Materials Science
Symmetry Engineering Enables Deterministic Ensemble Polarization Control in Hexagonal Boron Nitride
Aqiq Ishraq, Eric Herrmann, Pragya Agnihotri et al. (+15 more)
Optical polarization is a key degree of freedom in
quantum photonic and sensing technologies, enabling efficient light–matter coupling and directional emission. However, in defect ensemble-based
quantum systems, orientational averaging across many defects suppresses optical anisotropy and eliminates a deterministic polarization axis. Here,
symmetry engineering is introduced as a strategy to restore collective optical anisotropy in negatively charged boron vacancy (V_B^-) ensembles hosted in hexagonal boron nitride (h-BN). By imposing anisotropic in-plane tensile strain through lithographically defined nano-ridge arrays, in-plane
symmetry is broken and the ensemble emission dipole is aligned with the ridge-defined strain
symmetry axis. Polarization-resolved photoluminescence reveals that the polarization visibility scales with the magnitude of strain anisotropy, while the emission orientation rigidly follows the engineered
symmetry axis across devices with varying strain direction. This deterministic
alignment is independent of
crystal orientation, establishing the optical foundation for polarization-defined ensemble
spin readout and programmable photonic integration in two-dimensional
quantum materials
Read the full abstract ▾
All authors Aqiq Ishraq, Eric Herrmann, Pragya Agnihotri, Alexander Hutchinson, Ramiro M. dos Santos, Shahidul Asif, Lottie Murray, Abhijith Puthiya Veettil, Luke Stockl, Muhammad Hassan Shaikh, Collin Maurtua, Kenji Watanabe, Takashi Taniguchi, Matthew Doty, Cyrus E. Dreyer, Anderson Janotti, Xi Wang, Chitraleema Chakraborty
Machine Learning
ZeroMAG: Zero-Shot Multimodal Adapter Generation for Plug-and-Play EEG Foundation Models
Yubo Wang, Jingying Ma, Xinliang Zhou et al. (+8 more)
EEG foundation models (EFMs) capture reusable knowledge from large-scale
EEG data, while many
EEG recordings also include companion physiological signals that provide complementary information beyond the
EEG-only interface. The challenge is to preserve this pretrained knowledge while extending the EFM to heterogeneous
multimodal recordings through an adaptation inferred from unlabeled target data. We introduce ZeroMAG, a zero-shot
multimodal adapter generation framework that extends a frozen
EEG encoder and prediction head using unlabeled target recordings, without target labels or target-side optimization. The target
datasets are held out from all model training and selection in the ZeroMAG pipeline. ZeroMAG organizes companion modalities around a configuration-invariant adapter, constructs a modality-subject-task condition from unlabeled recordings and task context, and generates adapter weights in a function-constrained latent space learned from source adapters. Across six held-out target
datasets and three EFM backbones, ZeroMAG improves balanced accuracy by 7.22 percentage points over
EEG-only inference and 4.89 points over direct weight regression, while coming within 0.50 points of supervised
multimodal adaptation on average. Ablations further show that removing functional supervision from either representation learning or conditional generation degrades generated-adapter performance, confirming the contribution of both components.
Read the full abstract ▾
All authors Yubo Wang, Jingying Ma, Xinliang Zhou, Yangxuan Zhou, Jiquan Wang, Sha Zhao, Yiyuan Yang, Yi Ding, Ziyu Jia, Chenyu Liu, Cuntai Guan
Quantum Physics
What Must a Quantum-Memory Decoder Know About Temporally Correlated Noise?
Danesh Morales-Hashemi, Graeme Smith and Ningping Cao
What must a
quantum error-correction decoder know about temporally correlated noise, and what does it cost to learn? We study this question for fixed stabilizer
memory experiments with known system-environment interactions. We formulate calibration in terms of the fidelity differences that select a Pauli recovery and relate missing information to the resulting fidelity loss. Our main example uses weak coherent $X$ rotations in a class of CSS codes including Shor, Steane, and odd-distance rotated surface codes. All physical
qubits share a sign either fixed throughout a run or independently redrawn each interval. The two models have identical syndrome statistics and the same spatiotemporal Pauli process (SPP), obtained by Pauli-twirling every noise interval. Yet at suitable storage times, they require different logical corrections. In the weak-noise limit, each model is almost perfectly correctable when known. Without model information, even the best final recovery has a worst-case fidelity loss approaching $1/2$ relative to the noise-aware optimum, despite access to the full syndrome record. For this pair of models, at fixed code and fixed positive target loss below $1/2$, we prove matching calibration bounds. With logical preparation and final readout, complete QEC cycles require a number of noise intervals scaling as the inverse rotation angle raised to the code's $X$ distance. Allowing two uninterrupted intervals before one extraction reduces the optimal cost to inverse-square scaling. Physical Pauli twirling of every noise interval changes both recovery performance and its information requirements. Discarding the record of applied random Pauli operations makes the SPP sufficient for optimal recovery. Keeping this record can improve the attainable fidelity, but reaching that optimum may require additional noise information.
