Machine Learning
Growing and Transferring Pathways for LLM Reinforcement Learning
Haolin Tian, Yikang Li and Tonghan Wang
Reinforcement learning (RL) can substantially improve the
reasoning capabilities of large
language models, but an effective training run requires considerable resources. Reusing the computation acquired through RL across models would extend the value of that training. We propose GrowRL, an RL method that organizes learned updates into compact computational pathways through structural growth. GrowRL freezes the pretrained backbone and adds an independent causal
attention branch and a feed-forward network branch at each
Transformer layer. Zero-initialized output projections preserve the pretrained policy at initialization, and RL updates only the added pathways. The learned pathways remain explicit after training, allowing transfer to structurally compatible target models and subsequent adaptation with the target backbone frozen. The architecture also supports sequential transfer of additional pathways. By keeping the learned
attention and feed-forward computation separate from the pretrained weights, GrowRL provides an explicit structure for preserving and reusing the outcome of an RL run. Experiments across five backbones show competitive mathematical
reasoning and out-of-distribution performance with a small number of trainable parameters. Learned pathways provide immediate gains across all three transfer directions, with performance after only 10 target-side updates already close to that after 100 updates in every case. These results support computational pathways as parameter-efficient and reusable units of reinforcement learning adaptation.
Read the full abstract ▾
Mesoscale and Nanoscale Physics
Emergence of Nonlocal Hydrodynamics in Many-Body Radiative Heat Transport
Mauricio Gomez Viloria, Florian Herz, Riccardo Messina et al. (+1 more)
We develop a nonlocal hydrodynamic theory of radiative heat transport in many-body systems by coarse graining their microscopic electrodynamic interactions. Long-range radiative exchange naturally generates a spatially nonlocal constitutive relation in which the heat current at a given position depends on temperature gradients over an extended region, while energy conservation retains its local conservative form.
Fourier transport emerges as a controlled long-wavelength limit when the interaction kernel has a finite second spatial moment. When this moment fails to converge, spatial nonlocality survives coarse graining and gives rise to non-
Fourier transport. The resulting framework establishes a systematic bridge between discrete many-body fluctuational electrodynamics and a continuous description of radiative heat transport.
Read the full abstract ▾
All authors Mauricio Gomez Viloria, Florian Herz, Riccardo Messina, Philippe Ben-Abdallah
Differential Geometry
The triangular metric on projective surfaces
León Carvajales, Xian Dai, Beatrice Pozzetti et al. (+1 more)
In \cite{ThurstonStretch} Thurston introduced stretch paths and stretch maps between
hyperbolic surfaces. In this paper we extend similar constructions to the space of convex real projective structures on surfaces. Using Fock-Goncharov-like parameters, we introduce stretch paths and construct explicit stretch maps between the projective surface. We show that stretch maps are $e^t$–Lipschitz with respect to a triangular metric we introduce on the projective surface. This allows us to gain control on several length functions on an open set of the space of convex real projective structures (those where there is no “local back–tracking"). As an application we obtain pointwise upper bounds for the
Hilbert entropy of some convex real projective structures in terms of Fock-Goncharov-like parameters.
Read the full abstract ▾
All authors León Carvajales, Xian Dai, Beatrice Pozzetti, Anna Wienhard
Probability
Poincaré inequalities, logarithmic growth and doubling
Joaquim Martin
We investigate when a Poincaré inequality forces the underlying measure to be doubling, with particular
attention to the space \(X\) controlling the oscillation. We identify \(L\log L\) as the sharp threshold: under a natural local growth condition on the measure, doubling is forced precisely when \(L\log L(0,1)\not\hookrightarrow X\). We also obtain a doubling criterion under a weaker
geometric assumption when the oscillation norm is sufficiently stronger than \(L\log L\). We show that this threshold is sharp in every fixed Euclidean dimension, using a dimension-free local \(L\log L\)-Poincaré inequality for log-concave measures and non-doubling Gaussian examples. We then turn to infinite-dimensional spaces, where doubling is impossible for the measures under consideration. Nevertheless, the \(L\log L\)-Poincaré inequality continues to hold, and we establish sharp characterizations for infinite product models, Gaussian measures on
Hilbert spaces under power-type small-ball estimates, and Wiener measure with the uniform norm.
Read the full abstract ▾
Analysis of PDEs
A geometric lower bound for the clamped plate eigenvalue problem in low dimensions and extensions
Francesca Bianchi, Francesco Bozzola and Lorenzo Brasco
We show that the spectrum of the Dirichlet-biLaplacian can be quantitatively bounded from below in terms of the inradius only, for a general open set in dimensions $2$ and $3$. For the same dimensions, we also extend this result to fractional powers $s$ of the
Laplacian, with $N/2<s<2$. The resulting estimate comes with a constant having the correct asymptotic behavior as $s$ approaches the two endpoints $N/2$ and $2$. The proofs exploit capacitary techniques in the vein of Maz'ya.
