads with low and predictable latency. For quantum low-density parity-check (qLDPC) codes, iterative belief propagation (BP) repeatedly updates messages over sparse Tanner graphs, creating substantial memory-access and data-movement demands. We map normalized Min-Sum BP decoding of the [[144,12,12]] Bivariate Bicycle qLDPC code onto a DPU-based Processing-in-Memory (PIM) architecture. Within each DPU, 11 tasklets cooperatively decode one syndrome, while multiple DPUs process independent syndrome instances in parallel. Using uPIMulator and a data-qubit Pauli error model with ideal syndrome measurements, we compare throughput, per-syndrome processing time, logical error rate (LER), and single-syndrome tail latency against a 16-logical-CPU baseline. At a component-wise physical error probability of p=0.001 and one BP iteration, the projected aggregate kernel throughput of 2,560 DPUs reaches 1.071 x 10^7 decodes/s, compared with 1.22 x 10^6 decodes/s for the CPU, an 8.8x improvement. From two iterations onward, the measured LER remains below the physical error probability for every evaluated value of p. For one to five iterations, the maximum sampled serialized X+Z DPU compute latency remains below the 1 ms decoder-side reference for trapped-ion QEC, reaching approximately 0.873 ms at five iterations. These results show that near-memory processing can provide high aggregate throughput and sub-millisecond compute latency for qLDPC BP decoding under the evaluated conditions.
Read the full abstract

All authors Jeonggeun Seo, Youngsun Han, Leanghok Hour, Dongmin Kim

General Relativity and Quantum Cosmology

Compact Core–Shell Equilibria in Gravitational Vlasov–Poisson Systems with Positive and Negative Mass

We construct regular, spherically symmetric, compact energy-cutoff equilibria for gravitational Vlasov–Poisson systems sourced by positive and negative mass distributions in the context of stellar dynamics. Motivated by Bondi's notion of negative mass and its relation to the weak field limit of Einstein gravity with a signed mass–energy source, we show that the internal structure of the steady states depends qualitatively on the type of negative mass present. In the Bondi convention, the signs of inertial, passive gravitational, and active gravitational mass are reversed together, preserving the equivalence principle. The resulting equilibria consist of a central overlap core containing both mass species, surrounded by a finite positive mass shell and an exterior vacuum. Thus, although a pure Bondi negative mass gas cannot form a compact steady state, Bondi negative mass can be spatially confined by a suitable positive mass distribution. In the gravitational-charge convention, only the passive and active gravitational masses change sign, so that the two species obey opposite free-fall laws and unlike masses repel. The resulting equilibria are spatially segregated into a negative mass core, a vacuum gap, a positive mass shell, and an exterior vacuum. For cutoff exponent $n=-1/2$, the radial matching is explicit, while for the regular cutoff $n=1/2$ the matter regions satisfy Lane–Emden-type equations. These results provide a kinetic framework for investigating the distribution of negative mass in astrophysical and cosmological settings.
Read the full abstract

Computation and Language

OpenStamp: A Watermark for Open-Source Language Models

With the growing prevalence of large language model (LLM) generated content, watermarking is considered a promising approach for attributing text to LLMs and distinguishing it from human-written content. A prominent class of techniques embeds subtle but detectable signals in generated text by modifying token sampling probabilities. However, such methods are unsuitable for open-source models, where users have white-box access and can easily disable watermarking during inference. In this work, we introduce OpenStamp, a watermarking technique that encodes the watermarking logic directly into the model weights by modifying only the final projection, or unembedding, layer. Through experiments across two models, we show that OpenStamp achieves superior detection performance, with minimal degradation in model capabilities compared to prior methods. The implanted watermark is explicitly designed, and empirically confirmed, to be more robust to paraphrasing attacks and harder to scrub off through post-hoc fine-tuning than prior open-source watermarks. To enable developers to watermark their models, we release our code alongside watermarked versions of 4 popular open-source models.
Read the full abstract

Quantum Gases

Microscopic Theory of Drag in a Bose Condensate Interacting with a Moving Reservoir