Read the full abstract ▾
Cosmology and Nongalactic Astrophysics
Baryonic Imprints on DM Halos: the concentration-mass relation and its dependence on 28 model parameters in the CAMELS suite
Cami Delgado and Dhayaa Anbajagane
The
evolution of
baryons in
halos, as well as their gravitational impact on the total-matter phase space, has a rich phenomenology and is vital in answering many questions in both cosmology and astrophysics. We use the CAMELS simulation suite to quantify the impact of twenty-three different
galaxy formation parameters/processes (as well as five cosmological parameters) on the concentration–mass relation, $c_{\rm vir}$ - $M_{\rm vir}$, which characterizes the radial density distribution in
halos. We construct a simulation-informed nonlinear model for concentration as a function of
halo mass,
redshift, and twenty-eight cosmological/astrophysical parameters. This is done using 2048 simulations of the IllustrisTNG
galaxy formation model from the CAMELS suite. We extract the imprints of
galaxy formation across a wide range in mass $M_{\rm vir} \in [10^{11}, 10^{14.5}]$ $M_{\odot}/h$ and in
redshift $z \in [0, 3]$, finding clear mass- and
redshift-dependent features in each of the different feedback mechanisms. From our measurements, we construct a three-component phenomenological model that can include astrophysical imprints in the concentration–mass relation in a fast and simple manner.
Read the full abstract ▾
Computer Vision
DuoMatching: Joint-Marginal Distribution Matching for Few-Step Video Generation
Jiahao Zhan, Yan Wang, Yongrui Ma et al. (+5 more)
Streaming
video generation has benefited from distribution matching
distillation (DMD), which matches the joint distribution of
video frames to a
video teacher's approximation of the real
video distribution. Although this joint matching mitigates drift during autoregressive rollouts, limitations remain in visual quality and semantic
alignment. To address these limitations, we propose DuoMatching, a distribution matching framework that approximates the real
video distribution through a unified joint-marginal formulation. On top of existing joint matching formulations, the additional marginal matching objective provides dedicated frame-level supervision from an image generator, transferring complementary visual and semantic priors from it. To apply this frame-level supervision in
video generation, we introduce LatentBridge to resolve the latent representation mismatch between the
video student and the image teacher. Latent Variation Sampling further distributes such frame-level supervision across distinct temporal segments, reducing redundancy. Experiments demonstrate that DuoMatching improves visual quality, composition, and semantic
alignment while largely preserving motion dynamics. Human evaluations show overall preference rates above 80% against all evaluated baselines. The project page is available at https://johnzhan2023.github.io/DuoMatching/.
Read the full abstract ▾
All authors Jiahao Zhan, Yan Wang, Yongrui Ma, Qunliang Xing, Ruchang Yao, Runtao Liu, Shijie Zhao, Tianfan Xue
Algebraic Geometry
Simpson's closedness conjecture in arbitrary rank
Tianzhi Hu
For a compact Riemann surface $X$, Simpson associated to each stable graded
Higgs bundle $(E,\theta)$ a locus $W_{[(E,\theta)]}^1 \subset M_{\mathrm{dR}}(X,n)$, consisting of flat bundles that admit a Simpson filtration whose associated graded
Higgs bundle is $(E,\theta)$. He conjectured that $W_{[(E,\theta)]}^1$ is Zariski closed in $M_{\mathrm{dR}}(X,n)$. We prove this
conjecture in arbitrary rank. The proof is based on the Quillen geometry of the determinant-of-
cohomology line bundle over the
moduli of holomorphic bundles. To any holomorphic family of flat bundles, we associate a determinant-of-
cohomology line bundle equipped with a canonical holomorphic determinant connection, and derive explicit formulas for its
curvature. For a family of flat bundles in $W_{[(E,\theta)]}^1$, we then prove that the determinant connection form is exact. This exactness forces the associated determinant frame to extend as a nowhere-vanishing frame across any one-parameter degeneration, which controls the limiting filtration and yields the desired closedness.
Read the full abstract ▾
Probability
An infinite-dimensional Skorokhod problem with a diagonalization approach
Louis T. Clarke, Guodong Pang and Ruoyu Wu
We study an infinite-dimensional Skorokhod problem on the positive cone of $L_1([0,1])$, where a regulator process constrains a free càdlàg path to remain nonnegative through an oblique reflection mechanism determined by a positive linear integral operator $\boldsymbol{F}$ on $L_1([0,1])$ with spectral radius less than $1$. This constrained process $Z$ remains in the positive cone, and the regulator $Y$ is nondecreasing with respect to the cone order and satisfies a pointwise complementarity condition, increasing only on those space-time regions where the corresponding component of $Z$ vanishes. This formulation extends the classical multidimensional Skorokhod problem on the nonnegative orthant to an infinite-dimensional setting. Under suitable assumptions on the operator $\boldsymbol{F}$ that guarantee a dominant positive eigenfunction, we proceed through a diagonalization argument to give an alternative representation of the system in which the $L_1$ norm of the linear operator is less than 1. From this characterization, we establish existence and uniqueness of solutions for arbitrary input paths $X\in D([0,T],L_1([0,1]))$. We further show that the associated Skorokhod map is well defined and Lipschitz continuous with respect to both the uniform and Skorokhod $J_1$
topologies. As a consequence, solutions may be obtained as limits of multidimensional Skorokhod problems, providing a constructive characterization of the infinite-dimensional reflection mechanism. These results provide a framework for the study of constrained
stochastic systems with infinitely many interacting components and provide a foundation for reflected
stochastic processes evolving in function spaces.