Read the full abstract ▾
Machine Learning
SLEM: Spectrum-Aware Latent Expert Merging
Dohyeon Kim, Bedionita Soro, Bruno Andreis et al. (+1 more)
Mixture-of-Experts (MoE) models scale model capacity efficiently by activating only a small subset of experts for each input, while still requiring all parameters to be stored in
memory, resulting in substantial
memory overhead. Expert merging techniques have recently emerged as a promising compression paradigm that reduces this overhead by consolidating multiple experts into a smaller set. However, existing approaches have two key limitations: (i) direct convex interpolation in parameter space is not generally guaranteed to yield meaningful interpolation of expert functions; and (ii) estimating expert similarity or merging coefficients often relies on calibration data, introducing dependence on the calibration distribution. To address these limitations, we propose Spectrum-Aware Latent Expert Merging (SLEM), a calibration-free expert merging framework that occurs in a structured latent representation of expert weights using a VAE. SLEM groups experts within and across layers based on latent similarities and derives merging coefficients from expert weight spectra. It then performs geometry-aware latent interpolation and decodes the resulting representations into merged expert weights. Compared with existing expert merging methods, SLEM achieves comparable or better performance on general knowledge
benchmarks and higher average performance on generation
benchmarks involving mathematical
reasoning and code generation.
Read the full abstract ▾
All authors Dohyeon Kim, Bedionita Soro, Bruno Andreis, Sung Ju Hwang
Superconductivity
Molecule-intercalated van der Waals Josephson junctions uncover complex multiband superconductivity
Daniel Margineda, David Caldevilla-Asenjo, Covadonga Álvarez-García et al. (+6 more)
Intercalation of molecular species provides a route to tailor the electronic structure in van der Waals
superconductors, yet intercalated compounds have rarely been integrated into phase-sensitive
quantum devices. Here, we realize hybrid molecular Josephson junctions based on TaS2/NbSe2
heterostructures. Selective intercalation of the TaS2 electrode induces monolayer-like Ising superconductivity, while the junctions exhibit asymmetric Fraunhofer interference, field-induced and spontaneous Josephson diode effects. The zero-field diode disappears upon reversing the electrode stacking and re-emerges after intercalating the lower NbSe2 electrode. These findings indicate that the hybrid superconducting electrode hosts unconventional superconductivity underlying the nonreciprocal transport, while molecular intercalation acts as a tuning knob for its internal phase structure. This behavior is consistent with a time-reversal-
symmetry-breaking multiband superconducting state characterized by a nontrivial relative phase between superconducting condensates. Our results establish hybrid molecular Josephson junctions as a
quantum-device platform for phase-sensitive investigation and integration of molecularly engineered
superconductors, opening opportunities to uncover and engineer unconventional superconducting phases.
Read the full abstract ▾
All authors Daniel Margineda, David Caldevilla-Asenjo, Covadonga Álvarez-García, Daniel Tezze, Raul Arenal, Maxim Ilyn, Celia Rogero, Luis E. Hueso, Marco Gobbi
Computer Vision
MiMaAlign: Hierarchical Micro-Macro Alignment for Physically Plausible Video Generation
Yansong Liu, Litong Gong, Weijie Li et al. (+2 more)
Text-to-
video (T2V)
diffusion models have made remarkable progress in generating high-fidelity visuals, yet their outputs still often exhibit physically implausible phenomena, including discontinuous dynamics and implausible event interactions. Existing representation
alignment methods mainly transfer generic
video features or short-range motion cues, treating physical plausibility as a single-level representation problem. To address this issue, we propose MiMaAlign, a hierarchical micro-macro
alignment framework that transfers complementary dynamics-oriented and event-oriented knowledge into different abstraction depths of a pretrained
video diffusion model. Specifically, micro-level continuous dynamics
alignment imposes constraints by aligning
diffusion features with a continuous-time motion representation learned by a Neural Ordinary Differential Equation (Neural ODE), while macro-level event-commonsense
alignment provides representations of object interactions and event semantics through an instruction-prompted
vision-
language model (VLM). With training-only auxiliary objectives, we align the complementary supervision from the two teachers with shallower and deeper generator features, respectively, enabling the generator to internalize the corresponding physical priors. Empirical evaluations show that, on CogVideoX-2B, MiMaAlign achieves a relative improvement of up to 21.0% in the Physical Commonsense (PC) score on the VideoPhy
benchmark over the baseline, and also delivers substantial gains on the highly challenging VideoPhy2
benchmark, while maintaining general
video quality on VBench and being supported by blind user comparisons.
