Artificial Intelligence
The Argument and the Letterhead: Source-Position Coherence in AI Evaluation
Michele Loi
An argument can be surprising coming from a particular speaker without being a bad argument. Do AI evaluators keep these judgments apart? Two preregistered descriptive studies and a later Jev supplement collected 2,976 usable evaluations of six fixed texts about US AI policy, Germany's debt brake and Swiss
nuclear energy. Each text was presented under several source attributions. The key comparison asks whether the gap between two sources changes when the argument changes. On Sol, for example, a national-security argument received mean ratings of 0.359 under CODEPINK and 0.639 under College Republicans; a civil-rights argument received 0.742 and 0.721. A constant preference for one source cannot explain that pattern. Related interactions appeared across topics and recent model configurations, including those with
reasoning enabled, while several comparisons yielded small effects. The later European Jev supplement yielded five interactions below the adopted absolute reference of 0.05; its distinct rubric and interrupted collection limit comparison with the chat systems. Some written evaluations explicitly invoked a mismatch between a source and its attributed position. Taken together, the numerical and verbal evidence supports source-position coherence as a plausible explanation, alongside competing accounts involving credibility, authenticity and interpretation of the task. The paper develops this inference through controlled comparisons, reports conditional post hoc p-values in an appendix, and documents the human decisions and delegated checks behind an AI-conducted study.
Read the full abstract ▾
Artificial Intelligence
Textual User Taste: Natural-Language User Context for Foundation-Model Recommender System at Scale
Ghazal Fazelnia, Paul Gigioli, Eliza Klyce et al. (+16 more)
Foundation model recommender systems require user context that can be consumed by large
language models, reasoned over, and refined through natural-
language interaction. Traditional behavioral
embedding vectors remain highly effective for
retrieval and ranking, but they are opaque to users and not natively expressed for
language model workflows. We present Textual User Taste, a system that generates structured natural-
language taste profiles from listening behavior, interaction signals, content metadata, and optional user feedback, and deploys them to millions of Spotify users. We describe the end-to-end production lifecycle required to generate, evaluate, optimize, and maintain these representations at industrial scale, including
prompt development and compression, user steering, and integration with downstream personalization systems. Because no unique ground-truth taste profile exists, we introduce a multi-faceted evaluation framework to evaluate taste profiles as a production representation: they carry user-specific predictive signal independently, and when integrated with behavioral
embeddings, improve MRR by 0.6% for future-track prediction and NDCG@7 by 2.2% for search ranking. Our evaluation also reveals that taste profiles support positive natural-
language steering, while exposing important limitations, including challenges with negation and short-term temporal adaptation. These findings position taste profiles not as replacements for behavioral
embeddings, but as an interpretable and steerable interface between evolving user context and foundation-model recommender systems.
Read the full abstract ▾
All authors Ghazal Fazelnia, Paul Gigioli, Eliza Klyce, Sharon Zheng, Katie Zelvin, Ye Myat Thein, Anurag Deshpande, Seda Davtyan, Kate Remeika, Maya Hristakeva, Erik Franco, Karen Banzon, Peng Ge, Jacqueline Wood, Nandini Singh, David Murgatroyd, Mounia Lalmas, Yves Raimond, Andreas Damianou
Atomic Physics
Raman Transitions in Collinear Laser Spectroscopy for High-Precision Frequency Measurements
Julien Spahn, Hendrik Bodnar, Kristian König et al. (+2 more)
We demonstrate stimulated Raman spectroscopy on dipole-forbidden fine-structure transitions in a fast ion beam. It is applied to the $\mathrm{S}_{1/2}\rightarrow\mathrm{D}_{5/2}$ clock transition in $^{88}\mathrm{Sr}^{+}$ and combined with a subsequent E1 transition to allow for its detection by standard fluorescence detection. Building on this, the $\mathrm{D}_{3/2}\rightarrow\mathrm{D}_{5/2}$ fine-structure splitting is measured directly using a Doppler-free scheme employing two subsequent Raman transitions. These measurements demonstrate a path toward precision collinear
laser spectroscopy of narrow transitions on short-lived isotopes beyond the linewidth limit of conventional electric-dipole spectroscopy, which can be used to probe fundamental aspects of atomic and
nuclear physics.
Read the full abstract ▾
All authors Julien Spahn, Hendrik Bodnar, Kristian König, Wilfried Nörtershäuser, Robbe Van Duyse
Atomic Physics
Reconstructing Molecular-Ion Quantum States with Two-Color Strong-Field Scattering
Dina S. Eissa, Felipe Morales, Abraham Camacho Garibay et al. (+6 more)
Strong-field ionization of molecules can prepare coherent superpositions of ionic states whose amplitudes and relative phases govern the ensuing ultrafast dynamics, but accessing this information experimentally remains a challenge. Here we introduce two-color strong-field scattering, in which a strong mid-infrared field ionizes the molecule and launches an electron, while a weak second harmonic field resonantly couples vibronic states of the parent ion during the electron's continuum excursion, thereby introducing structural changes on the sub-
laser-cycle timescale. Scanning the relative phase between the two colors controls the ionic
evolution and maps it onto phase-dependent sub-angstrom bond-length changes retrieved from the rescattering signal. Using the example of N$_2$, we apply the two-color setup to probe the coherence between different ionization channels and to reconstruct the initial ionic superposition. Our results establish two-color strong-field scattering as a powerful structural approach to characterize the complex amplitudes of ion
quantum wavepackets.