Read the full abstract ▾
Mathematical Physics
Virasoro Conformal Blocks and modular functor from Liouville CFT
Guillaume Baverez, Colin Guillarmou, Antti Kupiainen et al. (+2 more)
In this article, we give a global construction of the Virasoro conformal blocks on Teichmüller space of Riemann surfaces of genus $g$ with $m$ marked points, when the central charge is $c>25$. We prove that they are global holomorphic sections of a holomorphic line bundle and they satisfy the Ward identity, which encodes their conformal invariance. For the sphere with $3$ points, the space of conformal blocks is one-dimensional. For a given marked Riemann surface and a marked pair of pants decomposition, it is in general an infinite dimensional
Hilbert space, isomorphic to $L^2(\R_+^{3g-3+m})$ with respect to some measure involving the DOZZ structure constants. We show that it carries a projective
unitary representation of the mapping class group. More generally, there are
unitary isomorphisms for each Moore-Seiberg move, changing a marked pair of pants decomposition into another one.
Read the full abstract ▾
All authors Guillaume Baverez, Colin Guillarmou, Antti Kupiainen, Rémi Rhodes, Yuxiao Xie
Mathematical Physics
Conformal Bootstrap for surfaces with boundary in Liouville CFT. Part II: spectral resolution and bootstrap
Colin Guillarmou, Rémi Rhodes and Baojun Wu
This paper is the second part of the proof of the conformal bootstrap for Liouville conformal field theory on compact surfaces with boundary. It is devoted to the spectral theory of the half-annulus
semigroup and to the resulting bootstrap formula. Building on the Segal axioms and gluing properties established in Part~1~\cite{GRW1}, we identify the generator of the half-annulus
semigroup with the boundary
Hamiltonian and establish its spectral decomposition by scattering methods. This yields a direct-integral decomposition of the boundary state space into irreducible Virasoro representations, governed by the boundary spectral measure. Combining this decomposition with Ward identities for boundary deformations, we express general correlation functions as integrals over spectral parameters attached to cutting curves, with integrands given by products of bulk and boundary structure constants and the corresponding conformal blocks. These results also underpin the analyticity and
symmetry properties of conformal blocks~\cite{GRSSbloc}, and contribute to the construction of mapping class group representations on spaces of conformal blocks~\cite{PapierBlocs}.
Read the full abstract ▾
Robotics
RATE: Risk-Aware Tactile Encoding for Contact-rich Robotic Manipulation
Yuyao Jiang, Haichao Liu, Jiarui Zheng et al. (+3 more)
Tactile sensing is particularly valuable for contact-rich robotic manipulation. Recent work has made substantial progress in tactile representation learning for robotic manipulation. However, similar tactile observations can arise from interaction conditions with very different task-risk implications, such as sensor noise, task-necessary variations, and emerging undesirable contact. Without context-grounded risk information, these cases can be ambiguous to downstream policies, leading to unnecessary corrections to benign variations or delayed responses to genuinely risky contact. To address this limitation, we propose Risk-Aware Tactile Encoding (RATE), which learns tactile representations that encode task-conditioned interaction risk. Specifically, history-conditioned prediction captures interaction context, while alert supervision associates this context with task-conditioned risk. The learned risk-aware representation complements conventional tactile features through a lightweight residual adapter. Experiments in both simulation and the real world demonstrate substantial improvements in task success, with controlled ablations confirming the complementary benefits of alert-guided learning and predictive temporal modeling.
Read the full abstract ▾
All authors Yuyao Jiang, Haichao Liu, Jiarui Zheng, Zihan Ding, Weihao Yuan, Ziwei Wang
Robotics
Autonomous Robotic Navigation for Endovascular Brain-Computer Interface Access
Harry Robertshaw, Weijie Qi, Nikola Fischer et al. (+3 more)
Endovascular
brain-computer interfaces (BCIs) avoid craniotomy but require precise device delivery through anatomically variable cerebral veins. This work presents the first demonstration of in vitro autonomous robotic
navigation for endovascular BCI access in the cerebral venous system. Soft Actor-Critic controllers were trained in silico for two sequential tasks spanning the right internal jugular vein to the superior sagittal sinus, using
geometric augmentation of one training anatomy.
Navigation was evaluated in a training anatomy and an anatomically unseen hold-out model over 250 in silico episodes and five fluoroscopy-guided in vitro robotic runs per task-anatomy condition, comprising 1,000 simulated episodes and 20 physical runs overall. Task recurrent predictors were also evaluated for online identification of impending
navigation failure. In silico success rates for Tasks A and B were 85.6% and 98.4% in the training anatomy and 42.0% and 91.6% in the hold-out anatomy, respectively. Fourteen of 20 physical runs were successful (70% overall), including 80% success for Task B in the hold-out phantom. In silico the predictors detected 99.3-100.0% of failures with false-alarm rates of 0.8-6.7%. During in vitro evaluation, predicted risk increased before failed episodes, but elevated probabilities during some successful runs showed reduced calibration after transfer. These results demonstrate the feasibility of autonomous cerebral venous access and show how online failure prediction could support human oversight, while also identifying anatomical generalization and sim-to-real calibration as priorities before preclinical translation.