Read the full abstract ▾
All authors Yansong Liu, Litong Gong, Weijie Li, Tiezheng Ge, Ming Zeng
Quantum Physics
Quantum state texture of maximal-resource Fourier states under Hawking radiation
Shengyan Ma, Yiding Wang, Xiaofen Huang et al. (+1 more)
Quantum state texture (QST) is a recently proposed
quantum resource that characterizes the nonuniformity of a density
matrix in a given computational basis. In the resource theory of QST, the
Fourier states f_k are maximally resourceful states, attaining the maximum QST measure. We study the QST behavior of three maximally resourceful states in a two-mode Dirac field under Hawking radiation from a Schwarzschild black hole, using a fidelity-based measure. We consider three cases: only mode B undergoes radiation, only mode A undergoes radiation, and both modes undergo radiation simultaneously. Our results show that for single-mode radiation, if the mode unaffected by radiation is orthogonal to the single-mode textureless state, the QST of the overall output state remains maximal and is independent of the Hawking temperature; we call this phenomenon orthogonality protection. When both modes undergo radiation simultaneously, orthogonality protection vanishes, and the QST of all three
Fourier states decreases with increasing Hawking temperature. Notably, the QST measure of f_3 under two-mode radiation is higher than that under only mode B radiation, indicating that QST can be redistributed among modes: QST gained by one mode under radiation can partially compensate for QST lost by the other. We further provide a general resource-theoretic explanation of orthogonality protection, showing that this phenomenon is independent of the specific form of the QST measure. These results reveal how QST degrades under Hawking radiation and provide a new perspective on the redistribution of
quantum resources in black hole backgrounds.
Read the full abstract ▾
All authors Shengyan Ma, Yiding Wang, Xiaofen Huang, Tinggui Zhang
Artificial Intelligence
PXRD-Bench: How Close Are We to Determining Crystal Structures from Powder Diffraction?
Dongyeop Woo, Nofit Segal, Nayoung Kim et al. (+3 more)
Determining
crystal structures from powder X-ray diffraction (PXRD) would enable routine characterization of newly synthesized materials, including those absent from reference databases. Generative models offer a promising approach, but existing evaluations remain fragmented and largely assume single-phase samples with known composition. To fill this gap, we introduce PXRD-Bench, a unified
benchmark that extends evaluation to unknown per-element atom counts and recovery of both structures in two-phase mixtures. It provides approximately 4.1 million training structures with simulated PXRD, tests generalization to unseen compositions and structural prototypes, and evaluates recovery from experimental PXRD after excluding structurally matching training entries. To provide a strong reference baseline, we also introduce Bragglite, a PXRD-to-structure model developed through a systematic study of PXRD modeling choices. It achieves a 44.2% Top-1 match rate on experimental RRUFF patterns with known composition, 24.6% above the strongest prior baseline. Further analysis yields several key findings: some models fail to benefit from experimental PXRD, while generalization to unseen structural prototypes and recovery from two-phase mixtures remain challenging. Together, our
benchmark, systematic studies, and Bragglite clarify the capabilities and limitations of current PXRD-to-structure models.
Read the full abstract ▾
All authors Dongyeop Woo, Nofit Segal, Nayoung Kim, Minkyu Kim, Rafael Gomez-Bombarelli, Sungsoo Ahn
Optics
Asynchronous Terahertz-field sampling with soliton microcombs
Alexandra Sakharova, Alexander E. Ulanov, Konstantin Wenzel et al. (+6 more)
Terahertz time-domain spectroscopy records the electric field of terahertz transients and underpins applications from materials science to non-destructive testing. Asynchronous optical sampling enables systems without moving parts, but its acquisition rate remains fundamentally limited by the repetition-rate of the driving femtosecond
lasers. Here, we demonstrate asynchronous terahertz-field sampling with two chip-based
soliton microcombs. Their ultrahigh 25 GHz pulse repetition-rate drastically reduces dead-time during acquisition, and stabilization via sideband injection-locking enables coherent averaging without active feedback loops. At 10 kHz acquisition rate, we achieve a dynamic range of 74 dB in a single 100 $\mu$s scan, and 100 dB after 50 ms of coherent averaging with a 1.6 THz bandwidth. Acquisition rates of up to 100 kHz are demonstrated, limited only by detection electronics. Together with a proof-of-concept silicon wafer thickness measurement, these results establish
soliton microcombs as compact sources for rapid terahertz-field sampling.
