Distributed Computing
Cross-Facility LLM Pre-training on HPC: Elastic Aggregation, Data Leasing, and Queue-Aware Placement
Zarè Palanciyan, Thomas van Osch, Douwe van der Wal et al. (+2 more)
Academic compute is fragmented: allocations are granted per facility, and facilities differ in accelerators and software stacks, schedule jobs independently, and share neither a network nor a filesystem. We present a system that pools such allocations to pre-train a single
language model across three supercomputers on two continents, up to 7,400km apart: Snellius (NVIDIA H100), LUMI and Frontier (both AMD MI250X). It combines (i) DiLoCo-style two-loop training with an elastic,
token-weighted Nesterov outer step for which zero, one or many live sites are all normal states; (ii) DARL, a data-leasing protocol whose heartbeat-backed leases guarantee that, within an epoch, no sample is trained twice or lost under crashes, late joins and work stealing; and (iii) queue-aware placement driven by unprivileged sbatch –test-only probes. Training Qwen3-0.6B on C4 for 20,000 optimizer steps, the three-site run reaches a held-out perplexity of 34.7, against 28.2 for a centralised baseline. Per-round overhead (weight exchange and checkpointing) stays near 110 s regardless of the number of local steps H, so its share of wall-clock time falls from 32% at H=100 to 5.7% at H=1,000 and 3.1% at H=2,000. In a 23.8 h three-site run with four site departures, 1.2% of granted data blocks were reclaimed and none was duplicated or lost. In an idealized queue-model projection, queue-aware placement shortens time-to-target by 18-43% compared with waiting for all sites to be allocated. Cross-site pre-training is thus operational rather than competitive: it turns fragmented allocations into one training run at a measured cost.
Read the full abstract ▾
All authors Zarè Palanciyan, Thomas van Osch, Douwe van der Wal, Olivera Kotevska, Tim Kok
Machine Learning
Beyond Random Splits: Evaluating Drug-Target Affinity Models Under Chemically and Biologically Motivated Distribution Shifts Copy
Minjae Chung, Clara Li, Malar Paavai Muthukumaran et al. (+8 more)
Drug-target affinity (DTA) prediction is widely used to prioritize candidate compounds before costly experimental screening. DTA models are often compared under a single data split, even though deployment may require extrapolation to new chemical series, new
protein targets, or both. We ask whether the distribution shift used for evaluation changes which architecture appears best. We curate 718,800 unique drug-
protein pairs from the ChEMBL and BindingDB
datasets. We compare a Morgan-fingerprint +
protein-CNN baseline with 12 controlled architectures that combine four drug representations with three ESM-2 interaction modes. Mean validation RMSE increases from 0.950 and 0.945 under scaffold and fingerprint-cluster OOD to 1.299 and 1.321 under
protein-cluster and dual OOD. Model rankings are similar across the two chemical shifts (tau = 0.79), but agreement with scaffold OOD falls under
protein OOD (tau = 0.39) and reverses under dual OOD (tau = -0.55). Held-out evaluation, repeated seeds, group-aware bootstrap analysis, and a size-matched control support the same conclusion: architecture selection depends on the form of extrapolation, not only on average error or training-set size. DTA
benchmarks should therefore match the chemical and target shifts expected at deployment.
Read the full abstract ▾
All authors Minjae Chung, Clara Li, Malar Paavai Muthukumaran, Shaunna Wang, Aniket Ramkrishnan Iyer, Shaun Qien Yeau Tan, Harinishree Sathu, Micky C. Nnamdi, J. Ben Tamo, Benoit Louis Marteau, May Dongmei Wang
High Energy Physics (Phenomenology)
Big Bang Nucleosynthesis Constraint on $B$-mesogenesis
James M. Cline, Jonas Frerick, Sebastian A. Naja et al. (+1 more)
$B$-mesogenesis is a low-temperature baryogenesis mechanism, that uses $CP$-violation of the $B$
mesons in the standard model to fulfill the Sakharov conditions. It requires a late period of matter domination by a scalar field $\Phi$ that decays into $b\bar b$
quarks, reheating the Universe below the QCD phase transition. We investigate the impact of this mechanism on Big Bang Nucleosynthesis, with respect to the reheating temperature $T_{rh}$, and the mass of $\Phi$. Injection of
hadrons skewing
neutron/
proton interconversions are the most important effect, followed by
neutrino heating from $\Phi$ decays. For $M_\Phi\lesssim 100\,$GeV, $T_{rh}$ must be $\gtrsim 10\,$MeV.
Read the full abstract ▾
All authors James M. Cline, Jonas Frerick, Sebastian A. Naja, Yong Xu
Machine Learning
Measure Less, Know More: Self-Supervised Test-Time Feature Acquisition
Eeshaan Jain, Linus Bleistein, Bart Deplancke et al. (+1 more)
Recent progress in
multimodal, high-dimensional learning has enabled foundation models to process heterogeneous, large-scale data. However, at test time, acquiring all features or modalities can be prohibitively costly and often redundant. Sequentially selecting informative modalities is therefore critical, yet challenging when the downstream task or prediction target is unknown. To this end, we introduce ECHO-$k$, a task-agnostic and self-supervised learning principle for modality acquisition: we use a deep model's internal pretrained representations (e.g., from a foundation model) as proxy targets that summarize cross-modal information. We provide theoretical guarantees in a stylized linear setting that motivate a reinforcement learning (RL) policy for sequential modality selection. Across task-agnostic and label-free acquisition baselines, ECHO-$k$ consistently improves budgeted downstream performance across diverse foundation-model backends. Our method provides a principled route to cost-aware test-time deployment, with implications for any
multimodal system where measurements are expensive or time-constrained, and downstream tasks unknown a priori.