We derive a microscopic theory of drag for a Bose-Einstein condensate interacting with a moving reservoir. Starting from interactions between a condensate and a drifting fermionic bath, we integrate out the bath degrees of freedom within the Born–Markov approximation to obtain an effective Gross–Pitaevskii equation with a drag potential. From this potential we derive an effective drag force and obtain a closed expression for the drag coefficient, determined by the reservoir density fluctuations and the condensate density profile. As an application, we simulate the drag for an exciton–polariton condensate interacting with a drifting electron gas.
Read the full abstract

All authors Hassan Alnatah, Shuang Liang, Shouvik Mukherjee, Qi Yao, Ashton S. Bradley, David W. Snoke

Quantum Physics

From integrability to many-body quantum chaos through a Markovian bath

Recent examples have confirmed the common belief that quantum chaos is always suppressed in the presence of an environment. Here we show that this is not always the case. We compute the Lyapunov exponent of a $q$-body Majorana Sachdev-Ye-Kitaev (SYK) model coupled to a Markovian bath. Provided that the jump operators describing the bath are isotropic random $k > 2$-body Majoranas fields, the Lyapunov exponent is positive for any strength of the coupling to the bath. Interestingly, this also applies to $q=2$ where the SYK is integrable which indicates that many-body quantum chaos can be induced by the environment. Moreover, for $q > 2$, where the unitary dynamics is quantum chaotic, and sufficiently large $k$, the Lyapunov exponent increases with the coupling to the bath. Explicit analytical results are obtained in the $q=2$ and large $q$ limits. Our results put forward an alternate route to induce and control the generation of scrambling in quantum many-body systems which is of potential relevance in the design of quantum information devices.
Read the full abstract

Data Structures and Algorithms

Online Differentially Private Consistent Clustering

We study differentially private (DP) $k$-means and $k$-median clustering in the online streaming setting. In this model, points arrive sequentially, and at each time step, we need to output a set of $k$ centers that optimizes the clustering objective for all points seen so far. We give a generic reduction that transforms the (sensitive) input stream into a private stream, which is a semi-coreset of the input stream. This implies that any (non-private) online clustering algorithm, run as a post-processing step, can achieve good utility for the original clustering objective. Our algorithm matches or improves upon the approximation ratio, space usage, and running time of existing algorithms [Epasto et al., 2026, Dupré la Tour et al., 2024]. A key aspect of our reduction is that it inherits desirable properties of the underlying non-private clustering algorithm, such as consistency [Lattanzi and Vassilvitskii, 2017]–a property not satisfied by previous DP algorithms.
Read the full abstract

All authors Edith Cohen, Vadym Doroshenko, Badih Ghazi, Pritish Kamath, Alexander Knop, Ravi Kumar, Ethan Leeman, Pasin Manurangsi, Adam Sealfon, Marika Swanberg

Algebraic Geometry

Degree of irrationality of properly elliptic surfaces

In this paper, we study the degree of irrationality of properly elliptic surfaces with a section. We prove $\min\{\chi(\mathcal O_S),\,2\operatorname{gon}(C)\} \leq \operatorname{irr}(S) \leq 2\operatorname{gon}(C)$. The lower bound is obtained from the canonical bundle formula and the Cayley–Bacharach property. We show that this bound is sharp. We also study the behavior of the degree of irrationality in moduli. A very general properly elliptic surface with a section over a curve of genus at least two has degree of irrationality at least four, whereas special families with $\chi(\mathcal O_S)=1$ or $2$ have degree two. Finally, we investigate properly elliptic surfaces with $\chi(\mathcal O_S)=0$, proving a generic lower bound of four and showing that $\operatorname{irr}(C\times E)=4$ for every hyperelliptic curve $C$ of genus at least two and every elliptic curve $E$. Our paper also includes special properly elliptic surfaces without a section. For Dolgachev surfaces, we exclude degree two for a very general member and construct special examples of degrees two and three.
Read the full abstract

High Energy Physics (Experiment)

Observation of the $\Xi_c^0 \to pK^-$ decay and measurement of its decay asymmetry