Read the full abstract ▾
All authors Dina S. Eissa, Felipe Morales, Abraham Camacho Garibay, Vyacheslav Leshchenko, Cosmin I. Blaga, Misha Ivanov, Pierre Agostini, Maria Richter, Louis F. DiMauro
Quantum Physics
Circuit-level benchmarks of GKP-concatenated qLDPC Codes
Yuan Yao, Ruipeng Xing, Jianshuo Gao et al. (+1 more)
Scalable fault-tolerant
quantum error correction based on Gottesman-Kitaev-Preskill (GKP) codes requires finite-rate outer codes that can exploit analog GKP information under realistic
circuit noise. However,
quantum low-density parity-check (qLDPC) outer codes have not been systematically compared as candidates for
circuit-level GKP concatenation. Here, we
benchmark BB and tricycle codes using a unified simulation framework that progresses from code-capacity noise to repeated noisy syndrome extraction and schedule-resolved displacement propagation. Across all three noise models, analog-informed belief-propagation with ordered-statistics decoding (BP-OSD) yields higher finite-size crossing estimates than hard-decision decoding. In the full
circuit model, the analog-informed crossings are \(\sigma_{\mathrm{th}}\simeq0.212\) for the selected BB sequence and \(0.142\) for the tricycle sequence, corresponding to squeezing requirements of approximately \(10.46\,\mathrm{dB}\) and \(13.94\,\mathrm{dB}\), respectively. These results show that analog GKP information continues to improve decoding under
circuit-level noise. Under the common simulation assumptions used here, the selected BB sequence also yields a higher finite-size crossing estimate than the selected tricycle sequence.
Read the full abstract ▾
All authors Yuan Yao, Ruipeng Xing, Jianshuo Gao, Xiao Yuan
Software Engineering
Is there a future for models in the LLM era?
Marco Calamo, Massimo Mecella and Monique Snoeck
In an era where software development is deeply tied with Large
Language Models, does Model Driven Engineering (MDE) still make sense? This raises the question of the extent to which MDE can be successfully combined with an LLM approach to address the downsides of each approach separately. In this paper, we try to answer that question by investigating a central Research Question: Can
Agents powered by Large
Language Models improve code generated from UML diagrams and text specifications by Model Driven Engineering (MDE) tools? To investigate this problem, we developed a novel LLM-powered Multi-Agentic approach called ARTHUR (Architecture Refactoring Through Hybrid UML
Reasoning), a framework that aims at combining the reliability of MDE and the ease of use of LLMs. ARTHUR is designed to refactor legacy Java code produced by traditional, rule-based, MDE code generators from UML diagrams. To asses the answer to our question, we have refactored the legacy code of several projects from our custom
dataset crafted for this purpose. \name{} which made it possible to add support for modern frameworks like Spring Boot, while ensuring compliance with Model-Based Testing techniques to verify that the code still corresponds to the initial model's specifications. We then measured the results obtained in terms of time and cost, passing test rate, and compile@k, pass@k and \texttt{pass$^k$} metrics. We also observed the effect of generating code directly from the conceptual model without the refactoring. Our preliminary test results show that MDE could not be more far from retirement, after all.
Read the full abstract ▾
Game Theory
Truthful-in-Expectation MMS Allocations for Chores
Zehan Lin, Biaoshuai Tao, Xiaowei Wu et al. (+1 more)
We study truthful-in-expectation (TIE) mechanisms for allocating indivisible chores alongside ex-post maximin share (MMS) guarantees. For goods, Bu and Tao (FOCS 2025) established a (1/n)-approximation for TIE mechanisms, and this was substantially improved by Babaioff, Feige, and Manaker Morag (FOCS 2026), who established an \Omega(1/\log n) approximation, where n is the number of
agents. The corresponding problem for chores has received less
attention. The best-known result is due to Aziz, Li, and Wu (MAPR 2024), who gave a TIE mechanism with an O(\sqrt{\log n}) MMS approximation guarantee that holds only in expectation. They also established a 6/5 lower bound for TIE mechanisms for two
agents. In this paper, we present the first TIE mechanism for chores that achieves a constant ex-post MMS approximation guarantee. Specifically, our mechanism guarantees an ex-post ratio of 1.97 for any number of
agents n, which improves to 4/3 for n=2 and 3/2 for n=3. On the hardness side, we tighten the two-
agent lower bound to 4/3, showing that our mechanism is optimal for n=2. More generally, we establish a lower bound of 13/12 on the ex-post MMS approximation ratio achievable by TIE mechanisms for every n\ge 3.
Read the full abstract ▾
All authors Zehan Lin, Biaoshuai Tao, Xiaowei Wu, Yuhao Zhang
Computation and Language
Measuring Collapse and Correction in Homogeneous-Panel LLM Debate
Xin Li, Mengbing Liu and Chau Yuen
Multi-
agent large
language model (LLM) debate is often evaluated by whether final answers improve, but a changed answer is not necessarily a better one: the same discussion can correct an initially wrong majority or collapse an initially correct one. Standard final-accuracy evaluations conflate these opposing transitions. We introduce an auditable protocol for homogeneous debate on multiple-choice questions (MCQs) that records each run as a transition ledger over collapse, correction, onset, and signed intervention utility. On 6,925 MMLU-Pro debates, the protocol records 253 collapses together with the corrections that offset them, and the correction side changes how interventions should be judged. Replay experiments reveal the central tradeoff: freezing every debate prevents 251 collapses but loses 804 corrections under equal weights, and a leave-one-model-out probe-gated freeze that is tuned for net utility never fires, so collapse prevention alone can recommend the wrong policy. A compact pre-debate 8-probe screen is a triage signal: its unadjusted family-level association with conditional-collapse risk is high (G=7, Spearman $\rho$ = 0.893, exact two-sided p = 0.0123), but initial-majority accuracy is a close comparator ($\rho$ = -0.821; family partial $\rho$ = 0.767, p = 0.0877), so we do not treat it as calibrated or capability-adjusted prediction. Round-level traces localize many collapses to the first debate round, where early disagreement can precede both collapses and corrections.
Reasoning-mode models, including one model at two
reasoning-effort levels, also show both collapses and corrections. We release replayable schemas, coders, audits, cost cards, and zero-API rebuild scripts so future model rows can be compared under the same denominators and signed utility ledger.