Read the full abstract ▾
All authors Eeshaan Jain, Linus Bleistein, Bart Deplancke, Charlotte Bunne
Robotics
I2CD: Direct Image-to-Convex Decomposition for Simulation-Ready Collision Geometry
Qian Wang, Liam Merz Hoffmeister, Brian Scassellati et al. (+1 more)
Physics simulators and motion planners require convex collision geometry, yet image-to-3D generative models output dense, frequently non-
manifold visual meshes. Bridging the two today takes a slow, brittle reconstruct-then-decompose pipeline of repair, decimation, and approximate convex decomposition. We present I2CD, which predicts a convex decomposition directly from a single RGB image. Rather than train a new image-to-3D model, I2CD freezes the pretrained Hunyuan3D-2 image-conditioned
diffusion transformer and shape decoder and trains only a lightweight cross-
attention head (38M parameters, under ten GPU-hours) whose learned "convex-slot"
tokens emit the halfplane parameters of $K$ convex polytopes. The output is compact, convex by construction, and loads into physics engines without any post-processing, in ${\sim}0.5$s per image. On $227$ held-out OmniObject3D and Google Scanned Objects instances, I2CD attains the highest volumetric IoU among eight reconstruct-then-decompose pipelines while running $6$-$37\times$ faster end-to-end. In a cross-simulator study in MuJoCo, PyBullet, Genesis, and Isaac Sim, every engine uses I2CD geometry as delivered, whereas raw generated meshes "load" everywhere but are silently replaced by a different collision shape in most cases or need seconds to minutes of per-object preprocessing. On a physical xArm7, I2CD produces planner-ready geometry for a $20$-object cluttered scene in $11$s versus $328$s for the strongest baseline, at comparable pick-and-place execution success ($85$ vs. $90$ of $100$ trials).
Read the full abstract ▾
All authors Qian Wang, Liam Merz Hoffmeister, Brian Scassellati, Daniel Rakita
Machine Learning
Causal Representation Learning with Instantaneous and Lagged Relations via Nonstationarity
Tatsuya Yamada, Hiroshi Morioka and Yoshinobu Kawahara
Causal representation learning for time-series data aims to identify latent states and their causal relations from observations. In this setting, an important challenge is to model both lagged causal relations across observation intervals and faster causal effects that appear as instantaneous relations within an interval, while accounting for nonstationarity in time-series data. However, methods that jointly handle these causal relations and nonstationarity remain limited. To address this gap, we establish sufficient conditions for identifying latent states up to component permutation and component-wise invertible transformations, and their instantaneous and lagged causal structures up to the same permutation, using an observed auxiliary variable, such as time or a condition label, associated with changes in transition-noise distributions. Based on these results, we propose iCReN, a framework that uses contrastive learning with discrete or continuous auxiliary variables to learn latent representations and estimate their instantaneous and lagged causal structures. Experiments demonstrate accurate recovery of latent states and both instantaneous and lagged causal structures on synthetic data and the utility of the learned representations for downstream
forecasting on real-world data.
Read the full abstract ▾
Computational Geometry
Point and Line Nearest-Neighbor Searching in 3-Space
Pankaj K. Agarwal, Esther Ezra and Micha Sharir
This paper presents data structures for nearest-neighbor (NN) searching problems involving points, lines, segments, and triangles in 3-space, achieving significantly better performance than the previously best-known results for these problems. For example, we present a linear-size data structure for answering NN queries with lines or segments amid $n$ points in 3-space with $O^*(n^{1/2})$ query time (where the $O^*(\cdot)$ notation hides subpolynomial factors). We also present a data structure of $O^*(n^4)$ size that answers such queries in $O^*(1)$ time. For the converse problem, in which we seek the nearest neighbor of a query point amid $n$ lines, segments, or triangles in 3-space, we present a linear-size data structure with $O^*(n^{2/3})$ query time. These results constitute a significant improvement over previous solutions. We obtain improved solutions for the two extreme regimes of (near-)linear storage and of fast query time. These results also yield trade-off bounds between the query time and the size of the data structure. Our results rely on several combinatorial and algorithmic results on arrangements of surfaces in 3-space and 4-space, particularly on recent results on vertical decompositions of substructures in such arrangements established by the authors.
Read the full abstract ▾
Materials Science
Magnetic-Order-Driven Nonlinear Photocurrents in PT-Broken Antiferromagnetic MnS2 and Altermagnetic MnSe2
Fu Li, Rajyavardhan Ray, Xiaoxiong Liu et al. (+6 more)
Nonlinear photocurrent responses generally require inversion-
symmetry breaking, but its origin can be
magnetic rather than structural. In this study, we investigate the bulk photovoltaic effect in
antiferromagnetic MnS2 and altermagnetic MnSe2, two pyrite-type
semiconductors whose
magnetic ordering breaks inversion
symmetry P and its combination with time reversal PT, while the accompanying
lattice distortion is minute. We calculate the shift current (SC) and injection current (IC) responses under linearly and circularly polarized light with and without
spin-orbit coupling (SOC) and identify their
symmetry-allowed tensor components using the corresponding
magnetic and
spin point groups, respectively. Linear SC and circular IC are
symmetry-allowed in both MnS2 and MnSe2 irrespective of SOC, whereas linear IC and circular SC remain forbidden in MnS2 but become allowed in MnSe2 only when SOC is included. These results highlight the importance of SOC and the corresponding MPG analysis for a complete description of nonlinear charge photocurrents in
magnetic materials. Beyond charge responses, MnSe2 supports linear shift and circular injection
spin photocurrents in the nonrelativistic limit, whereas all
spin-photocurrent responses are forbidden in MnS2. Our findings therefore establish inversion
symmetry breaking induced by compensated
magnetic order as a route to generating and controlling nonlinear photocurrents.