Read the full abstract ▾
All authors Alexandra Sakharova, Alexander E. Ulanov, Konstantin Wenzel, Milan Deumer, Alexander Dohms, Bastian Ruhnke, Lars Liebermeister, Robert B. Kohlhaas, Tobias Herr
Artificial Intelligence
XAI-Guided Fourier Domain Generalisation for Pure-Simulation Sim-to-Real Deep Reinforcement Learning
Wei Yao, Chuyao Wang, Ziwei Wang et al. (+1 more)
Deep reinforcement learning has shown strong potential for
vision-based autonomous driving, but policies trained in simulation often suffer severe performance degradation when deployed in the real world. Existing sim-to-real methods commonly address this gap through image-transfer-based domain adaptation or policy-agnostic domain randomisation. However, the former requires laborious acquisition of target-domain data and is often limited to the deployment domain, whereas the latter introduces unguided perturbations that may not reflect the underlying sim-to-real discrepancy, leading to suboptimal transfer performance. To overcome these limitations, we propose XFD, an explainable artificial intelligence (XAI)-guided
Fourier domain generalisation framework for pure-simulation sim-to-real reinforcement learning. The proposed framework extracts explainability attribution maps from simulated observations, and converts these attributions into policy-conditioned
Fourier-domain interventions. By perturbing appearance-sensitive spectral components while preserving task-defining spatial structures, XFD constructs robustness-enhancing visual domains entirely within simulation. Subsequently, the proposed robust policy optimisation guides the policy towards attending to task-critical features while suppressing sensitivity to irrelevant visual information. Extensive experiments in simulation and zero-shot real-world deployment demonstrate that XFD consistently improves traveled distance and reward over standard policy and randomly domain-randomised baseline, especially on unseen maps with cluttered backgrounds. Attribution-based analysis further proves that the proposed policy preserves more consistent task-relevant visual evidence across simulation and real-world observations.
Read the full abstract ▾
All authors Wei Yao, Chuyao Wang, Ziwei Wang, Nabil Aouf
Networking
CALIPER: Calibrating HPC/AI Network Simulators
Seyyidahmed Lahmer, Daniel Seibel, Srijeet Basu et al. (+9 more)
Network simulation plays an important role in HPC/AI system research by enabling exploration of network architectures, communication strategies, and system configurations at scales that may be impractical to study experimentally. Existing approaches to network simulator calibration and validation lack a principled methodology that combines parameter
alignment, constrained inference, and independent validation for collective-centric HPC/AI networks. We present CALIPER, a methodology for systematic calibration and validation of HPC/AI network simulators. CALIPER distinguishes directly observable system parameters from auxiliary parameters which are configured via calibration and constrains the latter to feasible domains consistent with the target system. We evaluate CALIPER's hierarchical, scale-aware calibration and its effects on parameter identifiability and generalization using collective communication
benchmarks on a reference system with up to 64-nodes at the Jülich Supercomputing Centre. Validation results show that the calibrated parameter configurations identified by CALIPER predict system performance with less than 5% mean errors across scale-up and scale-out network domains. CALIPER demonstrates that hierarchical, constrained calibration can produce simulator configurations whose predictive accuracy persists beyond the observations used for fitting.
Read the full abstract ▾
All authors Seyyidahmed Lahmer, Daniel Seibel, Srijeet Basu, Arno Troch, Jayesh Badwaik, Subhomoy Bhattacharyya, Abubakr Nada, Debjyoti Bhattacharjee, Vinay Kumar Baapanapalli Yadaiah, Michael Peeters, Andreas Herten, Darayus Adil Patel
Classical Analysis and ODEs
Schatten–Lorentz properties of commutators with matrix weights
Tuomas Hytönen and Xingyan Quan
We study the membership of commutators $[b,T]$ of Calderón–Zygmund operators in the Schatten–Lorentz classes in the
matrix-weighted setting. Working on a space of homogeneous type $(X,\rho,\mu)$ with a
matrix $A_2$ weight $W$, we characterise, for $p\in(1,\infty)$ and $q\in[1,\infty)$, the Schatten–Lorentz norm $\|[b,T]\|_{S^{p,q}(L^2(W))}$ in terms of an oscillation (Besov-type) norm $\|b\|_{\mathrm{Osc}^{p,q}}$ of the symbol: for all $p\in(1,\infty)$, finiteness of $\|b\|_{\mathrm{Osc}^{p,q}}$ is necessary; and for $p>p(\eta)$, where $p(\eta)$ depends on the regularity of the kernel of $T$ and a dimensional parameter of the space, the two-sided estimate $\|[b,T]\|_{S^{p,q}(L^2(W))}\approx\|b\|_{\mathrm{Osc}^{p,q}}$ holds. Extending a recent unified framework for unweighted and scalar-weighted versions of such results by one of us to the
matrix-weighted setting, our approach combines the dyadic representation of singular integral operators with the nearly weakly orthogonal (NWO) sequences of Rochberg and Semmes.
Read the full abstract ▾
Logic
Remarks on ultrafilters exponentiation
Lorenzo Luperi Baglini and Boris Šobot
Recently, two exponentiations of ultrafilters have been introduced in [7]. We rephrase them in nonstandard terms, and use the nonstandard formulations to characterize some of their properties. We also study their relations with divisibility by introducing factor-picking functions and highly exponential ultrafilters.