Read the full abstract ▾
Differential Geometry
Pointwise energy monotonicity for stable Higgs bundles
Tianzhi Hu
We study the Dai–Li
conjecture concerning the pointwise monotonicity of the energy density along the $\mathbb C^*$-flow of stable $\mathrm{SL}(n,\mathbb C)$
Higgs bundles. We prove that the
conjecture holds in rank two: for every stable $\mathrm{SL}(2,\mathbb C)$
Higgs bundle, the energy density is pointwise nondecreasing along the $\mathbb C^*$-orbit. In contrast, we show that this phenomenon is genuinely rank-dependent. For every rank $n\geq 3$, we construct stable $\mathrm{SL}(n,\mathbb C)$
Higgs bundles for which the energy density fails to be monotone along the $\mathbb C^*$-flow.
Read the full abstract ▾
Computer Vision
Evaluating Hierarchy-Aware Deep Learning for the Recognition of Tironian Notes
Yule Kang, Thomas Gorges, Janne van der Loop et al. (+4 more)
Tironian notes are generally regarded as the first Latin shorthand system and are notable for their large, fine-grained symbol inventory. Their high visual similarity and large class set make manual reading time-consuming, leaving manuscripts that contain Tironian notes inaccessible to many researchers. Automatic recognition is also challenging because models must distinguish subtle differences in stroke shape and sign structure while realistic training data remain scarce. However, standard flat classifiers do not explicitly use visual or structural relations between related signs. This paper investigates whether structural relationships between Tironian notes can support automatic recognition. We use the Supertextus Notarum Tironianarum (SNT) by Martin Hellmann, which provides idealized sign forms and a hierarchical organization of Tironian notes. We compare flat ResNet18, ConvNeXt, Shifted Window
Transformer (Swin), and
Vision Transformer (ViT) classifiers with Hierarchical Deep Convolutional Neural Network (HD-CNN)-style coarse-to-fine models and hierarchy-aware routing models based on visual class cleaning and similarity-based re-clustering. The models are evaluated on handwritten samples and manuscript-domain samples from Vergilius Turonensis, both with and without limited few-shot adaptation to the manuscript domain. The results show that the relative performance of flat and hierarchical models depends on adaptation. On Vergilius Turonensis, HD-CNN achieves the best non-adapted Top-1 result with 45.43%, while flat classification reaches the best Top-1 result after few-shot adaptation with 82.09%. Overall, the results indicate that hierarchical structure can support Tironian note recognition, especially under non-adapted conditions.
Read the full abstract ▾
All authors Yule Kang, Thomas Gorges, Janne van der Loop, Franziska Marske, Nikolaus Weichselbaumer, Tino Licht, Vincent Christlein
Distributed Computing
Weaver: A System for AI-RAN Compute Sharing with Foundation Model Training
Leyang Xue, Tianxin Wang, Xin Zhe Khooi et al. (+6 more)
The emergence of AI-RAN infrastructure, which equips
cell sites with GPU-accelerated hardware, creates an opportunity to colocate non-RAN workloads with primary RAN processing. We explore using this spare capacity for decentralized training of foundation models (FMs), one of the most compute-intensive AI workloads. We present the first characterization of spare GPU capacity in AI-RAN systems at both micro-scale–across transmission slots within a
cell site–and macro-scale–across sites. Our analysis finds that 40-85% of GPU capacity is unused; although this capacity is temporally bursty at individual sites, it is spatially complementary across sites. To safely and efficiently harness these resources, we present Weaver, a system that opportunistically trains FMs alongside latency-critical RAN workloads without degrading RAN performance. Weaver adopts a RAN-first design: a spare-compute controller integrated into the MAC scheduler uses compute-aware scheduling to smooth RAN GPU demand and exposes more usable spare GPU capacity. A two-level elastic training framework then adapts to dynamic, heterogeneous spare capacity within and across sites. Experiments on an O-RAN-aligned system prototype show that Weaver creates up to 4.9x more usable spare compute and utilizes up to 83% of the available spare capacity. On a multi-site testbed, Weaver improves training throughput by 2.1-3.7x over baseline approaches.
Read the full abstract ▾
All authors Leyang Xue, Tianxin Wang, Xin Zhe Khooi, Jiaxun Yang, Dheeraj Mahendiran, Yufeng Xia, Mun Choon Chan, Myungjin Lee, Mahesh K. Marina
Optimization and Control
Projection-Free Algorithms for Nonsmooth Stochastic Convex-Concave Saddle-Point Problems
Khanh-Hung Giang-Tran and Soroosh Shafiee
We study nonsmooth convex-concave saddle-point problems over compact convex sets, assuming access to
stochastic subgradients of the payoff function. We develop single-loop projection-free
algorithms that use linear minimization oracles over the primal and dual domains. Unlike prior projection-free approaches that rely on smoothing, our methods are purely subgradient-based and handle nonsmoothness directly. This design makes the framework modular: when a linear minimization oracle is not used over a specific domain, the corresponding update can be replaced by a standard projection oracle without changing the single-loop structure. Thus, our framework covers both fully projection-free and hybrid oracle configurations. We prove anytime strong saddle gap guarantees under standard unbiased
stochastic oracle assumptions with bounded variance. Our
algorithms achieve an $O(\epsilon^{-2})$ iteration complexity, matching projected
stochastic subgradient methods. We also provide matching lower bounds for the corresponding oracle models, establishing minimax optimality. Our results show that, for nonsmooth
stochastic minimax problems, the computational advantages of linear minimization oracles need not come at the expense of statistical or oracle efficiency.