A search for the Cabibbo-suppressed decay $\Xi_c^0 \to pK^-$ is performed using $pp$ collision data corresponding to an integrated luminosity of $5.4\,\mathrm{fb}^{-1}$, collected by the LHCb experiment at a centre-of-mass energy of $13\,\mathrm{TeV}$. The decay is observed for the first time and its branching fraction measured to be $(4.5\pm0.5\pm0.2\pm0.9)\times10^{-5}$, where the uncertainties are statistical, systematic and from the branching fraction of the normalisation channel $\Xi_b^- \to \Xi_c^0 (\to p K^- K^- \pi^+) \pi^-$. Using the decay chain $\Xi_b^- \to \Xi_c^0(\to pK^-)\pi^-$, the decay asymmetry parameter of the $\Xi_c^0 \to pK^-$ decay is determined to be $\alpha_{\Xi_c^0}=0.32\pm0.15\pm0.01$.
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, 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, C. Buti, J.S. Butter, J. Buytaert, 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, A. Casais Vidal, 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. 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, 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, O.D. Durmus, 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, 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, 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, M. Karacson, 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. 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, 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, 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, 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, F. Oliva, E. Olivart Pino, M. Olocco, R.H. O'Neil, J.S. Ordonez Soto, D. Osthues, J.M. Otalora Goicochea, P. Owen, 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, M. Piccini, L. Piccolo, 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, 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, C. Trippl, 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, 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.H. Wang, Z. Wang, Z. Wang, J.A. Ward, A. Wasili, 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, 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, T. Zies, D. Zuliani, X. Zuo

Computer Vision

CommerceVibe: Learning to Design E-Commerce Creatives as Executable Visual Code via Dual-Feedback Reinforcement Learning

High-quality e-commerce creatives are essential for presenting products and conveying marketing messages. Recent diffusion models enable scalable creative generation and produce visually compelling images, but their flattened raster outputs often contain distorted text and inconsistent product details, requiring refinement before deployment. Moreover, without explicit structure, the resulting creatives are difficult to edit and reuse, while complex design requirements remain challenging to encode as verifiable training signals. To address these challenges, we present CommerceVibe, which represents creatives as executable visual code and formulates generation as conditional HTML/CSS program synthesis. Given product images, design requirements, and product information, it produces renderable, editable, and reusable creatives. We further introduce dual-feedback reinforcement learning, in which rule-based feedback evaluates rendered programs for text readability, product visibility, and layout validity, while visual feedback from a vision-language model (VLM) assesses rendered creatives against input specifications across six perceptual and commercial dimensions. Together, these complementary feedback signals improve both constraint satisfaction and perception-dependent quality. We perform supervised fine-tuning (SFT) of Qwen3.5-9B on over 28,000 e-commerce examples, followed by dual-feedback reinforcement learning. On a 1,300-case benchmark, the optimized CommerceVibe model achieves a weighted score of 94.0/100, compared with 87.3 for the SFT-only variant, and outperforms strong external models. Blind evaluations by five e-commerce design experts further validate these improvements. CommerceVibe supports controllable, editable, and scalable e-commerce creative production.
Read the full abstract

All authors Yajiao Xu, Jin Zhang, Jiangbo Ai, Tao Jiang, Mo Xu, Lina Huang, Chengfu Huo

Earth and Planetary Astrophysics

Modeling Carbon-Based Planets with MAGRATHEA

A planet's interior structure not only influences its habitability, but it also provides information about the planet's formation and history. In the quest to characterize the variety of planetary discoveries, the most common practice is to rely on our understanding of the Earth's differentiated interior structure, with an iron core and silicate mantle as a starting point. However, recent work by arXiv:2408.07761 [astro-ph.EP] shows that these assumptions may not be true for all planetary systems as the composition of the condensed material can change drastically towards a carbon mantle instead of a silicate one, once the C/O ratio of the system approaches and exceeds approximately 0.9. In an effort to more fully understand the range of possible planetary bodies, we added the ability to model carbon-based mantles to our open-source planetary interior solver, MAGRATHEA. The newly added carbon mantle includes three relevant allotropes of carbon: graphite, diamond, and BC8 (body centered cubic with eight atoms per cell).
Read the full abstract