Read the full abstract ▾
Computation and Language
Ontology-Based Post-Coordinated Expression Assembly: from Clinical Narrative to Queryable Knowledge Graphs
Johann Pignat, Christophe Gaudet-Blavignac, Julien Ehrsam et al. (+7 more)
Clinical oncology documentation records decisive information in free text, where it remains readable but not retrievable at scale, leaving cohort identification dependent on incomplete diagnostic codes or manual review. The current standard, coding individual terms, captures what a document mentions, not what it asserts: meaning lies in how those terms relate to one another. We present a method for deriving queryable knowledge graphs from clinical narratives, conforming to the Extended Grammar of SNOMED CT across multiple ontologies. The method specifies how entities should be annotated and related. It defines five relation classes, each bound to a construction rule that assembles the entities and relations into post-coordinated expressions. A deterministic implementation, UNIGRAPHE, applies those rules to construct knowledge graphs from annotated text. Applied to expert-annotated real-world oncology documentation, UNIGRAPHE reconstructed 5,299 annotations into 2,196 post-coordinated expressions and shows that relation structure is the missing piece to make clinical narrative formally queryable.
Read the full abstract ▾
All authors Johann Pignat, Christophe Gaudet-Blavignac, Julien Ehrsam, Jean-Philippe Goldman, Cyrille Duret, Phil Cheng, Mina Bjelogrlic, Michel Cuendet, Christian Lovis, Olivier Michielin
Logic in Computer Science
Some complete $\omega$-powers of a one-counter language in the difference hierarchy
Jacques Duparc, Olivier Finkel and Dominique Lecomte
Fix a finite alphabet $\Sigma$ having at least two letters, and natural numbers $n,k\geq 1$. We denote by $D_n(\mathbf{\Pi}^0_k)$ the class of differences of $n$ subsets of the Cantor space $\Sigma^\omega$ in $\mathbf{\Pi}0_k$, and by $D_n-(\mathbf{\Pi}^0_k)$ the class of complements of sets in $D_n(\mathbf{\Pi}^0_k)$. We prove, for any natural numbers $n,k\geq 1$, the existence of a finite alphabet $\Sigma$ having at least two letters and a finitary
language $U_{n,k}$ over $\Sigma$ accepted by a one-counter automaton such that the $\omega$-power $$U_{n,k}^\infty:={ w_0w_1\cdots\in\Sigma^\omega\mid\forall i\in\omega ~~w_i\in U_{n,k}}$$ is $D_{2n-1}(\mathbf{\Pi}^0_k)$-complete. We also prove a conditional result for the dual classes $D^-_{2n}(\mathbf{\Pi}^0_k)$: under the assumption that there exists a finitary
language $V_{n,3}$ accepted by a one-counter automaton, such that its $\omega$-power $V_{n,3}^\infty$ is $D^-_{2n}(\mathbf{\Pi}^0_3)$-complete, there would also exist, for any integer $k>3$, a finitary
language $V_{n,k}$ accepted by a one-counter automaton, such that its $\omega$-power $V_{n,k}^\infty$ is $D^-_{2n}(\mathbf{\Pi}^0_k)$-complete. Finally we add another conditionnal result about Wadge degrees of rational $\mathbf{\Delta}^0_2$ omega powers, which completes the recent results of [FGL26].
Read the full abstract ▾
Nuclear Theory
Representation dependence of the neutron-proton kinetic energy in adiabatic deuteron stripping
Jin Lei
Deuteron stripping reactions such as (d,p) are a main tool for studying the single-particle structure of
nuclei, and they are usually analyzed with the adiabatic distorted-wave approximation (ADWA). With nonlocal nucleon optical potentials, ADWA cross sections change by tens of percent, up to a factor of two, depending on which realistic
neutron-
proton interaction describes the deuteron. Continuum-discretized calculations change much less, and exact Faddeev calculations reduce the spread by about a factor of ten. Should this sensitivity be read as a probe of
neutron-
proton physics at short distances? One identified route for it in ADWA is a single number, the
neutron-
proton kinetic energy averaged over the region where the two nucleons interact. I change the AV18 interaction only at short distances, keeping
neutron-
proton scattering, the deuteron binding energy, and the long-range tail of the deuteron unchanged. This number then rises from 218 to 314 MeV, more than half of its spread among five realistic interactions, while the coefficient that sets the low-momentum shape of the form factor moves by only one percent of its realistic range. The number therefore reflects how the interaction is written, not anything measured. It enters ADWA through two features that the other calculations treat differently: ADWA weights the deuteron where the two nucleons are close, and it assigns the deuteron energy to the pair however fast the two nucleons move. For this short-range change, an average that keeps the full Gaussian shape of the nonlocal potential, instead of expanding it to lowest order, rises by only 6 MeV against 96 MeV.