Read the full abstract ▾
Machine Learning
Multi-Attractor GNNs: Set-Valued Expressivity Beyond Unique Equilibria
Jialin Liu
Recurrent and
equilibrium graph neural networks (GNNs) often enforce a unique fixed point or use one training target per graph. Yet many combinatorial and scientific problems admit multiple valid solutions, with no preferred one. A designated target can then impose an arbitrary selection rule. For tasks invariant to node relabeling, a symmetric graph may have a symmetric solution set but no symmetric solution. We show that multiple
equilibria enable one weight-tied message-passing GNN to represent set-valued equivariant maps: different initializations approach different valid solutions. Under stated regularity assumptions, we first construct globally Lipschitz, permutation-equivariant dynamics that converge almost surely to valid solutions and reach every solution branch with positive probability. We then establish approximate realization by recurrent message passing with continuous component maps, with arbitrarily small update and limiting errors and arbitrarily high probability. This goes beyond standard universality arguments: although message passing alone cannot distinguish symmetric nodes, the evolving state keeps nodes distinguishable at every finite step without auxiliary node identifiers. Such dynamics can be learned without solution labels using problem-specific energies. On Ising ground states, structural module detection in
protein graphs, and chemical reaction steady states, the learned updates produce multiple high-quality predictions with high numerical convergence rates. They achieve better average solution quality than the tested unique-
equilibrium, single-target, and feedforward baselines, while remaining competitive with much larger
diffusion-based solvers.
Read the full abstract ▾
Medical Physics
Physics-informed self-supervised generation of digital brain MRI phantoms from weighted images using differentiable MRI simulation
Kseniya Belousova, Anna Konanykhina, Zilya Murzina et al. (+2 more)
Purpose: To develop a physics-informed, self-supervised framework for generating digital
brain MRI phantoms directly from conventional weighted MR images without requiring ground-truth parametric maps or anatomical
segmentation. Methods: The framework predicts T1, T2, and
proton density (PD) maps from T1-, T2-, and PD-weighted images and reconstructs the input images through an MRI signal model. Three generative architectures (variational autoencoder (VAE), generative
adversarial network (GAN), and flow-based model) were compared. Models were pretrained on 3,739 synthetic
brain slices generated using digital phantoms and an analytical MRI signal model, followed by fine-tuning on 90 real
brain slices from three healthy volunteers. The best-performing architecture was subsequently fine-tuned using the differentiable MR-Zero numerical MRI simulator and evaluated on 30 held-out real slices. Results: The flow-based model demonstrated the highest overall performance and preserved fine anatomical details better than the VAE and GAN. After analytical-model fine-tuning, it achieved MS-SSIM values of 0.955-0.985 and PSNR values of 26.68-32.63 dB across T1-, T2-, and PD-weighted images. Fine-tuning with MR-Zero increased T1-weighted reconstruction quality from 0.955 to 0.977 (MS-SSIM) and from 26.68 to 30.60 dB (PSNR), and provided high robustness of metrics across different MR image weightings. The resulting digital phantoms also enabled simulation of images using previously unseen acquisition protocols. Conclusion: The proposed framework enables physics-informed generation of reusable digital
brain MRI phantoms from weighted images using limited real-world data. Combining synthetic pretraining with differentiable numerical MRI simulation provides a practical approach for physically grounded MRI data augmentation without requiring reference parametric maps.
Read the full abstract ▾
All authors Kseniya Belousova, Anna Konanykhina, Zilya Murzina, Walid Al-Haidri, Ekaterina Brui
Computation and Language
When Words Speak Louder than Images: Towards Understanding Language Bias in Vision-Language Models
Yizhou Fang, Siyue Chen, Zimo Qi et al. (+3 more)
Despite substantial progress across downstream applications,
vision-
language models (VLMs) remain susceptible to
language bias, often prioritizing linguistic cues over visual evidence and consequently producing incorrect predictions. Prior studies have proposed various approaches to understanding and mitigating
language bias in VLMs, yet their findings often conflict due to the difficulty of tracing how
language bias propagates within black-box VLMs. Building on the word completion task, we trace how
language bias propagates through VLM inference by (1) proposing a diagnostic framework that decomposes the inference process into four distinct yet interdependent stages to trace the propagation of
language bias; and (2) examining how two key factors underlying
language bias, i.e., linguistic priors and cross-modal coverage, evolve across these stages and ultimately give rise to incorrect predictions. The linguistic prior captures the strength of statistical bias induced by the
language model component of a VLM and represents the origin of
language bias, whereas cross-modal coverage measures the extent to which linguistic cues cover the visual content. By decomposing inference into four stages and characterizing the interplay between linguistic priors and cross-modal coverage across these stages, we propose a systematic framework for tracing the propagation of
language bias throughout the inference process; and uncover the underlying mechanism of
language bias by revealing the interplay between linguistic priors and cross-modal coverage.
Read the full abstract ▾
All authors Yizhou Fang, Siyue Chen, Zimo Qi, Zhiyu Xue, Xi Chen, Guangliang Liu
Quantum Physics
Generalized LIMDDs: Succinctness and Canonicity for Decision Diagrams Modulo a Group
Arend-Jan Quist, Alexis de Colnet, Thomas Reps et al. (+1 more)
A reduced ordered decision diagram is the minimal automaton of a function on words of fixed length: its nodes are the residuals, and reduction is the Myhill–Nerode quotient. We study what happens when that quotient is coarsened by a group. Fix a group $G$ acting on residuals, merge two nodes when their residuals lie in one $G$-orbit, and record the group element on the edge. For functions $\{0,1\}^n\to\mathbb{C}$ and $G$ the Pauli group this is the Local Invertible Map Decision Diagram. We take $G$ from the two-parameter family generated by $C^{\leq k}R_q$, the phase rotation of order $q$ with up to $k$ control
qubits, with and without the bit flip $X$. We show that this gives exponential succinctness improvements compared to Pauli-LIMDD, and we determine the succinctness order of the family completely. The separating objects are hypergraph states, which can always be efficiently represented by some member of the family. We settle the tractability of five queries and eight transformations, which is invariant across the family, and shows the same behavior as Pauli-LIMDD. We give a five-rule reduction system whose normal forms are unique for every member of the family, and we show that this canonical form is computable in
polynomial time in the size of the LIMDD. We show that, when coarsening beyond the (anti-)diagonal groups, the calculation of a minimal sized normal form turns out to be non-local and it might to rebuild the whole diagram.