Read the full abstract ▾
All authors Fu Li, Rajyavardhan Ray, Xiaoxiong Liu, Babu Baijnath Prasad, Chen Shen, Yaqian Guo, Jeroen van den Brink, Hongbin Zhang, Harish K. Singh
Probability
Local Asymptotics for Entrance Laws of Reflected Lévy Excursions
Zhi-Hao Cui, Hao Wu and Wei Xu
This paper establishes uniform local large-deviation asymptotics for the entrance law $\underline n_t(dx)$ of excursions of a Lévy process with negative drift and locally regularly varying Lévy measure $\nu$, reflected at its running infimum. In the fixed-time regime, we prove that as $x\to\infty$, $$ \underline n_t\big((x,x+\delta]\big) \sim \underline n(\zeta\wedge t)\, \nu\big((x,x+\delta]\big), $$ uniformly in $\delta\in[\delta_0,\infty]$ for every $\delta_0>0$. Our principal result concerns the large-time regime and shows that, as $t\to\infty$, $$ \underline n_t\big((x,x+\delta]\big)\sim \int_x^{x+\delta} V(z) \,\nu(\beta t+dz), $$ uniformly in $x\geq0$ and $\delta\in[\delta_0,\infty]$, where $V$ denotes the renewal function. Using a fluctuation-theoretic decomposition of the killed
semigroup, we further derive corresponding uniform local large-deviation asymptotics for the Lévy process killed upon its first passage below zero.
Read the full abstract ▾
Cryptography and Security
Passing the Test You Trained On: Re-evaluating Prompt-Injection Detectors for LLM Agents
Zhuowen Liu
LLM
agents increasingly screen tool outputs with small
prompt-injection detectors, and teams choose among detectors by their scores on public
benchmarks. We ask whether those scores predict how a detector behaves inside an
agent. We replay the ground-truth tool calls of two
agent benchmarks, AgentDojo and tau-bench, without an LLM to obtain tool outputs that are benign by construction, label injected outputs by differential replay, and evaluate fifteen detectors, including Meta's
Prompt Guard 2, and two task-aware LLM judges on these outputs and on the BIPIA
benchmark. Detection rankings transfer poorly between
benchmarks: the best detector on BIPIA catches 2% of AgentDojo injections at a 1% false-positive rate, and a detector that catches 72% of AgentDojo injections catches 15% on tau-bench. False-positive rates on tool outputs, which range from none to over 90%, do transfer between the two
agent benchmarks. Where training data is public, the form of the training inputs explains the results. The BIPIA leader was trained on full BIPIA inputs, but having seen InjecAgent's attack strings as short
prompts does not help it find them inside tool outputs; the best detector on both
agent benchmarks shares no data with any
benchmark and was trained on
agent-style inputs. Evaluations meant to inform deployment should use the
agent's own tool outputs, report detection at a low false-positive rate, and audit what the detector was trained on.
Read the full abstract ▾
Cosmology and Nongalactic Astrophysics
Dark Energy Survey Year 6 Results: fast and interpretable posterior predictive checks for correlated cosmic probes
C. Doux, J. Muir, A. Ferté et al. (+59 more)
We present a fast and numerically stable framework for conducting posterior predictive distribution (PPD) tests of internal consistency in correlated cosmological probes. Our method, developed as a validation check for the Dark Energy Survey (DES) Year 6 combined analysis of weak
lensing and
galaxy clustering, employs a Gaussian-mixture model to efficiently approximate the high-dimensional PPD and introduces a scalar consistency metric, $\Delta_{\rm PPD}$, defined as the fraction of the PPD at lower probability density than the observed data. This method avoids challenges related to calibration and numerical stability that impacted PPD methods previously used in DES Year 3. After demonstrating performance with a toy model, we confirm that when applied to DES Year 3 measurements, $\Delta_{\rm PPD}$ identifies the same inter-probe tensions as published, calibrated PPD results while requiring only minutes per test. We further validate the method on several noise realizations of simulated DES Y6 analyses, demonstrating that $\Delta_{\rm PPD}$ responds coherently to injected inconsistency, while highlighting two effects which can complicate the interpretation of PPD tests for increasingly complex models. First, internal tensions that can be absorbed as shifts in nuisance parameters become more difficult to detect. Second, poorly constrained directions in parameter space can induce projection effects in PPDs for disjoint sets of observables, producing reports of tension even for consistent data. We conclude that $\Delta_{\rm PPD}$ provides a practical PPD-based diagnostic for multiprobe unblinding, to be interpreted with care alongside other validation checks. This framework has already played an important role in DES analyses and is well suited for application to forthcoming Stage-IV surveys such as LSST, Euclid, and Roman, where fast, robust internal consistency tests will be essential.