Read the full abstract ▾
Computer Vision
Auditing Observation-Quality Shortcuts in the Evaluation of Person-Centered Motion for Video Anomaly Detection
Chaehyun Kim and Changwon Lim
Video anomaly detection (VAD) identifies deviations from normal patterns, but such deviations do not necessarily correspond to safety-critical events. We focus on human-centered emergencies and violence and investigate whether motion-based anomaly discrimination reflects task-relevant motion information or observation-quality differences arising from person detection, pose estimation, and tracking. We introduce an observation-quality-aware evaluation framework that combines quality diagnostics, quality balancing, motion orthogonalization, and robustness analyses. On human-centered categories of UCF-Crime, quality-only features match or exceed the discrimination of multi-cue motion in the original cohort, revealing a strong observation-quality shortcut. After cardinality matching brings all 11 measured quality variables below a prespecified marginal balance threshold, multi-cue motion consistently outperforms quality-only features across Logistic Regression, Random Forest, and RBF-SVM, yielding ROC-AUC gains of 0.046-0.141. Because nonlinear quality-only prediction remains above chance after matching, we interpret this result as additional discrimination beyond measured marginal quality rather than as evidence of quality-free or causal motion understanding. These findings show that high VAD performance should not be interpreted as evidence of abnormal-motion modeling without
auditing the reliability of the person-observation pipeline. Our study provides a systematic protocol for evaluating observation-quality shortcuts and for assessing whether motion-based discrimination survives explicit control of detection, pose-estimation, and tracking reliability.
Read the full abstract ▾
Cryptography and Security
Jincheng Ying, Li Wenlin, Wencheng Yang et al. (+1 more)
White-box
neuron pruning has shown that refusal in large
language models (LLMs) can depend on a small, removable set of feed-forward (FFN) channels. Existing attacks either need carefully paired “twin”
prompts or score each
neuron in isolation.
Pruning also has an unavoidable capability cost, so over-removal compounds the loss. TwinBreak removes $1\%$–$5\%$ of parameters and NeuroStrike removes $0.3\%$–$0.7\%$ of
neurons in targeted layers. We model safety as a small
circuit whose
neurons contribute jointly rather than independently. We propose CircuitBreak, a
neuron-
pruning attack that greedily searches for a sparse set of FFN
neurons whose joint ablation flips a refusal readout. CircuitBreak re-scores every candidate after the
neurons already removed and needs no twin
prompts. On ten instruction-tuned models, the greedy search converges after removing $29$–$150$
neurons ($0.0066\%$–$0.0388\%$ of the MLP). The pruned models raise the attack success rate from $2.24\%$/$7.35\%$ to $87.54\%$/$87.54\%$ on Llama-3.2-3B-Instruct/Llama-3.1-8B-Instruct and from $8.95\%$/$5.43\%$ to $88.50\%$/$88.18\%$ on Qwen2.5-7B-Instruct/Qwen2.5-14B-Instruct, as judged by Llama Guard 3 8B on StrongREJECT.
Read the full abstract ▾
All authors Jincheng Ying, Li Wenlin, Wencheng Yang, Yinhao Xiao
Human-Computer Interaction
Foresee: Making Procedural Futures Tangible through Generative Previews
Yusi Sun, Ying Jiang, Jiayin Lu et al. (+4 more)
Procedural guidance can show users how to act without explaining what an action enables later. We present Foresee, an interface combining next-step guidance with consequence previews, method comparisons, and cross-step access. A hybrid pipeline combines structured rendering, image editing, and
video generation. A literature-informed framework and six-participant formative study shaped its design. A 24-participant study compared AR guidance, AR with generated
video, and Foresee across digital painting, gameplay, and cocktail making. Exploratory painting comparisons showed higher correct completion without a recorded error than AR and higher post-task quiz accuracy than generated-
video guidance under task-wise correction. Interviews revealed that users could overlook valued explanations, match visible outcomes without understanding later possibilities, or wait for concrete demonstrations when symbolic cues were unclear. These findings inform generative previews that explain what actions enable or constrain while supporting ongoing execution.
Read the full abstract ▾
All authors Yusi Sun, Ying Jiang, Jiayin Lu, Jing Qian, Yin Yang, Yong-Hong Kuo, Chenfanfu Jiang
Other Condensed Matter
Deterministic positioning accuracy of B centers in hBN by electron irradiation
A. Núñez Marcos, C. Arnold, J. Barjon et al. (+3 more)
Hexagonal boron nitride (hBN) is a promising material platform for
quantum technologies, enabling both the realization of nanophotonic devices and the hosting of point defects that act as high-quality single-
photon emitters (SPEs). Among these defects, B centers, which reproducibly emit at 436 nm, can be generated individually at predefined positions using electron beam irradiation, a crucial capability for top-down
quantum photonics. Yet, the spatial accuracy of emitter creation strongly conditions the potential applications. Here, we generate arrays of individual B centers in a scanning electron microscope (SEM) with in-situ cathodoluminescence (CL) monitoring. We characterize the sources of random deviations from the target positions to identify the limitations to their deterministic positioning. In particular, we measure the distance between the electron beam and the emitter at the moment the emitter is created based on CL mapping. This approach provides a direct estimation of the spatial extent of the beam-emitter interaction. We find this contribution to be the main constraint to the emitter fabrication accuracy, yielding a standard deviation of 27 nm. Additional contributions, primarily from the finite density of UV-emitting carbon dimer precursors, increase the overall positional standard deviation to 40 nm. Our results highlight the possible improvements to the B center creation technique to reach sub-10 nm accuracy.