Systems and Control

AutoDRI: Bridging the Semantic Gap for Automated Design Rule Integration in CP-SAT-Based Cell Synthesis under Multi-Patterning

Design-rule integration (DRI) remains a major bottleneck for scalable (Constraint Programming with SAT) CP-SAT-based standard cell synthesis and rapid technology enablement at advanced nodes. It still depends heavily on manual effort and domain expertise. Moreover, existing low-level rule encodings are not expressive enough for emerging constraints such as cut-based rules under multi-patterning technology. This paper presents AutoDRI, a multi-agent framework for automated design-rule integration in standard cell synthesis. AutoDRI combines a geometric semantic library, a standardized conflict-set encoding, a constructive multicolor-cut modeling method, and a feedback-driven multi-agent flow to bridge the semantic gap between natural-language design rules and executable CP-SAT constraints. In the reported experiments, AutoDRI achieves near-perfect rule-integration correctness across 41 cell benchmarks under 10+ complex rules, including colored cut-mask spacing rules, reaching 33/33 correct integrations with Gemini-3-pro and 32/33 with GPT-5.4, while maintaining runtime comparable to manual hard-coding and passing KLayout DRC and Cadence LVS.
Read the full abstract

All authors Yuhao Ren, Yucheng Wang, Zihao Chen, Chung-Kuan Cheng, Zhiang Wang

Materials Science

Ionization Energies, Electron Affinities, Bandgaps, Exciton Binding Energies, and Polarization Energies of Orientation-Controlled Picene, [6]-Phenacene, and [7]-Phenacene Thin Films

Phenacenes, in which benzene rings are fused in a zigzag manner, are chemically robust $\pi$-conjugated hydrocarbons of interest for organic electronic applications and superconductivity. However, quantitative electronic parameters of phenacene thin films remain limited compared with those of acenes. Here, we determine the ionization energies (IEs), electron affinities (EAs), band gaps, exciton binding energies, and polarization energies of orientation-controlled thin films of picene, [6]-phenacene, and [7]-phenacene using ultraviolet photoelectron spectroscopy and low-energy inverse photoelectron spectroscopy. The energy parameters depend only weakly on molecular size. In contrast, both IE and EA exhibit large orientation-dependent shifts of approximately 1 eV: lying films show IEs of 6.41-6.43 eV and EAs of 2.16-2.42 eV, whereas standing films show IEs of 5.44-5.67 eV and EAs of 1.36-1.53 eV. Nevertheless, the band gaps of approximately 4 eV and exciton binding energies of approximately 1 eV remain nearly independent of molecular size and orientation. Analysis of polarization energies shows that the induction term is nearly orientation independent, whereas the electrostatic term strongly depends on orientation and originates primarily from molecular quadrupole moments. These results identify molecular orientation as a key factor governing frontier energy levels in phenacene thin films.
Read the full abstract

All authors Rintaro Makino, Mihiro Kubo, Keiichirou Yonezawa, Hiroyuki Yoshida, Satoshi Kera

Software Engineering

Decoupling is a Necessity: Transformation-Agnostic Decompiled Code Recovery under Optimization and Obfuscation

Reverse engineering is essential for software security analysis and vulnerability detection. Decompilation, the process of lifting binaries to high-level pseudocode, is central to this task. However, production binaries are hostile environments: aggressive compiler optimizations and adversarial obfuscation jointly mangle control structures, obscure variable intents, and disguise high-level program logic. Consequently, existing LLM-based decompilation tools frequently suffer from structural collapse and semantic hallucinations. We present ReSource, the first multi-phase LLM framework designed for transformation-agnostic source recovery. To tackle these intertwined distortions, ReSource conceptualizes the binary-to-source discrepancies into three orthogonal tiers, namely lexical, syntactic, and semantic, and decouples the recovery process accordingly. First, to ground the LLM and prevent logic drift, it retrieves empirical priors from a curated Semantic Distortion Database. Second, to resolve control-flow flattening, it integrates a lightweight predictor to reconstruct the source-level structural skeleton. Finally, a contextual lexical deduction stage refines identifiers to restore human readability. Evaluated on a massive benchmark of over 80,000 decompiled-source function pairs across three optimization levels and four obfuscation techniques, ReSource achieves an 83% Top-5 source retrieval accuracy and an average similarity score of 0.66. By maintaining robust semantic identifiability where state-of-the-art baselines (DeGPT, LLM4Decompile, and FidelityGPT) severely overfit or degrade, ReSource provides a scalable and reliable foundation for downstream security analysis.
Read the full abstract