Read the full abstract ▾
All authors C. Doux, J. Muir, A. Ferté, M. Raveri, D. Sanchez Cid, N. Weaverdyck, E. J. Baxter, W. d'Assignies, T. Karwal, A. Amon, D. Anbajagane, F. Andrade-Oliveira, M. R. Becker, J. Blazek, A. Carnero Rosell, C. Chang, J. M. Coloma-Nadal, M. Crocce, J. DeRose, S. Dodelson, G. Giannini, E. Krause, E. Legnani, N. MacCrann, J. Mena-Fernández, A. Porredon, J. Prat, C. Preston, M. Rodriguez-Monroy, P. Rogozenski, R. Rosenfeld, S. Samuroff, E. Sanchez, I. Sevilla-Noarbe, M. Soares-Santos, C. To, M. A. Troxel, M. Yamamoto, B. Yin, J. Zuntz, S. Allam, M. Aguena, D. Bacon, D. Brooks, R. Camilleri, R. Cawthon, L. N. da Costa, S. Desai, P. Doel, K. Glazebrook, K. Honscheid, D. Huterer, K. Kuehn, J. L. Marshall, R. Miquel, M. Paterno, A. A. Plazas Malagón, M. Smith, E. Suchyta, D. L. Tucker, V. Vikram, J. Weller
Quantum Physics
Lucio De Simone, Lorenzo Capra, Arthur Vesperini et al. (+3 more)
Quantum entanglement is a fundamental resource in
quantum information theory, yet its general characterization and quantification remain challenging, especially in multipartite systems. In this work we investigate
entanglement from a
geometric perspective, focusing on the Riemannian structure induced by the Fubini–Study metric on the projective
Hilbert space of multi-
qubit quantum states. By exploiting the local-
unitary invariance of this metric, we derive the
entanglement distance (ED), a
geometric measure that quantifies
entanglement as an obstruction to locally minimizing the sum of squared Fubini–Study distances generated by local operations. We analyze the properties of ED for pure multi-
qubit states and discuss its behavior under local operations and classical communication. In particular, we show that ED reproduces established
entanglement measures in well-defined and restricted settings. For pure states of two
qubits, ED reduces to an exact monotone function of the concurrence and, independently, to an explicit monotone function of the
entropy of
entanglement. These results provide a clear
geometric interpretation of standard bipartite
entanglement measures within the present framework, while highlighting the limitations of such correspondences beyond the two-
qubit case.
Read the full abstract ▾
All authors Lucio De Simone, Lorenzo Capra, Arthur Vesperini, Leonardo Rossi, Loris Di Cairano, Roberto Franzosi
Computer Vision
Corrupted but Correct: Why Vision-Language Models Lie to Themselves Internally
Arun Josephraj Arokiaraj, Chenyi Huang, Zekun Wu et al. (+1 more)
A targeted
adversarial perturbation can drive a
vision-
language model's (VLM's) teacher-forced training loss for a fixed target caption to near zero, yet the same model, allowed to generate freely, produces the original, correct description with no trace of the target. We call this dissociation the train/inference gap, and give it a precise mechanistic account on Qwen2.5-VL-7B-Instruct using a controlled two-stage PGD attack on 200 held-out COCO images. First, we show that image-level pixel statistics, including a correctly re-implemented, texture-based attackability measure from the CNN robustness literature, have essentially no predictive power over which images are corrupted (best predictor r=-0.050, p=0.484; ridge regression R^2=0.069). Second, using the logit lens, we localise the gap to a single autoregressive step: the rank of the target
token, conditioned on the correct first
token already being generated, is fixed at exactly 3,488 out of 152,064 vocabulary entries for every image and every condition, with zero variance. Third, tracking target-
token rank across all 28 LLM decoder layers reveals that the visual encoder corrupts every image's representation by a comparable margin regardless of eventual outcome, but the
language model decoder then differentially arbitrates: amplifying the corrupted signal for susceptible images and actively suppressing it, past its clean-image baseline, for resistant ones (p<0.001, rank-biserial r=0.579). A linear probe on the merger hidden state separates these two outcomes with AUC=0.858, though we flag a circularity concern in this estimate. Together these results argue that
adversarial robustness in autoregressive VLMs is substantially a property of the
language decoder's prior, not the visual encoder, with direct implications for where faithfulness evaluations and defenses for deployed VLM systems should be targeted.
Read the full abstract ▾
All authors Arun Josephraj Arokiaraj, Chenyi Huang, Zekun Wu, Adriano Koshiyama
Chemical Physics
Electronic Density versus Geometry for Machine-Learned Molecular Absorption Spectra
Siddharth Dhanpal, Peter Elliott, Paolo Emilio Trevisanutto et al. (+2 more)
Molecular optical absorption spectroscopy provides a direct probe of electronic structure and is widely used for molecular identification, interpretation of photophysical behaviour, and planning of spectroscopy experiments. Calculating the absorption spectra using first-principle excited-state methods, however, is computationally demanding, at least compared to ground-state calculations, which limits their routine application across large molecular sets. Machine-learning (ML) surrogates can reduce this cost and allow rapid spectral prediction. However, their performance depends strongly on how molecular information is represented. Here, we compare using the ground-state electron density versus the molecular geometry as inputs to a ML model for predicting absorption spectra, for a training set of 6874 molecules selected from the QM7
dataset. For each of these molecules, the density was calculated using density functional theory (DFT) and the absorption spectrum was calculated using linear-response (LR) time-dependent DFT (TDDFT). Utilizing the ground-state density as the input to the ML model is motivated by the Hohenberg-Kohn and Runge-Gross
theorems, and the fact that the ground-state density encodes information about bonding, charge localisation, and electronic delocalisation. Hence, it may be a more judicious starting point for the ML model compared to the geometry, as it effectively decouples the chemistry of the ground-state. The question we test is whether the benefits of using the density outweigh the (notprohibitive) penalty of requiring an additional single-point DFT calculation for the density. We find that the density-based convolutional neural network achieves a validation correlation of 0.9926, compared with 0.9795 for the best geometry-based graph model, reducing the residual decorrelation, by approximately 64%.