Read the full abstract ▾
All authors A. Núñez Marcos, C. Arnold, J. Barjon, S. Buil, J.-P. Hermier, A. Delteil
Materials Science
Modelling the dynamics of the martensitic transformation in Ni$_2$MnGa with machine-learning force-fields
Markus E. Gruner, Yulia Klunnikova, Mike J. Bruckhoff et al. (+2 more)
The martensitic transformations in stoichiometric Ni$_2$MnGa and related Heusler
alloys are governed by a delicate interplay between electronic structure, magnetism, and
lattice dynamics. While density functional theory (DFT) has successfully mapped the energy landscape of modulated martensites, it remains computationally prohibitive for capturing finite-temperature dynamics across relevant length and time scales. Here, we bridge this gap by constructing DFT-trained machine-learning force-fields (ML-FF) interpolated from ab initio molecular dynamics. We demonstrate that ML-FF trained on DFT data can reproduce key features of Ni$_2$MnGa, including the soft TA$_2$
phonon and a preference for nanotwinned martensites. Using classical molecular dynamics simulations, we explore the impact of adaptive nanotwinning on the austenite-martensite transition, which facilitates the dynamics of both, forward and backward transformation. We demonstrate, that the negative twin boundary energy along the entire transformations path predicted by DFT for $T=0$, supports the formation of ordered modulated martensites at finite temperatures. Our results pave the way for large-scale simulations of
magnetic shape
memory alloys with near-DFT accuracy.
Read the full abstract ▾
All authors Markus E. Gruner, Yulia Klunnikova, Mike J. Bruckhoff, Konstatin P. Skokov, Olga N. Miroshkina
Computation and Language
PATMT2: Two-stage domain adaptation for multilingual patent translation
Xiaowei Su and Xusen Cheng
We present PatMT2 (Patent Machine Translation, v2), a two-stage LoRA pipeline that adapts the Hy-MT2-7B dedicated translation model to four-
language (Chinese/English/Japanese/Korean) patent translation. On six translation directions, the Stage-2 checkpoint reaches a macro-average BLEU of 43.39 (BF16) to 43.57 (AWQ INT4), improving over the Hy-MT2-7B base by up to +8.8 BLEU and also exceeding Qwen3-32B and several external systems on the same average. We then measure the cost of adaptation: probing 38
languages absent from the fine-tuning data, the model still produces correct, fluent output on 34, while four low-resource
languages collapse into distinct, reproducible failure modes. We treat these as diagnostic cases rather than retention rates, and report their mechanisms without attributing them to a single cause. Paired bootstrap tests confirm the translation gains over all baselines (p < 0.001).
Read the full abstract ▾
Computer Vision
Teaching Vision-Language Models to Decline: Support Completion Against Mirage
Alfaj Uddin Ahmed, Md. Shaown Miah, Md Arid Hasan et al. (+2 more)
Vision-
language models answer questions about images they were never shown, a failure called mirage. Such an answer can be correct without being justified by the evidence, and the difference matters wherever a model's output is checked by someone who did not see its input. We locate the failure in the training objective: cross-
entropy separates a model that reads the image from one that ignores it by only the conditional mutual information between answer and image, and no training example rewards declining. We formalise support completion, TIDE, which also shows each training question with a blank image, a same-domain wrong image and an out-of-domain image, with targets that decline, and prove that under TIDE every image-blind predictor is suboptimal by at least 1.04 nats per example, that the optimum declines exactly on the counterfactual kinds while keeping its behaviour on valid images, and that for any finite model the error of its answer-or-decline decision, declining on valid input plus answering on invalid input, is bounded below by how well its representation separates valid from invalid evidence. On Qwen2.5-VL-7B under a frozen held-out protocol with two judges, the three-condition predecessor of TIDE reduced the invalid-evidence mirage rate from 0.759 to 0.097 and the blank-image rate from 0.700 to 0.003 at a false-abstention cost of 0.126; label-free grounding raised the rate to 0.859, a likelihood-advantage contrast moved it only to 0.718 because its empty-image term carries no gradient, and grounding applied after support completion re-opened the mirage. The four-condition protocol on three base models, with seeds and component ablations, is in progress and not reported here.