All authors Zhiping Zhou, Xiaohong Li, Ruitao Feng, Yao Zhang, Yuekang Li, Wenbu Feng

Computer Vision

Thread-Efficient Decoding for Neural Texture Compression

Neural texture compression (NTC) achieves higher compression ratios than BCn formats but suffers from GPU thread divergence, which significantly reduces runtime performance. In this work, we propose a shared decoder MLP architecture – trained with a gradual decoder freezing schedule – combined with texture clustering to reduce thread divergence by 25%-52% while preserving rendering quality. We evaluate our method on over 500 textures and multiple real rendering scenes, demonstrating up to 8.48x speedup on the Radeon RX 9070 XT GPU compared to non-shared baselines. Our key contributions include: (1) a unified shared decoder architecture that reduces divergence by grouping textures; (2) a training recipe with gradual decoder freezing that improves stability and reconstruction accuracy; (3) a semantic clustering strategy using CLIP embeddings that groups similar textures for effective decoder sharing; and (4) comprehensive performance and ablation studies validating our approach.
Read the full abstract

All authors Janarbek Matai, Sho Ikeda, Lukasz Lipski, Takahiro Harada

Combinatorics

Note on adjacency spectra for twice punctured hypercubes

Eigenspaces of hypercube graph adjacency matrices are well understood. We extend this to subgraphs obtained by removing a vertex or antipodal pair of vertices, motivated by a question on zonotopes.

Artificial Intelligence

Resource Constraints and Performance in Agentic AI Systems

Progress toward more autonomous AI increasingly depends on agentic systems that combine a language model with tools, memory, state management, and multi-step execution. These mechanisms shape both task capability and operational burden. We compare OpenClaw and NanoBot as complete agentic systems using a paired primary benchmark and a more detailed instrumented subset of paired prompts. In the primary benchmark, the rate of full task completion was 31% for OpenClaw and 25% for NanoBot, a six-percentage-point difference with a 95% task-bootstrap interval from -3 to 15 percentage points, providing no statistically established full-completion advantage for either system. In the instrumented layer, both systems achieved 26% full completion, while NanoBot reached at least partial completion on 43% of prompts compared with 26% for OpenClaw. OpenClaw took longer on 83% of prompts and had a higher recorded peak-memory value on every prompt, with geometric mean ratios of 2.98 for wall time and 19.44 for peak memory. Among the ten detailed-layer prompts on which at least one system achieved partial or full completion, NanoBot weakly dominated on eight; across all 23 prompts, however, ten of its eighteen dominance cases were cheaper joint failures. Outcome labels differ across the two evidence layers, showing why agent-system evaluation should connect capability and resource measurements to attempt-level execution and scoring provenance. These findings show that progress toward more autonomous AI should be evaluated through verified task completion, observed resource use and records linking each result to the execution that produced it.
Read the full abstract

Machine Learning

There and Back Again: Bidirectional Diffusion Bridges for Multimodality Translation

Multimodality translation (e.g., text-to-image) is a core generative AI task. However, existing approaches (1) follow generative paths that do not directly represent the source modality, limiting the flexibility of some sampling algorithms; and (2) are unidirectional, preventing inversion (e.g., image-to-text). We propose BIT: Bidirectional Image-Text Diffusion Bridges. In contrast to previous approaches, BIT starts directly from text and interpolates into images, providing (1) a source-aware generative path that enables diverse and flexible sampling algorithms; and (2) an endpoint-conditioned process that can be traversed from image to text, providing a unified, bidirectional generative framework. BIT is derived through stochastic calculus, yielding SDE forms amenable to simulation and tractable loss functions that scale to high dimensions. Our experiments show that BIT is competitive with denoising-diffusion and deterministic-flow baselines, and outperforms them on several vision–language and natural-science evaluations.
Read the full abstract