Read the full abstract ▾
All authors Arend-Jan Quist, Alexis de Colnet, Thomas Reps, Alfons Laarman
Disordered Systems and Neural Networks
Attraction-controlled torque organization and rotational states in frictional granular matter
Kiwamu Yoshii
In this study, we numerically investigate how interparticle attraction affects stress transmission, torque organization, and particle rotation in a two-dimensional frictional granular material. Although the attraction is a central force and exerts no direct torque on the particles, it strongly modifies the contact network and particle rotation. Attraction produces a low-rate stress state and selects the high-stress state near the shear-thickening regime of the dry system, while leaving the high-rate thickened state almost unchanged. By decomposing the shear stress, we find that the direct attractive contribution accounts for only a fraction of the total stress, while the repulsive contribution is strongly enhanced. Attraction increases the coordination number and torque amplitude and can reverse the local torque-sign correlation. At low packing fractions and intermediate shear rates, a stress-collapse state emerges with strongly suppressed torque and rotation despite a finite contact network. These results show that attraction controls macroscopic rheology through reorganization of the frictional contact network.
Read the full abstract ▾
Machine Learning
eval-unlearn: Benchmarking unlearning in Text-to-Image Diffusion Models
Mansi, Nikhil Raghavan, Zixia Huang et al. (+4 more)
The rising number of concept unlearning techniques for text-to-image (T2I)
diffusion models has produced a fragmented evaluation landscape. Methods are assessed under heterogeneous experimental conditions making principled cross-method comparison difficult. We present eval-unlearn, an open-source Python library providing a unified, reproducible benchmarking framework for concept unlearning in T2I
Diffusion models. eval-unlearn integrates twelve published unlearning techniques spanning fine-tuning, closed-form model editing, and inference-time intervention, alongside nine complementary evaluation metrics covering erasure efficacy,
adversarial robustness, generative quality, and concept retention. Its plugin architecture lets third-party techniques and metrics self-register without modifying the core framework, and its streaming, batched pipeline supports efficient evaluation of both standard NSFW concepts and arbitrary general concepts. As a further contribution, we release a public leaderboard on HuggingFace along with an interactive tool for real-time evaluation of unlearning techniques. The leaderboard compares nudity concept erasure case study across all twelve techniques, exposing significant accuracy-quality trade-offs that are obscured by heterogeneous evaluation. eval-unlearn is released under the MIT license; the package, code, leaderboard, and documentation are all available at https://eval-unlearn.readthedocs.io.
Read the full abstract ▾
All authors Mansi, Nikhil Raghavan, Zixia Huang, Kai Sheng Ong, Ji Shen Lim, Brandon Siao Xiang Ling, Francesco Leofante
Machine Learning
SpikeCredit: Temporal Credit Carrier for Reinforcement Learning with Sparse Rewards
Yingchao Yu, Pengfei Sun, Wenxuan Pan et al. (+4 more)
Reinforcement learning (RL) with sparse rewards is challenging because delayed outcomes provide little guidance about which intermediate computations caused success or failure. We argue that reliable credit assignment requires policy dynamics that preserve and expose credit-relevant information over time, a role we formalize as Temporal Credit Carriers (TCCs) and that spiking neural networks (SNNs) naturally fulfill through graded membrane traces and event-driven spikes. Based on this hypothesis, we propose SpikeCredit, an SNN-based framework for RL with sparse rewards that first performs task-adaptive TCC selection and then closes the loop between a fast TCC-reading pathway, where self-motion feedback constraint uses local behavior-grounded cues to constrain transition-level credit recovery, and a slow TCC-writing pathway, where credit-targeted trace
alignment feeds recovered credit back into the actor to make future TCC dynamics more credit-readable. Across sparse-reward MuJoCo tasks, SpikeCredit improves Last10 return over sparse SNN baselines by +1169% on Ant, +953% on Hopper, +723% on Swimmer, and +1781% on Walker2d, and exceeds the dense-reward baseline on Swimmer by +113%. Mechanistic analyses further show substantially stronger
alignment with dense rewards than the sparse SNN baseline. These results position spiking dynamics as credit-preserving substrates for sparse-reward RL.
Read the full abstract ▾
All authors Yingchao Yu, Pengfei Sun, Wenxuan Pan, Wei Chen, Yitian Hong, Kuangrong Hao, Yaochu Jin
Robotics
GuardPIBT: Counterfactually Gated Neural Guidance for Ultra-Large-Scale 3D Multi-Agent Path Finding
Yuan Zhou, Zhenyu Hou, Guangtong Xu et al. (+4 more)
Large-scale 3D multi-
agent path finding becomes increasingly difficult under dense traffic. Priority Inheritance with Backtracking (PIBT) scales well, but its one-step goal-directed ordering may become insufficient under dense interactions and large-scale congestion. We present GuardPIBT, which augments rather than replaces the PIBT executor: neural predictions only propose residual reorderings of PIBT's native candidates, while final actions remain determined by PIBT. First, local graph
attention models nearby interactions, while global source–goal transport features provide population-level coordination context for candidate reordering. Second, a counterfactual group gate filters reorderings whose closed-loop effects may degrade coordination. Third, for ultra-large populations, population-adaptive grouping preserves decision granularity, asynchronous cached inference amortizes neural computation, and selective repair resolves long-tail
agents. PIBT retains validity checking, priority inheritance, and backtracking throughout. Experiments with up to 100,000
agents demonstrate reliable completion across 2D and 3D environments, including all three 100,000-
agent warehouse runs with zero audited graph violations. The project website is available at {\color{magenta}https://guardpibt.github.io/GuardPIBT/}.