Read the full abstract ▾
All authors Harry Robertshaw, Weijie Qi, Nikola Fischer, Alejandro Granados, Thomas C. Booth, Sam E. John
Biological Physics
Migrating cell clusters as active droplets with asymmetric shape and contact angles
Hanno I. Hennighausen, Jaime A. Espina, Elias H. Barriga et al. (+1 more)
During embryonic development and cancer progression, clusters of
cells migrate collectively following external cues. Such migrating
cell clusters have been modeled as active droplets wetting a substrate. So far, this active wetting model assumed the cluster to take the symmetric shape of a spherical cap. Here, in contrast, we experimentally show that migrating clusters of frog neural crest
cells have asymmetric height profiles, with a lower contact angle at the front than at the rear. We then extend the active wetting model by deriving an equation for the cluster shape, which becomes part of the solution rather than an assumption. Our model shows that shape asymmetry can arise from the asymmetric tractions that drive cluster migration: They generate a pressure gradient across the tissue which, via the Young-Laplace equation, leads to a
curvature gradient on the cluster's upper surface. Finally, by fitting the theoretical predictions to the measured height profile, we estimate the tissue viscosity. Overall, our results provide a general mechanism for the emergence of asymmetric shapes in migrating
cell clusters, understood as active droplets on a wettability gradient.
Read the full abstract ▾
All authors Hanno I. Hennighausen, Jaime A. Espina, Elias H. Barriga, Ricard Alert
Differential Geometry
On the Definition and Smooth Examples of Weakly Kähler–Finsler Metrics
Xiaoshu Ge
The generalization of the Kähler metric in the complex Finsler setting admits four natural versions. Among these, the strongly Kähler–Finsler condition is known to coincide with one of the intermediate conditions. In this paper, we establish the equivalence between the weakly Kähler–Finsler condition and the remaining intermediate condition. Moreover, we obtain new necessary and sufficient conditions for a weakly Kähler–Finsler metric to be strongly Kähler–Finsler. Using these criteria, we construct, for the first time, smooth examples of weakly Kähler–Finsler metrics which are not Kähler–Finsler, via the Kobayashi metrics on smooth bounded strongly convex domains.
Read the full abstract ▾
Numerical Analysis
Learning the Willmore flow with a lightweight neural operator
Elie Bretin, Roland Denis, Tokuhiro Eto et al. (+1 more)
We extend to the Willmore flow the phase-field neural operator approach proposed by Bretin, Denis, Masnou, and Terii (2022) for mean
curvature motion in two and three dimensions. The resulting lightweight, single-time-step architecture is motivated by a Bence-Merriman-Osher-type expansion of the Allen-Cahn
semigroup, and is trained using trajectories generated by a hybrid minimizing movement scheme. In our experiments with non-oriented interfaces, unconstrained convolution kernels produce anisotropic dynamics, whereas radial kernels enhance the
evolution process. We also construct three-dimensional kernels from trained two-dimensional kernels via their radial
Fourier profiles, which provides an effective initialization for the non-oriented 3D model. Finally, we show applications to curve and surface reconstruction from unoriented point clouds, including reconstruction using the transferred model without the need for additional 3D training.
Read the full abstract ▾
All authors Elie Bretin, Roland Denis, Tokuhiro Eto, Simon Masnou
Differential Geometry
Multiplicity of Finsler geodesics and electromagnetic trajectories via localized deformations
Erasmo Caponio, Anna Valeria Germinario and Antonio Masiello
Given a Finsler
manifold $(M,F)$ and suitable nested regions $B_1\subset B_2\subset B_3$, with $M\setminus B_1$ connected and non-contractible, as in the model case of concentric balls in $\mathbb R^n$, $n\geq2$, we consider conformal deformations $\varphi_\lambda F$ which can be chosen so as to leave $F$ unchanged outside $B_3$ and grow on $B_2\setminus B_1$. This growth creates an energy barrier, diverging with $\lambda$, that low-energy curves cannot cross. The minimax families can therefore be chosen in the path space of $M\setminus B_1$, which contains compact subsets of arbitrarily large category. Consequently, for every $m\in\mathbb N$ and all sufficiently large $\lambda$, there exist at least $m$ geodesics joining two prescribed points, all avoiding $B_1$. This extends to the Finsler setting an idea of Capozzi, Fortunato and Greco. The same scheme applies to Randers metrics $\alpha+\beta$, with $\alpha$ deformed and $\beta$ unchanged, as well as to fixed-energy trajectories of non-relativistic electromagnetic systems and relativistic Lorentz-force trajectories, and light rays in spacetimes endowed with a causal Killing field. In the non-relativistic case only the scalar potential is perturbed, while the
magnetic field is preserved. When the Killing field is not everywhere timelike, the Fermat principle yields a possibly singular Randers–Kropina metric on a spacelike hypersurface. Under a natural admissible category condition, combining localized barriers with Randers approximation gives $m$ Randers–Kropina geodesics with pairwise distinct lengths, equivalently $m$ future-pointing light rays with pairwise distinct arrival times.
Read the full abstract ▾
Mesoscale and Nanoscale Physics
Electrostatic Doping of Moiré Superlattices Controls the Optical Fingerprint of a WSe_2 /Twisted Bilayer Graphene heterostructure
Houssem Eddine Hannachi and Sihem Jaziri
We theoretically investigate the optical response of the WSe2 monolayer vertically stacked on twisted
bilayer graphene (tBG) under electrostatic doping. In this
heterostructure, the doped moiré superlattice of tBG generates a spatially modulated electrostatic potential that couples to the electron and hole constituents of the WSe2
exciton through the screened Coulomb interaction, thereby providing a microscopic mechanism for
exciton coupling to the charge-density landscape of the moiré
Fermi sea. Focusing on twist angles below the magic angle (1.1{\deg}), where electronic correlations are strongly enhanced, we demonstrate that electrostatic doping induces pronounced modifications of the excitonic absorption spectrum, including the emergence of multiple satellite resonances. We show that these spectral features originate from the interplay between the
exciton center-of-mass and internal degrees of freedom, together with doping-induced hybridization between Rydberg
exciton states of different orbital
symmetry. The calculated spectra are in good agreement with previously reported experimental observations and theoretical predictions. Our results establish electrostatic doping in tBG as an effective means of engineering excitonic states in adjacent semiconducting monolayers and provide a pathway toward novel optical sensor devices.