All authors Gabe Guo, Elon Litman, Thanawat Sornwanee, Jose Blanchet, Stefano Ermon

Statistics Theory

Exploiting Exact Conditionals Improves Conditioning: Provably Fast Mixing Time Bounds By Sampling from the Marginal

The problem of sampling from a probability distribution arises in many applications such as posterior sampling in hierarchical Bayesian inverse problems and Gaussian processes for machine learning. Markov chain Monte Carlo (MCMC) algorithms are often used for sampling from a target probability distribution, but implementations can be computationally expensive, especially for large-scale problems. In certain applications, the target distribution naturally factorizes into a lower dimensional marginal distribution and a conditional distribution that allows exact sampling. We describe an MCMC algorithm called MarCo that exploits such a structure and generates a Markov chain via Metropolis-Hastings sampling from the marginal distribution, followed by sampling from the exact conditional distribution. By design, MarCo constructs a Markov chain on the joint space that inherits the convergence behavior of the marginal MCMC algorithm. This provides multiple theoretical and computational advantages. We prove that MarCo can achieve improved mixing time upper bounds compared to direct sampling from the joint distribution. Moreover, compared to one-block methods that also exploit marginal-conditional structure, we use the framework of Peskun-Tierney ordering to show that MarCo has a larger right spectral gap and smaller asymptotic variance, thus leading to superior convergence properties. Numerical results illustrate the performance benefits of MarCo and are provided for various problems, including a semi-blind image deblurring example.
Read the full abstract

All authors Abhijit Chowdhary, Federica Milinanni, Julianne Chung, Elizabeth Newman

Machine Learning

Beyond Pairwise Graphs in Science: Hypergraph Adaptive Wavelet Operators for Parametric PDEs

Physical systems are often modeled by solution operators that map input fields, parameters, geometries, or past states to steady or future physical states. Learning these maps is difficult, especially for time-dependent systems that must assimilate history and remain stable under autoregressive rollout. Many neural operators work best on regular, structured grids, while realistic simulations often require unstructured meshes or point clouds to resolve complex geometries; in such settings, grid-centric representations can lose accuracy. Graph neural operators handle these domains through message passing or spectral graph filtering, but pairwise edges do not directly capture group-wise couplings among mesh cells, local neighborhoods, or conservation volumes. We introduce the Hypergraph Adaptive waveLet Operator (HALO), which lifts the domain to a hypergraph and learns in its spectral wavelet domain. HALO avoids explicit hypergraph-Laplacian eigendecomposition through Chebyshev polynomial wavelet filters, giving localized spectral kernels at linear sparse-matrix cost. Its trainable dyadic wavelet scales are regularized toward tight-frame coverage, allowing the frequency response to adapt to each PDE while encouraging stable multi-scale spectral coverage. Across 2D and 3D benchmarks on structured and unstructured discretizations, HALO achieves best or near-best accuracy among frequency-, transformer-, DeepONet-, state-space-, and graph-based baselines and sustains stable multi-step rollouts. The same model scales to industrial aerodynamic geometries: on meshes of a few hundred thousand points it is on par with, or better than, the strongest fixed-discretization transformers, while remaining resolution-equivariant.
Read the full abstract

Machine Learning

SOMTab: Set-Order Mamba for Efficient Tabular In-Context Learning

Tabular foundation models based on in-context learning have recently emerged as strong alternatives to task-specific model fitting. However, the current performance frontier remains dominated by attention-heavy architectures, where attention is used throughout the modeling pipeline. This raises a natural question: is attention necessary at every stage of tabular in-context learning? We introduce SOMTab, a Set-Order Mamba architecture for efficient tabular in-context learning. SOMTab separates representation construction from query-conditioned retrieval. For row and column representations, it maps unordered table tokens into stable latent slots and applies Mamba-based state-space mixing to construct compact representations. For final prediction, it retains attention-based in-context learning to preserve query-conditioned retrieval from labeled context examples. We further introduce DCH-TailMix, a synthetic prior that combines degree-corrected graph heterogeneity with mixed heavy-tailed regimes to diversify synthetic dependency structures. Across tabular benchmarks, SOMTab approaches the performance of strong Transformer-based tabular foundation models while achieving faster inference and lower GPU memory usage, yielding a favorable efficiency–accuracy trade-off.
Read the full abstract