Read the full abstract ▾
All authors Yuan Zhou, Zhenyu Hou, Guangtong Xu, Xiaoqiang Ji, Yuqing Tang, Jialiang Hou, Fei Gao
Cryptography and Security
Continuous Assurance of Agentic Security Auditors for Software Delivery Decision Gates
Guy Lupo, Nguyen Hung Nguyen, Viet Vo et al. (+4 more)
Large
language model (LLM)-based repository auditors are increasingly deployed as security controls within continuous integration (CI) pipelines, where their findings admit, block, or delay software changes. As Agentic Software Development Life Cycle (SDLC) Security Controls, their non-deterministic behaviour changes the evidence, while organisational risk appetite and jurisdictional or data-sovereignty policy change its interpretation. Point-in-time audits therefore cannot maintain current assurance for merge decisions. We propose the Policy-Evidence-Execution Separation Pattern, implemented by the Trustworthy AI Posture (TAIP) Assurance Engine and operated as Continuous Control Posture Assurance (CCPA). By separating policy from stable execution and binding admitted evidence to a versioned Posture Tree, the same assurance logic operates across models, environments, and policy profiles. We evaluate the approach using unmodified RepoAudit on a fixed Python Null Pointer Dereference
benchmark. The retained evidence repository contains 80 RepoAudit executions across two OpenAI model configurations, gpt-4o-mini and gpt-4.1. TAIP recomputes assurance posture after policy, evidence, and model-context changes and is evaluated across increasing numbers of independent Decision Gateway contexts. The maximum observed policy-to-posture latency was 1.1 ms across three policy-class cycles in one execution. At 1,000 independent assurance contexts, full policy-triggered recomputation with one worker recorded a maximum aggregate refresh of 1.62 s, below the predeclared 5 s Decision Gateway budget. These single-host measurements concern assurance over retained evidence and exclude RepoAudit execution and provider inference.
Read the full abstract ▾
All authors Guy Lupo, Nguyen Hung Nguyen, Viet Vo, M.A.P. Chamikara, Guangdong Bai, Nazatul Haque Sultan, Alsharif Abuadbba
Quantum Physics
Gauge freedom and efficient algorithms for Lindbladian learning
Steven T Flammia, Savar D Sinha and Yu Tong
We study the problem of learning a local Lindbladian in the presence of state-preparation-and-measurement (SPAM) noise. Although recent work has developed scalable learning
algorithms under idealized access assumptions, SPAM can make distinct Lindbladians experimentally indistinguishable, thus making part of the Lindbladian fundamentally unlearnable. We give a sharp characterization of this obstruction for bounded-degree local Lindbladians. We identify a family of locality-preserving gauge transformations that commutes with trusted single-
qubit control, and use it to classify Lindbladian components. For the generically gauge dependent components, we construct Lindbladians to show that they can have $\Omega(1)$ uncertainty independent of system size, even after imposing complete positivity. We complement this characterization with SPAM-robust
algorithms for learning every universally gauge-invariant component. Our
algorithms use only trusted single-
qubit operations, require neither ancillas nor entangling control, and use $\widetilde{\mathcal O}\left(\epsilon^{-2}\log(N/\delta)\right)$ experiments and total
evolution time for constant locality and degree. Our guarantees allow global SPAM to become arbitrarily far from ideal as $N$ grows, requiring only nonvanishing local visibility. We also establish sufficient conditions for boundary-induced gauge invariance of additional
Hamiltonian coefficients under physical positivity constraints, though learning these additional parameters in general remains an open problem. More broadly, our gauge-aware framework and SPAM-cancellation techniques offer a practical toolkit for the scalable characterization of open
quantum systems.
Read the full abstract ▾
Robotics
Repair Before You Fuse: Frozen-Host Adaptation for Corrupted-but-Present Sensors
Gia-Huy Thai, Quang-Thinh Ly, Anh-Minh Phan et al. (+1 more)
Camera-
LiDAR detectors can continue to consume unreliable features even when both sensors remain present, synchronized, and calibrated. We introduce Boundary Feature Repair (BFR), a frozen-host adaptation framework that learns task-supervised residual corrections at modality interfaces the detector already consumes. BFR-C repairs each camera feature level read by fusion, whereas BFR-L aligns host-conditioned
LiDAR candidates to a selected boundary and routes site-wise innovations relative to the frozen anchor. Their jointly trained composition is BFR-CL. Zero-initialized per-channel scales make every variant an exact detector-level identity before optimization; only the repair modules train, while the encoders, fusion consumer, router, detection head, and host normalization statistics remain fixed. At inference, BFR requires neither clean references, corruption metadata, temporal history, nor online updates. Across the complete 20-corruption, five-severity KITTI-C grid, BFR-C reduces RCE from $14.07$ to $11.92$ on MVX-Net and from $14.29$ to $11.00$ on Focals Conv-F relative to their reproduced frozen baselines. On the latter host, BFR-L raises AP$_{\mathrm{cor}}$ from $73.65$ to $74.48$, while BFR-CL reaches $77.01$ AP$_{\mathrm{cor}}$ and $10.46$ RCE with $86.02$ clean AP. On nuScenes-R, BFR-CL raises the reproduced MoME baseline's mAP robustness ratio from $80.1$ to $81.4$. These results establish boundary repair as a targeted retrofit for corrupted-but-present sensing without retraining the deployed detector.