Read the full abstract ▾
Computation and Language
Author Representation Strategies for Zero-Shot Authorship Attribution: A Comparative Study of LLM-Based and Embedding-Based Approaches
Nudrat Habib, Tosin Adewumi, Sana Sabah Al-Azzawi et al. (+2 more)
Authorship Attribution (AA) requires capturing fine-grained stylistic characteristics, making it particularly challenging in zero-shot (ZS) settings where no task-specific supervision is available. In this work, we investigate the effect of author representations on ZS AA by evaluating a label-only prompting baseline together with three author representation strategies: representative writing samples, LLM-generated descriptions, and style
embeddings (LISA). The first three approaches perform attribution using LLM prompting, while the
embedding-based approach uses style
embeddings with cosine similarity. We investigate the influence of
prompt design and propose a two-stage
embedding-based attribution framework that combines candidate space reduction with
embedding-dimension selection. The results show that label-only ZS AA is ineffective, while incorporating author-specific representations consistently improves attribution performance. Among the evaluated approaches, the proposed two-stage LISA framework achieves the strongest overall performance, whereas LLM-generated style descriptions provide a substantially more compact representation of author style at the cost of some attribution performance. These findings demonstrate the importance of author representation in ZS AA, while indicating that current open-source LLMs remain insufficient for robust attribution without more effective representation learning.
Read the full abstract ▾
All authors Nudrat Habib, Tosin Adewumi, Sana Sabah Al-Azzawi, Marcus Liwicki, Elisa Barney
Robotics
DR-IPC: Disturbance-Resilient Integrated Planning and Control for LiDAR-Based Quadrotor Navigation
Peng Liu, Jingyan Wang, Qipeng Ye et al. (+5 more)
LiDAR-based quadrotor
navigation in cluttered environments remains challenging under external disturbances, particularly when obstacle-aware motion generation and disturbance-rejection control are handled in separate layers. This article presents disturbance-resilient integrated planning and control (DR-IPC), which combines lightweight path guidance with nonlinear model predictive control (NMPC) to directly generate angular velocity and thrust. An interconnected extended Kalman filter and nonlinear disturbance observer jointly provide filtered state estimates and reconstructed disturbances for NMPC prediction. The resulting formulation unifies nonlinear quadrotor dynamics, actuator constraints, local motion generation, and penalised safe-flight-corridor residuals without requiring a separate trajectory-optimization stage. Gazebo and MARSIM simulations, together with indoor and outdoor experiments, validate DR-IPC under wind, suspended payloads, narrow passages, ball impacts and reactive avoidance of a dynamic obstacle. In multi-goal
navigation with disturbances, DR-IPC increases the number of completed missions from 1/10 to 9/10 in Gazebo and reduces the altitude RMSE from 0.34 to 0.01 m in experiments. The complete system operates onboard at 100 Hz. Supplementary
videos are available on the project page https://drpp316.github.io/DR-IPC-Page/, and the source code will be released.
Read the full abstract ▾
All authors Peng Liu, Jingyan Wang, Qipeng Ye, Wen Li, Jinya Su, Zuo Wang, Shihua Li, Yunda Yan
Machine Learning
Nicolas Zucchet and Scott W. Linderman
Distillation has become a core primitive of large
language model training, but its properties are not yet well understood. We take an entropic perspective, studying how the
entropy of the student depends on the data and the divergence that define the
distillation objective. We prove that forward KL inflates the
entropy of the student above that of the teacher. Since cross-
entropy training is a special case, this yields an identity that we verify quantitatively in pretraining and supervised finetuning. Other divergences come with no such guarantee: reverse KL deflates
entropy until the gap between student and teacher gets too large, and interpolating between the two changes
entropy smoothly early in training but abruptly at convergence. The lower
entropy of on-policy
distillation comes from
token-level reverse KL, not from on-policy sampling. The divergence therefore acts as an implicit
entropy regularizer, whose role is clearest in self-
distillation: as conditioning on privileged information deflates
entropy, the divergence hyperparameters that work best are those that compensate for it.
Read the full abstract ▾
Quantum Physics
Breaking the chain: geometry-native state preparation with ASPIRE
Fredrik Hasselgren and Matthew L. Sims-Goh
Loading data into
quantum computers is a key bottleneck that threatens end-to-end
quantum advantage.