Computer Vision

StreamEMS: Streaming Video Understanding with Self-Evolving Memory Scheme for Vision-Language Models

Recently, many streaming video understanding methods have been proposed by constructing an external memory to store historical data for computational reduction. Most methods focus on optimizing the injection procedure of current data (write) and retrieving informative historical data (read) from memory, while overlooking the opportunity to further enhancing the representational capability of memory itself. In this work, we present StreamEMS, a general mechanism for improving streaming video understanding by re-structuring the historical data stored in memory through self-evolving memory scheme, enabling more informative and robust memory representations. Specifically, we first introduce a Semantic Evolution Module to evolve the memory into more information-dense representations by exploiting informative memory entities discovered via progressively shrinking semantic scales from coarse to fine. In addition, we further introduce a Prior-informed Evolution Module to evolve memory into more robust representations by leveraging prior memory distributions to refine the current memory state. We validate the effectiveness of our proposed designs on widely-used streaming video understanding datasets, i.e., OVO-Bench and StreamingBench, and the results showcase that our method performs better than other methods. Moreover, the advantage of our method becomes consistently evident even under high token usage drop rate settings, indicating the effectiveness and robustness of our method in unleashing the potential of the memory itself.
Read the full abstract

Probability

Precise universal edge asymptotics for planar $\beta=2$ Coulomb gases with radial external fields

We investigate the extremal statistics of planar $\beta=2$ Coulomb gases with radial external fields. For the rightmost eigenvalue and the spectral radius, we establish sharp Berry–Esseen bounds for their convergence to the Gumbel distribution, with explicit rates \[ \frac{25\log\log n}{4e\log n} \quad\text{and}\quad \frac{2\log\log n}{e\log n}, \] respectively. In addition, we derive sharp asymptotic equivalences for the large and moderate deviations of both statistics across all relevant scales. Analogous results hold for the smallest modulus.
Read the full abstract

Computer Vision

What Do Interaction Representations Actually Measure? Pre-Event Separability in Weakly-Supervised Violence Detection

Articulated human pose provides detailed body-configuration information beyond coarse spatial relationships, but whether this detail yields greater discriminative information when the downstream pipeline is held fixed remains unclear. We examine this through early violence detection. Holding the tracker, temporal head, supervision, folds, and evaluation fixed, we compare five interaction representations spanning coarse bounding-box geometry, a matched handcrafted pose analogue, enriched pose descriptors, and a matched-capacity encoder learned from raw joints, under video-level evaluation with cluster-bootstrap intervals. No pose-based representation outperforms coarse geometry, though with fifteen anomalous videos this subset cannot rule out small effects. Extending the pipeline to frozen visual encoders, and repeating the comparison on XD-Violence (137 anomalous videos, nine times our UCF-Crime sample), person-crop appearance and whole-frame context both exceed geometry by a wide margin, yet context matches appearance on UCF-Crime and exceeds it on the larger split: cropping to the interacting people yields no advantage over encoding the whole frame. This prompts a direct test of what the benchmark measures. Scoring anomalous videos using only frames preceding the annotated onset, under a control removing sequence length as a cue, retains 39-91% of above-chance separation on both benchmarks, including for seven hand-designed geometric channels. Inspection of the tightest pre-onset windows identifies concrete provenance artifacts: editorial title cards and platform watermarks absent from the surveillance footage supplying the normal class. Video-level AUC here is thus a composite of event evidence and pre-event source cues, a shared source of discrimination that can obscure differences between representations. The diagnostic requires only annotations these benchmarks already ship.
Read the full abstract