Read the full abstract ▾
All authors Gia-Huy Thai, Quang-Thinh Ly, Anh-Minh Phan, Tuan Dang
Distributed Computing
WavePP: High-Throughput Pipeline Parallel LLM Prefill under Prefix Reuse
Aaryam Sharma
Pipeline parallelism can improve prefill throughput by processing multiple request chunks concurrently across different stages of the model. However, keeping the pipeline fully utilized requires efficient scheduling and request preparation. In systems where stages retain and evict cache state independently, a local cache hit does not guarantee that the same prefix can be reused across the pipeline. Here, coordination overhead can impede request admission cadence and thus reduce overall throughput. In this paper, we present WavePP, a prefill runtime built on top of TensorRT-LLM that addresses these challenges by overlapping request admission with pipeline execution. WavePP asynchronously finds a prefix that can be reused across all stages, protects the cached state, and reserves space for the remaining input while earlier requests continue to execute. It subsequently plans the chunk sizes of each request dynamically to maximize pipeline fill. Each stage then completes the local preparation before executing the request. In the same system and pipeline
topology, WavePP improves TensorRT-LLM's prefill throughput in 37 of 40 tested settings on GLM 5.2 and MiniMax M2.7. At concurrency 128 with high cache reuse, these changes increase throughput by factors of 2.91 and 2.02, respectively. Across 28 Kimi K3 settings, WavePP also has the highest measured throughput in all 18 settings at concurrency eight or higher, compared with tensor/expert-parallel and pipeline-parallel baselines from TRT-LLM, SGLang, and vLLM.
Read the full abstract ▾
Computation and Language
Rubric-Aware On-Policy Self-Distillation for LLM Personalization
Yilun Qiu, Xiaoyan Zhao, Chengbing Wang et al. (+6 more)
LLM personalization aims to generate responses aligned with individual users' preferences and needs. User-specific rubrics make these expectations explicit, providing direct supervision on what a satisfactory answer should cover. Existing rubric-guided approaches, however, exploit such guidance only at a coarse granularity, either by using rubrics to supervise the prediction of relevant aspects for subsequent generation or by reducing aspect coverage to a single response-level reward for reinforcement learning. This leaves a gap between specifying what a personalized answer should contain and teaching the model how to generate it. To bridge this gap, we propose GRASP, a rubric-aware on-policy self-
distillation framework for LLM personalization that turns user-specific rubric aspects into fine-grained,
token-level supervision. Specifically, GRASP pairs a rubric-free student with a rubric-informed teacher that additionally receives the target user-specific rubrics. By aligning their next-
token distributions along on-policy trajectories generated by the student, GRASP transfers the teacher's rubric-conditioned guidance into the student, translating user-specific semantic requirements into dense
token-level supervision. Since rubric-informed teachers can still produce inadequate supervision, we further introduce Rubric-based Teacher Validation (RTV), which retains only instances where the teacher sufficiently covers the target aspects, improving both supervision quality and training efficiency. Experiments on the LaMP-QA
benchmark for personalized question answering demonstrate that GRASP achieves state-of-the-art performance across multiple backbones, supporting the effectiveness of rubric-guided
token-level supervision for personalization. To ensure reproducibility, our code is available at https://github.com/SnowCharmQ/GRASP.
Read the full abstract ▾
All authors Yilun Qiu, Xiaoyan Zhao, Chengbing Wang, Cilin Yan, Rui Zu, Wanyang Zhang, Xiaolong Jiang, Jiayin Cai, Yang Zhang
Artificial Intelligence
Imprint Reader: From Weight-Update Readout to Behavioral Intervention
Guanxu Chen, Qihao Lin and Jing Shao
As
language models take a growing role in AI development, a natural aspiration is for them to reflect on their own learning process, as humans do, and use that reflection to improve themselves. At the same time, these models have an advantage that human learners lack, since training leaves parameter-level traces that can, in principle, be inspected directly. However, current models cannot decode these traces into an explicit account of what they have learned. To this end, we introduce the Imprint Reader, a model trained with Semantic Mount-and-Read Tuning (SaRT) to describe frozen weight updates. SMaRT mounts each update onto the Reader and uses an anchor-free meta-query to elicit a natural-
language description, while no-change and random-perturbation controls discourage unsupported claims. On held-out updates, the joint Reader reaches judge-based Pass@100 of $2\%$ for knowledge and $16\%$ for behavior. These results demonstrate the feasibility of natural-
language readout while pointing to reliability across updates as the next step. Beyond free-form generation, the Reader provides a differentiable proxy for the gap between a specified target behavior and a candidate weight update. Its coordinate-aligned gradients support intervention through MetaEdit. At a $0.5\%$
pruning rate, Reader-guided selection raises measured harmful-
prompt refusal from $57.9\%$ to $64.1\%$ under a safety-maintenance target. Using behavior descriptions without target-task training data, MetaEdit increases the frequency of backtracking and sub-goal expressions in mathematical
reasoning traces and raises BFCL Overall from $41.69\%$ to $44.60\%$.
Read the full abstract ▾
Machine Learning
Latency and accuracy tradeoffs in Spiking Neural Networks
Zhanglu Yan, Zixuan Zhu, Kaiwen Tang et al. (+3 more)
Spiking neural networks are attractive for low-power
speech command recognition, yet their latency has received far less
attention than their energy efficiency, and their multi-timestep execution is widely assumed to make them slower than quantized neural networks. This paper challenges the assumption that more local timesteps necessarily imply higher network latency. By overlapping computation across adjacent layers at the timestep level, SNNs may complete execution in less time than comparable bit-serial QNNs. However, this overlap relies on spikes firing on incomplete inputs, and a spike once generated cannot be withdrawn, so its error persists and reduces accuracy. Waiting for more input before firing would seem to improve accuracy at the cost of reduced overlap. Yet we find and prove that this intuition fails at some layers, where even a small increase in waiting can change spike timing and downstream computation, making the network both slower and less accurate. We therefore propose a Pipeline Delay Search method which selects each layer's delay by balancing task-level accuracy gains against added network latency. We then adapt the selected configurations through spike-based quantization-aware training and bounded tuning of firing thresholds and initial membrane potentials. Together, these steps form Falcon, a framework for Fine-grained Analysis of Latency and Controlled firing which systematically analyzes and optimizes SNN latency under a spatial analog compute-in-
memory mapping with shared digital engines. We evaluate Falcon on GSCV2 and SSC, achieving competitive accuracies of 96.31 and 83.02 at modeled network-core latencies of 119.64 and 124.00us, respectively. Together, our analysis and results show that SNNs can compute more yet finish faster, and wait longer yet predict worse, highlighting why Falcon matters for both latency and accuracy.