Quantum-
circuit compilers for structured states like
matrix product states (MPS) are a critical enabler to manage this in
quantum chemistry, finance, and many-body
quantum simulation. Approximate preparation of MPS typically relies on layered sequential staircases of nearest-neighbour gates. We present Adaptive State Preparation by Iterative Removal of
Entanglement (ASPIRE), an
algorithm that approximately compiles state-preparation
circuits by selecting gates based on the state's underlying
entanglement geometry and the available connectivity of the target
quantum processor. Our protocol scores candidate
qubit pairs on the removability of their
entanglement and identifies layers of parallelised long-ranged gates that remove the most
entanglement. We demonstrate our protocol empirically across a range of settings, illustrating its success for preparing
Hamiltonian ground states and multivariate amplitude-encoded functions, and accounting for
qubit connectivity, hardware noise, and both NISQ and early fault-tolerant settings. Throughout, ASPIRE typically achieves higher fidelities and shallower
circuits than existing methods for states with long-ranged
entanglement, even when classical optimisation is accounted for. Our investigation positions ASPIRE as a competitive, resource-frugal state-preparation protocol, with particular promise for applications tolerant to imperfect fidelity such as preparation of guide states for
quantum phase estimation.
Read the full abstract ▾
Cosmology and Nongalactic Astrophysics
Probing Early Phases of the Epoch of Reionization with the $\mathrm{kSZ}^{2}\times21\,\mathrm{cm}^{2}$ Cross-Correlation
Wen-Qing Guo, Yin-Zhe Ma, Adam Lidz et al. (+2 more)
Observations of the Epoch of Reionization (EoR) provide key insights into the nature of the first luminous sources and the thermal state of the intergalactic medium (IGM). Cross-correlations between the kinetic Sunyaev–Zel'dovich (kSZ) effect and neutral hydrogen (HI) 21-cm emission offer a promising probe of this era, sensitive to both the ionization and thermal histories of the early Universe. We investigate the cross-correlation of the squared kSZ field with the squared 21-cm intensity-mapping field ($\mathrm{kSZ}^{2}\times21\mathrm{cm}^{2}$) over $z\simeq9$–$16$, focusing on the impact of spatial fluctuations in the HI
spin temperature. This higher-order statistic avoids line-of-sight cancellations in the kSZ signal, while squaring the 21-cm field enables non-vanishing correlations with pairs of equal and opposite high-$k_\parallel$ modes accessible outside the foreground wedge. We forecast the cumulative signal-to-noise ratio ($\mathrm{S/N}$) for several combinations of 21-cm and CMB experiments, including realistic instrumental noise. For a fiducial heating history and an optimistic foreground-avoidance scenario ($k_\parallel\leq0.1\,h\,\mathrm{Mpc}^{-1}$), we find $\mathrm{S/N}\simeq5.89$ for SKA-Low AA$^\ast\times$CMB-HD and $\simeq3.62$ for HERA$\times$CMB-HD. If the foreground wedge extends to the horizon, these decrease to $2.94$ and $2.85$, respectively. Our results demonstrate that $\mathrm{kSZ}^{2}\times21\mathrm{cm}^{2}$ is sensitive to the thermal
evolution of the IGM as well as its ionization structure, providing complementary constraints on the high-
redshift thermal and reionization histories.
Read the full abstract ▾
All authors Wen-Qing Guo, Yin-Zhe Ma, Adam Lidz, Steven Murray, Hemanth Potluri
Machine Learning
Beyond Trained Models: Compiling GNNs for a Sound Explainer Benchmark
Steve Azzolin, Francesco Paolo Nerini, Stefano Teso et al. (+4 more)
Explainers for Graph Neural Networks (GNNs) are commonly evaluated by their plausibility, i.e., how well their explanations recover a predefined ground truth, such as a motif planted in the data. This protocol implicitly assumes that a GNN trained on such data relies on the intended motif. Although prior work has questioned this assumption, plausibility remains widespread. First, we show that the assumption is violated on several widely used
benchmarks, where, e.g., degree statistics alone suffice to solve the task. Then, we remove this confounder by replacing training with compilation. We achieve this by introducing $\mathsf{Gracr}$, the first compiler translating graded modal logic formulas into GNN weights, yielding models that replicate the behaviour of the corresponding formulas. Since the behaviour of the model is now known by construction, we can define its ground truth explanation formally and compute it exactly. Building on this, we introduce $\mathsf{Gracr}\mathsf{Bench}$, a
benchmark of compiled GNNs for the evaluation of explainers against this exact ground truth. Experiments on eleven explainers across six tasks show its effectiveness for fine-grained diagnostic evaluation: notably, we discover that most explainers are not robust to indirect influences or alternative implementations of the same formula. These results position $\mathsf{Gracr}\mathsf{Bench}$ as a novel, rigorous evaluation setting for graph post-hoc explainability.
Read the full abstract ▾
All authors Steve Azzolin, Francesco Paolo Nerini, Stefano Teso, Francesco Bonchi, Bruno Lepri, André Panisson, Andrea Passerini
Computation and Language
Structured Composition of Verifiable Atomic Insights for Table-to-Report Generation
Teng Lin, Xinyu Liu and Nan Tang
Table-to-report generation refers to the task of automatically generating article-level analyt- ical reports from relational tables and is an essential capability for automated data science and decision support. Its central challenge lies in systematically discovering verifiable com- posite insights across tables, attributes, and analytical perspectives, and organizing them into coherent, complete, and traceable evidence chains. Existing methods primarily rely on sequential, reactive data
agents or direct Large
Language Model(LLM) generation. They suffer from exploration bias: early local observations constrain subsequent actions, causing models to focus prematurely on local analyzes and miss cross-table or cross-dimensional evidence. We propose ComInsight, which reformulates insight discovery as the composition of atomic evidences. We first define an atomic insight as the smallest executable analytical unit conforming to a predefined analysis pattern and enumerate all valid atomic insights from database schema and content. These atoms are then organized into a multi-relational insight graph, where nodes represent verified data facts and edges encode logical, temporal, or hierarchical relations. Finally, a set of composition operators systematically fuses atomic nodes into higher-order composite conclusions. Every composite output is accompanied by executable SQL and fine-grained provenance, ensuring full verifiability. Across three
benchmarks InsightBench, DDR-Bench, and T2R-Bench, ComInsight consistently outperforms strong baselines in factual correctness, novelty, and structural completeness. We believe ComInsight offers a reliable, efficient, and explainable path toward table-to-report generation.