Read the full abstract ▾
All authors Siddharth Dhanpal, Peter Elliott, Paolo Emilio Trevisanutto, Alin M. Elena, Gilberto Teobaldi
Digital Libraries
Still funded, no longer counted: how NIH's 2025 award reviews changed what the government counts as minority health research
Fangfang Xie, Jingwen Zhang and Haining Wang
Funders know their portfolios through software that classifies award text. The US National Institutes of Health (NIH) reports its spending in more than 300 categories mined this way, and work on classification and indicators treats the text as the applicant's to write. In 2025 NIH required "DEI
language" removed from awards not supporting DEI activities, and so policed the words it also counts. We followed population names through 37,790 continuing awards, checking them against practice records. Names were informative: in new awards with trial baselines, a title naming Black populations predicted a 60-percentage-point higher enrolled share. Text features predicted which names survived, and practice records added little. NIH's minority health category followed the names: of continuing projects carrying it, 99.4% with unchanged summaries kept it, against 15.7% of those whose summaries no longer named a racial or ethnic population, while the projects kept their funding. Blinded reviewers judged 42 of a random 50 such losses to have met its definition in FY2024. Comparing both summaries of 70 name losses, they found aims concerning the population recast in 49. Read alone, 63 FY2025 summaries no longer met the definition. Awards naming sexual and gender minorities were recorded as terminated 37.8 points more often after adjustment for listed terms, institute and activity. We call the mechanism word targeting, its boundary population targeting, and its product uncounted science: funded research the count no longer records. The count followed the edited text, and the edited record cannot say whether the research changed with it.
Read the full abstract ▾
Number Theory
Hausdorff dimension for exactly approximable matrices
Yubin He, Bixuan Li and Baowei Wang
We establish the Hausdorff dimension for exactly approximable
matrices for all dimensions, including the inhomogeneous case. More precisely, let $\psi:\N\to(0,\infty)$ be a non-increasing approximation function and let $\tau_\psi$ denote its lower order. For $m,n\ge1$ and $\boldsymbol\theta\in\mathbb R^m$, denote $\Exact_{m,n}(\psi,\boldsymbol\theta)$ the set of $m\times n$
matrices that are $\psi$-approximable with shift $\boldsymbol\theta$, but are not $c\psi$-approximable for any $0<c<1$. It is shown that \[ \hdim \Exact_{m,n}(\psi,\boldsymbol\theta)=m(n-1)+\frac{m+n}{1+\tau_\psi},\ \ \ {\text{for}}\ \tau_\psi>\frac{n}{m}. \] We adopt a probabilistic viewpoint, i.e. by an averaged counting argument to quantify the distribution of rational vectors and subspaces $$\left\{\frac{\mathbf p+\boldsymbol\theta}{q}: (\mathbf p, q)\in \mathbb{Z}^{m+1}\right\},\ \ \ \bigg\{\big\{A\in \Mat_{m, n}(\R): A\mathbf q-\mathbf p-\boldsymbol\theta=0\big\}: (\mathbf p, \mathbf q)\in \mathbb{Z}^{m+n}\bigg\}.$$
Read the full abstract ▾
Computer Vision
ChromaGS: Text-Driven Semantic Editing of 4D Gaussian Avatars
Antonio Canela and Jordi Sànchez-Riera
We present ChromaGS, a method for real-time,
language-guided color editing of animatable 3D Gaussian head avatars. Given a trained animatable avatar, users can instantly modify the color of semantic regions through natural
language, with edits applied at render time and no retraining required. Our key insight is to augment each Gaussian primitive with learned soft assignments to semantic regions and decompose colors into region-level base colors and Gaussian-level residuals. This decomposition enables coherent color transfer: modifying a region's base color propagates naturally through all associated Gaussians while preserving fine appearance details encoded in residuals. A two-stage
language pipeline translates text instructions into target colors, supporting both absolute specifications and relative adjustments. Unlike generative editing methods that may introduce unintended modifications, our approach provides deterministic, precisely localized semantic control. Experiments demonstrate faithful appearance preservation and intuitive interaction across diverse subjects. Project page and code are available at: https://a-canela.github.io/chromags/
Read the full abstract ▾
Computers and Society
Quantifying Ethereum Energy Consumption via Network Mapping
Yahn Costa Hackspacher, Cornelius Ihle, Vasundhara Shaw et al. (+4 more)
Ethereum's electricity use fell by about 99.95% after the move from proof of work to proof of stake. Service providers still need to report operational energy use, e.g. under the EU
Markets in Crypto-Assets Regulation (MiCAR). Existing estimates either apply one typical wattage to every node or start from aggregated monitoring counts. Both ignore attributes that nodes already advertise on the peer-to-peer network: client software, ARM or x86 hardware, hosting location, and validator role. We crawl the consensus and execution layers, assign each peer a wattage from those attributes using published measurements, and estimate the remaining incomplete peers with a Random Forest. On 6,934 peers from two Nebula crawls (19 and 22 June 2026), reachable nodes sum to 415 kW, or 3.63 GWh if that draw were held for a year. The same Lighthouse+Nethermind x86 wattage on every peer yields 431 kW. Observed attributes lower the total by 3.9%, mainly because nodes at Hetzner and other non-AWS clouds draw less than that home-desktop figure. AWS accounts for 15.6% of watts from 12.2% of peers, and validator-flagged nodes for 31.4% of watts from 25.7% of peers. The 415 kW snapshot is about 46% of the Cambridge Centre for Alternative Finance (CCAF) estimate of about 0.90 MW. Both use about 60 W per node, so the gap is mostly how many nodes each estimate includes. Rules cover 3,110 peers and the forest the other 3,824. On held-out labeled peers with client, architecture, and OS hidden, the forest's mean absolute error against the rule wattage is 4.3 W. Twenty-four-hour measurements on a gaming desktop differ from the predictions. After subtracting a 33 W idle graphics card that Ethereum clients do not need, both differences fall to about 19%.