Read the full abstract ▾
All authors Zhanglu Yan, Zixuan Zhu, Kaiwen Tang, Yuyang Cai, Qianhui Liu, Weng-Fai Wong
Machine Learning
On-Policy or Off-Policy Learning? A Systematic Study of Distillation Dynamics
Julianna Piskorz, Antonin Berthon and Mihaela van der Schaar
On-policy learning has been argued to reduce catastrophic forgetting, produce sparser parameter updates, and improve generalisation. However, existing comparisons between supervised fine-tuning and reinforcement learning vary many factors simultaneously, making the contribution of rollout policy difficult to isolate. We study the effect of rollout policy in a controlled strong-to-weak
distillation setting, by independently varying rollout policy,
token-level KL direction, and learning rate across the Llama3 and Qwen2.5 model families and
reasoning tasks spanning scientific, medical, and arithmetic domains. Our analysis reveals a nuanced picture of
distillation dynamics in which rollout policy does not necessarily play a central role. Instead,
token-level KL direction more clearly shapes task performance and output coverage, while learning rate governs forgetting and update sparsity. Analysis of KL gradients and experiments along a continuous student-teacher rollout-policy spectrum explain this pattern: forward KL is remarkably robust to rollout policy, with its performance stable and strong despite changes to the rollout policy, whereas reverse KL is substantially more sensitive and favours student-generated rollouts. On-policy data nevertheless improves generalisation to harder variants of the Countdown arithmetic task under both KL directions, although this advantage does not reliably persist after subsequent RLVR. Our broader conclusions remain robust to removing gradient clipping, using sampled KL
estimators, and training on tasks requiring longer
reasoning chains. Overall, our results challenge the view that on-policy rollouts are inherently preferable and show that their value depends critically on the objective, evaluation setting, and optimisation hyperparameters.
Read the full abstract ▾
Fluid Dynamics
A Conservative Multi-Level Adaptive Velocity-Space Method for the Discrete Unified Gas-Kinetic Scheme
Weijie Ren, Hang Yu, Zhengyu Tian et al. (+2 more)
Multiscale gas flows often span a wide range of Knudsen numbers, from near-continuum to highly rarefied and strongly nonequilibrium conditions. Such flows require kinetic solvers that remain accurate across these regimes. The discrete unified gas-kinetic scheme (DUGKS) provides a unified description, but its computational cost and
memory consumption are dominated by the size of the discrete velocity space. To alleviate this bottleneck, this work develops the multi-level velocity-space adaptive discrete unified gas-kinetic scheme (MLVS-DUGKS), reducing velocity-space redundancy while preserving accuracy and conservation. The method identifies the velocity-space requirements of individual physical
cells from their distribution functions and groups these requirements into a small number of shared representative velocity spaces. This organization retains sensitivity to different local flow states without constructing an independent velocity space for every physical
cell. A linear moment-constrained correction is incorporated to maintain discrete compatibility and conservative data transfer between representative spaces.
Benchmark simulations spanning near-continuum to highly rarefied regimes show close agreement with the uniform-velocity-space DUGKS (UVS-DUGKS). For the cylinder and two-cavity problems, the average number of discrete velocity points is reduced by approximately one order of magnitude, yielding speedups of $7.00$–$7.93$ and GPU-
memory reductions by factors of $6.73$–$7.49$. As a result, the MLVS-DUGKS provides an efficient tool for the simulation of multiscale nonequilibrium gas flows.
Read the full abstract ▾
All authors Weijie Ren, Hang Yu, Zhengyu Tian, Wenjia Xie, XiaoQiang Fan
Machine Learning
AIM-ZO: Activation-Informed Subspace Maintenance for Zeroth-Order LLM Fine-Tuning
Yue Xie, Zhi Zheng, Yunpeng Ba et al. (+5 more)
Zeroth-order (ZO) optimization offers a
memory-efficient alternative for LLM fine-tuning by estimating updates only from forward evaluations of perturbed parameters, without backpropagation or activation storage. However, in billion-parameter LLMs, isotropic perturbations often waste many forward evaluations on weakly informative directions. To make these evaluations more informative, existing ZO methods restrict perturbations to low-dimensional subspaces. Yet the quality of these subspaces is critical: overly compressed or poorly maintained spaces can miss useful update directions. To obtain a high-quality subspace for ZO updates, this paper proposes AIM-ZO, a ZO fine-tuning method based on Activation-Informed Subspace Maintenance. AIM-ZO uses forward activations as local directional information and continuously integrates them into a broad, evolving subspace over training. To access broader gradient-relevant structure while keeping individual perturbations low-dimensional, AIM-ZO activates only a smaller set of shared and sampled directions, decoupling the maintained width from the active width. We evaluate AIM-ZO across 5 LLMs and 11 downstream tasks under matched forward-evaluation budgets; its six-task average exceeds the strongest fully evaluated ZO baseline by 1.26 percentage points on OPT-2.7B and MeZO by 2.85 percentage points on OPT-30B. Our code is available at https://github.com/EkkoXy/AIM-ZO
Read the full abstract ▾
All authors Yue Xie, Zhi Zheng, Yunpeng Ba, Xuyang Wu, Xialiang Tong, Zhichao Lu, Tao Zhong, Zhenkun Wang