Read the full abstract ▾
Artificial Intelligence
From Benchmarks to Production: A Text-to-SQL System for Complex Financial Data
Arijit Sehanobish, Bruno Gomes Coelho, Guillaume Michel et al. (+3 more)
General-purpose Text-to-SQL systems achieve strong performance on academic
benchmarks like Spider and BIRD, where schemas are relatively shallow and column values are often human readable. In production financial databases, where concepts are stored as opaque integer keys rather than human-readable strings, these methods fall below 50%, as even simple queries require multiple joins and filter predicates reference opaque IDs. We present Financial LINking Text-to-SQL (FLINT), a domain-specialized Text-to-SQL system that closes this gap through three key components: (1) a lookup
agent that dynamically resolves natural-
language concepts to question-specific reference table constraints, (2)
embedding-based
retrieval of structurally similar query templates from a compact, expert-authored bank, and (3) schema linking that prunes a large table schema to the relevant subset by traversing foreign-key chains, rather than relying on name similarity alone. We evaluate on two
datasets totaling 359 questions over production financial schemas. FLINT outperforms various state-of-the-art baselines using the same LLM. The system is deployed in production as part of a financial data
retrieval service.
Read the full abstract ▾
All authors Arijit Sehanobish, Bruno Gomes Coelho, Guillaume Michel, Sophia Zhi, Valerie Faucon-Morin, Kristen Howell
Signal Processing
FedSwitch: Federated Region Classification From Wireless Channel Measurements
Mattia Piana, Stefano Rini and Stefano Tomasin
Wireless Internet-of-Things (IoT) networks can leverage locally observed channel state information (CSI) at base-stations (BSs) to perform region classification, i.e., infer the region of origin of a transmitter (TX), thereby enabling channel-based authentication. However, centralizing these measurements incurs substantial communication overhead and requires sharing location-dependent
radio fingerprints. To address these difficulties, we propose an edge-native
federated learning (FL) framework where BSs collaboratively train a shared classifier via a parameter server (PS) without exposing raw data, thereby supporting physical-layer security tasks like spoofing detection. To overcome the statistical heterogeneity arising from diverse BS locations and propagation conditions, we introduce FedSwitch, a personalized FL
algorithm that automatically transitions edge nodes from collaborative training to local fine-tuning upon loss stagnation. We provide theoretical convergence guarantees for the proposed collaborative training phase and derive analytical bounds linking classification errors to statistical discrepancies. Extensive evaluations on synthetic channel measurements demonstrate that FedSwitch significantly improves region-classification accuracy over standard FL baselines.
Read the full abstract ▾
Computer Vision
Feedforward Novel View Synthesis for Heterogeneous Cameras
Meng Wei, Cheng Zhang, Boying Li et al. (+4 more)
Feed-forward novel view synthesis has recently shown promising results from sparse posed images, but most existing methods assume that context and target views share a fixed camera family. This homogeneous-camera assumption breaks in practical multi-sensor systems, where perspective, fisheye, and panoramic cameras may coexist and where the target projection may be unseen during training. We study feed-forward NVS across heterogeneous central cameras and identify a key ambiguity introduced by tokenization: a visual
token aggregates a projection-dependent bundle of pixel rays, while existing camera encodings mainly expose absolute rays or
token-center relations. To address this, we combine
token-center relative Camera Positional Encodings and proposed local raymaps, a
token-level representation that explicitly describes the intra-patch ray distribution summarized by each
token. We further propose projection-aware 2D RoPE, which replaces raw image-grid coordinates with ray-induced angular coordinates so that relative positional
reasoning is aligned across camera projections. Together, these components treat diverse cameras as calibrated samplings of a shared ray space rather than separate visual domains. On ScanNet++ with heterogeneous-camera system, our method improves over camera-conditioned baselines under mixed-camera evaluation and demonstrates zero-shot generalization to panoramic views.
Read the full abstract ▾
All authors Meng Wei, Cheng Zhang, Boying Li, Yihang Chen, Jianmin Zheng, Hamid Rezatofighi, Jianfei Cai
Optics
Quantitative spin-orbit engineering enables subwavelength-scale programmable polarization states for versatile photonic functionalities
Shuaijie Yuan, Xuefeng Zhang, Jin Yang et al. (+7 more)
Precise and programmable control of optical polarization provides a versatile vectorial degree of freedom for light-matter interaction and optical information encoding. Existing polarization state generators (PSGs), however, suffer from poor switchability and modest fidelity at the micrometer scale, which severely constrains their practical applicability. To overcome these limitations, here we present a quantitative
spin-orbit engineering that enables the dynamic synthesis of arbitrary polarization states at subwavelength dimensions. The core concept is to tailor conjugate optical
vortices that independently address the orthogonal
spin components of a vectorially polarized light. This spatial modulation scheme annihilates intrinsic topological charges and reshapes the
spin-orbit interaction, thereby coherently constructing polarization ellipses on demand. Leveraging this mechanism, we experimentally synthesize diffraction-limited arbitrarily polarized beams spanning the entire Poincare sphere with fidelity over 99%. As proof-of-principle demonstrations, we apply the miniaturized PSG platform to Mueller-
matrix polarization microscopy, pixel-level polarization encryption, and parallel polarization-information encoding, illustrating its broad utility of programmable polarization control for subwavelength polarimetric imaging and reconfigurable multiplexed polarization encoding. Compared with state-of-the-art PSGs, our approach uniquely integrates rapid programmability, high fidelity, and subwavelength localization in a single architecture. These multifunctional capabilities of this work position the platform as a promising route for advancing Mueller
matrix polarization microscopy, multidimensional optical storage, and secure optical information processing, and beyond.