Read the full abstract ▾
All authors Yahn Costa Hackspacher, Cornelius Ihle, Vasundhara Shaw, Dennis Trautwein, Geerd-Dietger Hoffmann, Bela Gipp, Moritz Schubotz
Computer Vision
Depth Hypothesis Guided Iterative Refinement for Event-Image Monocular Depth Estimation
Daikun Liu, Teng Wang and Changyin Sun
Event cameras hold excellent dynamic properties, showing great potential for
monocular depth estimation (MDE). However, existing methods mainly improve performance by optimizing contextual features, but still struggle with the ill-posed and nonlinear nature of direct full-depth regression. In this paper, we propose HypoDepth, the first event-image
monocular depth iterative refinement framework. By introducing a discrete Depth Hypothesis Volume (DHV), we transform the depth regression problem into a constrained depth search task. Specifically, we construct a 3D cost volume between the DHV features and contextual features and perform a multi-scale correlation search to guide stable residual optimization. This lightweight cost volume enables efficient global-to-local refinement across multi-resolution. Our method outperforms existing approaches on DSEC and MVSEC with state-of-the-art results and strong zero-shot generalization. Meanwhile, our tiny model achieves an excellent balance between accuracy and efficiency, enabling real-time performance on resource-limited devices.
Read the full abstract ▾
Quantum Physics
An exact semidefinite characterization of the stoquasticity cone of permutationally invariant Bell operators
Jan Li, Owidiusz Makuta, Evert van Nieuwenburg et al. (+1 more)
Bell correlations have been detected in atomic ensembles ranging from hundreds to hundreds of thousands of particles, witnessed by the permutationally invariant (PI) Bell operators. For all the operators realized so far, the maximally violating state can be brought, in a suitable basis, into a nonnegative-amplitude form. In that basis, the Bell operator is stoquastic. The coefficients parametrising these operators that at the same time lead to stoquasticity form a convex cone, the so-called stoquasticity cone. Its characterization had remained open beyond two-body operators, as for many-body operators the number of inequalities grows with the system size $n$. In this work we show that, in the thermodynamic limit, the stoquasticity cone admits an exact semidefinite characterization, valid at any operator order. The key observation is that, along each off-diagonal, the
matrix elements are the values of a single univariate
polynomial. Being univariate, this condition is equivalent to a linear
matrix inequality, with no relaxation hierarchy. Therefore, deciding stoquasticity becomes a single semidefinite feasibility problem. As its lowest-degree instances, this description recovers the known three-hyperplane characterization for two-body operators and yields, for three-body operators, a curved boundary.
Read the full abstract ▾
All authors Jan Li, Owidiusz Makuta, Evert van Nieuwenburg, Jordi Tura
Combinatorics
$k$-Pairing: A Generalization of the Partition Pairing Theorems
Xiaorui Niu and Diane Y. H. Shi
We extend the partition pairing theory of Andrews and Dastidar by replacing pairs with groups of $k$ equal parts. Two weight-preserving bijections give combinatorial interpretations of the joint pairing index–width distribution and the negative-rank enumerations. The first combines conjugation with the Stockhofe–Keith correspondence and sends the $k$-pairing index and width to the number of parts and the largest part, respectively. Consequently, their joint distribution is independent of $k$ and is given by a Gaussian
polynomial. We also obtain finite refinements that record the residual statistics. The second bijection sends simply $k$-paired partitions of negative $k$-pairing rank to $k$-regular partitions with marked internal gaps. In a fixed nonzero residue class, sign cancellation leaves rectangular partitions, while gap markings correspond to overlining choices with the smallest part not overlined. This gives direct combinatorial explanations of the divisor counts and the factor $1/2$ in the overpartition enumeration, extending the odd-divisor and odd-overpartition results of Andrews and Dastidar. Finally, motivated by the diagonal pairing of Andrews and Dastidar, we extend the construction to ordered tuples of $k$ Young diagrams. Applying the two-wing transfer operation to pairs of component diagrams defines an equivalence relation on these tuples. We show that two tuples are equivalent if and only if they have the same
cell-multiplicity function, and that every equivalence class contains a unique representative whose component diagrams are nested. We also determine the cardinalities of the equivalence classes and identify the nested representatives with plane partitions of rectangular shape.
Read the full abstract ▾
Graphics
The Shape of Speech: A Geometric Measure of Coarticulation for Speech-Driven 3D Facial Animation
Danzel Serrano and Przemyslaw Musialski
Speech-driven 3D facial animation can reproduce recognizable mouth poses. However, it can simplify the motion between them, and that motion carries coarticulation, the way the sounds around each sound shape its articulation. We introduce a
geometric measure of this trajectory shaping: lip-path length compared with the shortest route through the vowel, consonant and vowel positions of a
speech segment. In contrast to the endpoint chord, this consonant-aware route accounts for obligatory transit and avoids degeneracy, while preserving invariance to uniform motion gain. The measure needs only a forced
alignment, so it applies where no ground truth exists. We demonstrate it on four state-of-the-art methods, one per architectural family, real-time and offline. All four trace flatter lip trajectories than captured
speech. Against frame-rate-matched ground truth, DiffPoseTalk, ARTalk and FaceFormer show clear deficits, equivalent on this measure to removing 15-60% of real
speech's fast articulatory component. CodeTalker is marginal on the primary measure and clear on a companion measure. A pre-registered study with 97 viewers and 3,523 judgments underpins the measured direction: controlled damping of real motion lowers the score and is penalized, whereas exaggeration shows no detected penalty over the tested range. Viewers also prefer real
speech in 73.4% of sentence comparisons and, in the aggregate, on single words. Together, the measure, its calibration and the study identify a perceptually relevant loss of trajectory shaping and a concrete target for improving synthesized articulation.