Read the full abstract ▾
All authors Alfaj Uddin Ahmed, Md. Shaown Miah, Md Arid Hasan, Syed Ishtiaque Ahmed, Sayeed Shafayet Chowdhury
Machine Learning (Statistics)
Online Individualized Treatment Selection under Partial Feedback: False Discovery Control and Minimax Optimality
Runzhe Gu, Zinan Zhao, Wenguang Sun et al. (+1 more)
We study online individualized treatment selection when only the outcome under the assigned treatment is observed. Our objective is to preferentially treat individuals whose potential outcomes under treatment exceed a
covariate-specific control reference, while limiting false discoveries among treated individuals. We propose Partial-Feedback Adaptive Individualized Treatment Selection (PAITS), which initializes learning from historical data and combines sequential estimation of local null risks, randomized exploration, and feedback-driven updates of the treatment selection threshold. Under regularity conditions, we establish asymptotic false discovery rate control and sublinear gap-weighted assignment regret relative to a population oracle. We further establish a minimax lower bound on assignment regret under prescribed uniform nuisance-estimation requirements and give conditions under which PAITS attains the corresponding constrained minimax rate up to logarithmic factors. These results quantify the tradeoff between accurate treatment assignment and nuisance learning under partial feedback. Simulation studies and a real-data analysis using the Jester recommendation
dataset illustrate the finite-sample performance of PAITS.
Read the full abstract ▾
All authors Runzhe Gu, Zinan Zhao, Wenguang Sun, Xintao Xia
Robotics
DeTac: Dual-branch Decoupling of Texture and Force Signals in Vision-Based Tactile Sensors
Zhirui Zhang, Yunkai Tang, Jiahang Li et al. (+6 more)
Vision-based tactile sensors enable fine-grained perception for dexterous robotic manipulation, but current marker-based designs tightly couple surface texture and contact force signals within a single observation. Existing data-driven decoupling methods rely on serial mask-based inpainting, which suffers from mask-sizing dilemmas caused by physical indentation effects, error retention, and lack of real paired data for supervision. To address these challenges, we propose DeTac, a novel dual-branch generative framework that concurrently extracts clean shading and marker images from a single mixed tactile image. DeTac leverages full-image synthesis to naturally bypass mask dependency and introduces a cross-branch synergistic optimization that enables mutual supervision, effectively preventing error retention. Furthermore, we construct a strictly aligned, real-world paired tactile
dataset using custom hardware to provide high-quality supervision. Extensive experiments demonstrate that DeTac achieves state-of-the-art decoupling accuracy, significantly facilitating downstream surface reconstruction and force estimation tasks while exhibiting cross-sensor generalization.
Read the full abstract ▾
All authors Zhirui Zhang, Yunkai Tang, Jiahang Li, Fan Fei, Jinnuo Zhang, Kairan Ding, Jieji Ren, Yandong Guo, Boxin Shi
Superconductivity
Photon echo from Maki-Thompson interference in a fluctuating s-wave superconductor
Hui Zhou and Yuan Wan
We study theoretically the nonlinear optical response of a fluctuating s-wave
superconductor driven by two successive terahertz pulses. We show that the response exhibits a
photon echo signal that arises from the
quantum interference between the time-reversed electron and hole trajectories in the Maki-Thompson process. By contrast, the transport of the fluctuating Cooper pairs responsible for the Aslamazov-Larkin conductivity does not generate an echo. The
photon echo signal can be readily measured using terahertz two-dimensional coherent spectroscopy, highlighting its potential to distinguish different fluctuation conductivity mechanisms.
Read the full abstract ▾
Differential Geometry
On the Weyl tube formula for nonlocal perimeters
Serena Dipierro, Jack Thompson and Enrico Valdinoci
We study the asymptotic behaviour of the nonlocal perimeter of tubular neighbourhoods. For a closed $C^{1,1}$ submanifold of Euclidean space, we show that a suitable rescaling of the nonlocal perimeter of its tube converges to an explicit constant times the Hausdorff measure of the submanifold. The constant is the fractional perimeter of the unit ball in the codimension of the submanifold. For finite point sets, we obtain an exact tube formula in terms of a generalised hypergeometric function. In contrast with the classical Weyl formula, this expansion has infinitely many terms, contains both noninteger and integer powers of the tubular radius, and has coefficients that depend on the relative positions of the points rather than only on their intrinsic geometry. We also establish a point-set expansion for general analytic kernels.
Read the full abstract ▾
Computation and Language
The interpreter as developer: An autoethnographic design study of building a computer-assisted interpreting tool through human-LLM co-creation
Zeming Zhao
Real-time AI captioning now serves audiences at
multilingual events, yet computer-assisted interpreting (CAI) tools are still built for interpreters, rarely by them. This article reports an autoethnographic design study in which a professional interpreter with no programming training built a local-first CAI system through expert-led 'vibe coding', co-creating with an LLM architect and a coding
agent. Analysis of the complete development dialogues shows how tacit professional knowledge, from
segmentation to booth infrastructure, was externalized into design after generic defaults failed in use. In a demanding public scientific event, three mechanisms of in-booth appropriation emerge under a firmly human-led division of labor; telemetry from a first fully instrumented assignment shows a fully local chain sustaining a median translation latency of 0.94 s across 971 utterances. The case is read as lead-user innovation newly enabled by generative AI, with implications for the technological literacy of interpreters and the future of interpreter training.
Read the full abstract ▾