Read the full abstract ▾
All authors Shuaijie Yuan, Xuefeng Zhang, Jin Yang, Min Cheng, Xinyu Ma, Jinhai Zou, Jinyong Leng, Zhongquan Nie, Bing Lei, Pu Zhou
Algebraic Geometry
Yunfeng Jiang and Hsian-Hua Tseng
Let $X$ be the
moduli space of stable trace-free rank $r$
Higgs bundles over a smooth projective curve $C$ with fixed determinant $\Lambda\in\Pic(C)$ and $\gcd(r,\deg\Lambda)=1$. We prove that the $\cc^*$-equivariant
quantum product of a divisor on $X$ is given by the Steinberg correspondence of the Lagrangian Steinberg cycle in $X\times X$. The Steinberg cycle is the image of the reduced virtual fundamental cycle of the $2$-pointed Kontsevich stable map space to $X$ under evaluation map. For $\Gamma=\Pic^0(C)[r]$, we give a character decomposition for $\Gamma$-invariant curve classes or finite orbit sums. In rank two and genus two we explicitly determine its fifteen endoscopic lines and the reduced spectral discriminant. %The resulting endoscopic support statement is cohomological, not a determination of the entire Chow cycle. For $\widehat X=[X/\Gamma]$, which is the
moduli space of stable $\PGL_r$-
Higgs bundles over $C$. We consider the orbifold
quantum cohomology theory from a gerbe-twisted theory. The complex K-theory $\KU^*(X)$ and twisted K-theory $\KU^*([X/\Gamma],\alpha)$ by the lifting $\mu_r$-gerbe $\alpha$ of the universal projective bundle is an isomorphism, and is given by the
Fourier-Mukai transformation. We prove that the
Fourier-Mukai transformation is compatible with the
quantum product by the divisors.
Read the full abstract ▾
Artificial Intelligence
Reasoning Models Are Accurate but Unsound on Identification
Arman Behnam and Binghui Wang
A
reasoning model asked whether a causal effect is recoverable from observational data can fail in two ways: it refuses an identifiable query or answers a nonidentifiable one. The latter is more consequential, as no observational data can validate the claimed formula. Measuring this failure requires queries that are provably non-identifiable, which prior evaluations lack, and grading that accepts correct formulas in any equivalent form, which string matching cannot provide. We build CERTID, a formal identification pipeline that addresses both limitations. CERTID uses the sound and complete causal identification
algorithm ID to certify whether an effect is identifiable from a given graph and query, and verifies returned formulas against structural causal models whose interventional distributions are known exactly. CERTID further develops theoretical results to mitigate structural leakage, repair non-identifiable queries, and establish grading guarantees. We evaluate three frontier
reasoning models (Gemini Flash, Gemini Pro, and GPT5.5) on 1,200 certified instances spanning 4 to 50 vertices. Accuracy proves a poor proxy for soundness: on identical instances, the false-claim rate on non-identifiable queries varies by seventeen-fold across models. We also find that models decide identifiability with 97-100% accuracy on graphs generated after the strongest model's training snapshot. Instances, the certification procedure, the verifier, and per-instance records are available at https://anonymous.4open.science/r/certid-D718.
Read the full abstract ▾
Cryptography and Security
PrivDev: Mapping Static-Analysis Data Types to DPV
Simon Bernbeck, Ricardo Ramalho, Matheus Amendoeira et al. (+1 more)
Static-analysis scanners can identify personal-data types in source code, but they lack mechanisms to connect these findings to standardized
privacy vocabularies. PrivDev maps 122 Bearer CLI data types to Data
Privacy Vocabulary Personal Data (DPV-PD) categories and links them to potentially relevant GDPR provisions. Our approach combines deterministic mapping for 43 exact-label matches with a
retrieval-grounded Large
Language Model (LLM) to resolve the remaining 79 non-trivial mappings. The resulting RDF knowledge graph contains 118 ODRL policy resources that were structurally validated using SHACL. The artifact passed five complementary validation gates that cover structural correctness, query consistency,
retrieval quality, LLM-based assessment, and human evaluation. In human evaluation, nine annotators produced 711 judgments, yielding a raw agreement of 0.72, Gwet's AC1 of 0.68, and Gwet's AC2 of 0.88. Our results indicate that the proposed mappings are plausible and reproducible, while also revealing ambiguities in scanner-defined data-type labels and coverage gaps in DPV-PD.
Read the full abstract ▾
All authors Simon Bernbeck, Ricardo Ramalho, Matheus Amendoeira, Juliana Alves Pereira