Read the full abstract ▾
Materials Science
Recrystallization as a tool to characterize and control the properties of cold sprayed materials: case study on 316L steel
Petr Haušild, Jaroslav Čech, Petr Králíček et al. (+2 more)
Owing to the unique buildup mechanisms, recrystallization in cold sprayed materials significantly differs from that in conventional counterparts. In this study, the relationship between deformation-induced microstructural heterogeneity, recrystallization behavior, and mechanical properties of cold sprayed materials was investigated using EBSD and nanoindentation. The 316L stainless steel was selected as a model material. The deposit was characterized in its as-sprayed condition and after annealing at temperatures between 900 and 1100 K. Thermomechanical estimates suggest that locally, temperature can rise over 1000 K and very high dislocation densities (up to 1017 m-2) form at particle boundaries during impact, providing a strong driving force for subsequent recrystallization. The EBSD observations confirmed the intense defect accumulation and grain fragmentation localized near the particle interfaces. Recrystallization initiated preferentially in these highly strained areas after annealing at 1000 K and progressed with increasing temperature and holding time, leading to complete recrystallization after 1 h at 1100 K. The recrystallized grain-size distribution was well described by a heavy-tailed log-normal distribution, which indicates extremely heterogeneous strain partitioning generated during particle impacts. Nanoindentation measurements showed that the high hardness of the cold sprayed deposit remained largely unchanged after annealing at 900 K but gradually decreased as recrystallization progressed. We further demonstrate that controlled post-spray recrystallization can serve as a powerful tool for characterizing deformation during deposition and for tailoring the microstructure and mechanical properties of cold-sprayed materials.
Read the full abstract ▾
All authors Petr Haušild, Jaroslav Čech, Petr Králíček, Tomáš Heller, Jan Cizek
Cryptography and Security
Persona Guardrail: A Production-Grade Defense Framework for Agentic Systems
Bijeeta Pal, Sridhar Reddy Maddireddy, Muhaimin Bin Munir et al. (+4 more)
Large
language model-based
agents are increasingly deployed to perform domain-specific tasks by interacting with enterprise knowledge, tools, and external services. Existing runtime guardrails primarily target
prompt injection and other attack-specific behaviors under a black-box threat model, but provide limited guarantees that
agents operate within their intended functionality. As a result, production
agents remain vulnerable to malicious requests and out-of-domain queries that existing defenses often fail to distinguish. We present Persona Guardrail, a production-grade runtime defense framework that enforces explicit functional boundaries for customer-facing agentic AI systems through synchronous input and output validation driven by semantic allowlist and blocklist specifications. We also introduce PAGE (Persona-Aware Guardrail Evaluation), a
benchmark for evaluating function-specific guardrails across benign,
adversarial, and out-of-domain interactions on both user and
agent turns. Compared with a generic LLM-based guardrail, Persona Guardrail improves overall accuracy from 85.7% to 95.9%, increases out-of-domain detection from 57.3% to 93.5%, and reduces the false-approved rate from 25.0% to 4.7%. Currently deployed in production, Persona Guardrail meets its latency budget while sustaining a very low false-block and false-allow rate under realistic production workloads. These results demonstrate that Persona Guardrail provides a practical, scalable, and production-ready foundation for securing agentic AI systems.
Read the full abstract ▾
All authors Bijeeta Pal, Sridhar Reddy Maddireddy, Muhaimin Bin Munir, Zoltan Puha, Max Zhurovich, Adi Raghavendra, Sean Tout
Plasma Physics
Hong Qin
The susceptibility tensor of a homogeneous collisionless magnetized
plasma possesses parity-time (PT)
symmetry. For a stationary gyrotropic
equilibrium, with the wave vector $\mathbf{k}$ and the
equilibrium magnetic field $\mathbf{B}_0$ in the $x$–$z$ plane, the kinetic response satisfies $\chi(\omega^*,\mathbf{k})=R\chi(\omega,\mathbf{k})^*R$, where $R=\operatorname{diag}(1,-1,1)$. We show that the susceptibility PT-
symmetry is a reduced-response-space manifestation of PT
symmetry of the underpinning kinetic system. We identify parity $P$ and time reversal $T$ on the electromagnetic kinetic eigenmode space and show that the
Hamiltonian operator $\mathcal{H}_{\mathrm{VM}}$ of the linearized Vlasov–Maxwell system is PT-symmetric, satisfying $PT\mathcal{H}_{\mathrm{VM}}(PT)^{-1}=\mathcal{H}_{\mathrm{VM}}$. The PT
symmetry of the homogeneous magnetized Vlasov–Poisson eigenproblem is also established. Many kinetic instabilities associated with nonmonotonic distributions, rings, loss cones, and resonant energetic-particle populations are fundamentally spontaneous PT-
symmetry breaking, which is achieved through and only through Krein collisions. Three examples demonstrate distinct utilities of this analysis. Definite-sign cyclotron residues protect thermal Bernstein modes from spontaneous PT-
symmetry breaking. The nonmonotonic Dory–Guest–Harris distribution permits sign-indefinite residues and Krein collisions, for which we derive an analytical instability threshold and growth rate. The recently discovered mode-pole resonance instability provides a distinct realization of spontaneous PT-
symmetry breaking and illustrates how PT
symmetry can guide the discovery of kinetic instabilities with applications to fusion energy. These results establish PT
symmetry as a structural principle of
plasma waves and instabilities and open new directions for studying spectral
topology in kinetic
plasmas.
Read the full abstract ▾
Machine Learning
OptiSelect: How does the Optimizer Shape Data Curriculum?
Simin Fan, Alireza Abdollahpoorrostam and Martin Jaggi
Online data selection has demonstrated substantial efficiency gains for LLM pretraining by training on the most valuable candidates within each batch. Since a candidate's value is realized through its effective model update, principled selection should account for the optimizer step, which reshapes the raw gradient before it updates model parameters. We formalize this optimizer-aware selection paradigm as OptiSelect and present the first systematic study of how the optimizer shapes data selection. Our theory establishes a selection gain principle in which the advantage of online selection is governed by the discriminability of the optimizer-induced utility scores. We prove that sign-based and polar-tangential preconditioners of Lion and
Muon would suffer from a discriminability collapse which caps attainable gains from OptiSelect, whereas diagonal-adaptive optimizers such as AdamW and Sophia admit strictly better upper bounds. The proposed principle also yields a quantitative derivation of the optimal candidate oversampling ratio. Pretraining experiments on 124M and 720M models are consistent with our theoretical analysis and show that AdamW's diagonal-adaptive scoring geometry remains the strongest scoring geometry even with
Muon as optimizer. We further demonstrate that OptiSelect retains its benefits under data rephrasing, a technique used in modern data processing pipelines. Our findings provide theoretical foundations and practical guidance for co-designing optimizers and data selection in LLM pretraining.
Read the full abstract ▾
Solar and Stellar Astrophysics
The origin of dispersion in the wind mass-loss rate–luminosity relation of massive O stars
Jiri Krticka and Iva Krtickova
Stellar winds of hot
stars are accelerated by the radiative force that mainly comes from the line transitions of heavy elements. As a result, the wind mass-loss rate is expected to scale mostly with the
stellar luminosity. This scaling clearly emerges from the observational analysis of O
stars and their winds, but with substantial scatter. The origin of this scatter is unclear. Part of the scatter of the observational relation between the hot
star wind mass-loss rate and luminosity may originate from the sensitivity of the mass-loss rate to other
stellar parameters besides the luminosity, such as the effective temperature, mass, and
metallicity. We tested this possibility using our wind models. We calculated tailored wind models for
stellar parameters corresponding to selected individual O
stars from the Large Magellanic Cloud as derived from a previous analysis. To model the wind, we used our global wind code that calculates the radiative force from the solution of the radiative transfer equation in the comoving frame. Wind models calculated for particular
stellar parameters reproduce the observational mass-loss rates better than simple scaling relations derived from model grids. This indicates that the dependence of the mass-loss rate on the other
stellar parameters except the luminosity might explain part of the dispersion in the relation between the mass-loss rate and luminosity. However, some dispersion still remains, which might indicate that wind variability is another source of the dispersion.
Read the full abstract ▾
Artificial Intelligence
Jumping the Line: Exploiting Length Predictions in LLM Scheduling
Yuyang Dai, Rana Shahout and Mahmood Sharif
Efficient request scheduling is increasingly important for reducing completion time in large
language model (LLM) serving. Size-based policies such as Shortest Job First prioritize shorter requests, but output lengths are unknown before generation, so practical schedulers rely on predicted lengths. We introduce JIL, an attack on prediction-based LLM schedulers that manipulates the scheduling signal to obtain higher priority and reduce completion time. Using TRAIL as a case study, JIL optimizes an
adversarial suffix that causes a lightweight output-length probe to underestimate a request's length. We evaluate JIL on two
datasets and four LLMs across varied request profiles and deployment configurations. JIL reduces predicted output lengths by up to 83.4 percent, and
adversarial requests complete up to 1.53 times faster on average in end-to-end serving experiments. The reduction in predicted length is substantially larger than the change in actual output length, revealing a mismatch between the scheduler's estimate and the request's realized size. Response utility varies across models and tasks, exposing a trade-off between scheduling advantage and response quality. We also evaluate scheduler-side defenses and find that grouping length predictions into coarse intervals reduces JIL's scheduling advantage and mitigates delays to benign requests.
Read the full abstract ▾
Quantum Physics
Fault tolerance of quantum circuits with tensor networks and symplectic geometry
Soham Ghosh, Holger Boche and Andrew Tanggara
We develop an operator-algebraic framework for analyzing fault tolerance in
quantum circuits and establish necessary and sufficient conditions for exact recovery of the ideal computation under a prescribed noise model. For prescribed
circuit families and noise models, we derive a semidefinite program (SDP) based test whose infeasibility certifies that there does not exist any recovery map restoring its intended operation. For non-adaptive stabilizer
circuits, analyzing the stabilizer
symmetries of its components with
symplectic geometry, we obtain closed-form detection and logical-effect
matrices that characterize
circuit distance algebraically. Applied to a finite Hastings–Haah honeycomb Floquet-code
circuit, these
matrices certify the dynamical encoding of two logical
qubits and distance four under Pauli noise including measurement errors. Using these
matrices, we then derive MacWilliams identities for
circuit weight enumerators and obtain linear-programming (LP) upper bounds on
circuit distance. Output-code constraints give additional Singleton-like bounds. Under independent and identically distributed (i.i.d.) depolarizing noise, the weight enumerators determine decoder failure probabilities and yield upper bounds on
circuit pseudothresholds. Finally, we derive LP distance bounds for families of flag syndrome-extraction
circuits specified by their CNOT orderings, by incorporating the $t$-flag criterion as additional linear constraints to the distance-bounding LP. An analytical CSS Hamming-code
benchmark yields a tight distance-three bound attained by the Chao–Reichardt one-flag construction.
Read the full abstract ▾
Sound
LayerIt: Towards a Framework for Time-Aligned, Composable Music Visualizations
Fernando Azeredo and António Sá Pinto
Music information
retrieval often relates signal-derived, algorithmic, and symbolic information across different coordinate systems. Existing visualizations typically leave notation separate from physical time, while composites that combine them are assembled by hand. We present LayerIt, an open-source Python library for composing independent representations with notation on a shared performance-time axis. Given a score warped into performance time and a note-level
alignment, LayerIt emits a single SVG that preserves the score's MEI structure and keeps added components identifiable and restylable. We demonstrate this through a challenging beat-tracking analysis, in which tracker output, signal representations, and notation can be inspected together to locate metrical disagreement and other errors against both notated structure and performed time.
Read the full abstract ▾