Computation and Language
Evaluating Language Model Safety Across Long Adversarial Conversations
Parisa Salmani and Peter R. Lewis
Conversational safety evaluations often test
language models with a single harmful
prompt, even though real-world systems interact with users through long, adaptive conversations. This study examines whether models continue to respond safely when an
adversarial user persists across multiple turns. We evaluate three open-weight, instruction-tuned models on two harmful
prompts across different conversation lengths and random seeds. In each setting, a second
language model acts as a persistent
adversarial user, while a safety classifier labels every response as safe or unsafe. Across all model-
prompt combinations, first-turn safe-response rates ranged from 85% to 100%. By depth 11, they dropped to 38-61%, and by depth 101, to 15-44%. This decline appeared across models and continued well beyond the short interactions typically used in multi-turn safety evaluations. These results provide proof-of-concept evidence that strong single-turn safety does not necessarily persist during sustained
adversarial interaction. They highlight the need for long-horizon evaluations and conversation-level safeguards that account for risk accumulating across turns.
Read the full abstract ▾
Machine Learning
Continual Learning of Dynamical Systems in Recurrent Neural Networks through Recyclable Unit Gating
Sima Hashemi, Daniel Durstewitz and Georgia Koppe
Dynamical Systems Reconstruction (DSR) aims to infer models from observed time series that reproduce a system's qualitative long-term behavior. Continual DSR (cDSR) requires learning new systems while preserving previously learned dynamics, yet even small parameter updates in recurrent models can qualitatively alter their behavior over long autonomous rollouts. We
benchmark established continual learning (CL) methods spanning parameter regularization, replay, and parameter isolation on the fully trainable and interpretable Almost-Linear RNN (AL-RNN). Parameter isolation preserves earlier dynamics most effectively, but excessive task-specific allocations can rapidly exhaust a fixed-size network. We therefore introduce Continually-Recyclable Unit-Gating (CRUG), which conserves capacity through compact allocation and forward transfer. Differentiable gates trained with an $L_0$-based penalty select task-specific units, while unused units are recycled for subsequent tasks. Directed connections allow later tasks to reuse earlier representations without affecting the dynamics of previously committed units. CRUG achieves the strongest reconstruction–capacity trade-off among the tested methods with zero forgetting and reliably learns a heterogeneous sequence of nonlinear and chaotic systems. Furthermore, we show that forward transfer is more pronounced and useful when tasks share similar underlying dynamics. Lastly, we demonstrate that CRUG's advantages extend beyond autonomous cDSR to sequential cognitive tasks.
Read the full abstract ▾
Computation and Language
Halluscoring 2026: The first shared task on llms hallucination detection and answer verification
Aisha Alansari, Abdessalam Bouchekif, Ahmed Hasanaath et al. (+9 more)
We present HalluScoring 2026, a shared task for evaluating
hallucination detection and factual verification in Arabic question answering under challenging generalization settings. Its four subtasks are organized into two tasks. Task~1 evaluates binary
hallucination detection, considering generalization to unseen questions (Subtask 1.1) and responses generated by unseen LLMs (Subtask 1.2). Task 2 extends the evaluation beyond detection by requiring the systems to additionally identify the correct factual answer from six related candidates, covering Islamic knowledge (Subtask 2.1) and general knowledge (Subtask 2.2). The shared task is based on two Arabic
datasets: HalluScore and HalluTruthQA. A total of 13 teams participated in the shared task, ten of which submitted system description papers. The results of Task 1 demonstrate that
hallucination detection remains challenging. On the Task 1 test sets, the top-ranked systems achieved AUC-ROC scores of 0.7717 for Subtask 1.1 (REGLAT) and 0.7670 for Subtask 1.2 (NAMAA). Under assisted evaluation, the highest combined detection and answer-selection scores for Subtasks 2.1 and 2.2 were 0.8824 and 0.8565, respectively.
Read the full abstract ▾
All authors Aisha Alansari, Abdessalam Bouchekif, Ahmed Hasanaath, Salah Eddine Bekhouche, Malak Alkhorasani, Mohammed-En-Nadhir Zighem, Saad Ezzini, Hichem Telli, Hend Al-Khalifa, Muhammad Abdul-Mageed, Hadid Abdenour, Hamzah Luqman
Quantum Physics
Time-Integrated Leakage as a Dynamical Benchmark for Multi-Mode Superconducting-Qubit Reset
Sushant Sharma and Aswath Babu H
Reset performance in superconducting
qubits is often summarized by the population that remains at the end of the pulse. This is an important number, but it does not describe what happens during the reset itself. A protocol may reach a small final residual while keeping the
qubit in a leakage state for a comparatively long part of the trajectory. In this work, we therefore use time-integrated leakage, together with the usual endpoint population, to study a multi-mode dissipative reset model under flux-frequency control. We examine finite thermal occupation, parameter sweeps, frequency disorder, an auxiliary-chain ablation, and a literature-based rate
benchmark. At the representative ablation point, adding the auxiliary chain reduces Lf from 8.56 to 0.78 ns and shifts the sustained Pf less-than- 10-2 crossing from 39.5 to 9.8 ns. Interestingly, the configuration without the chain still reaches the smaller long-time residual. The two metrics can therefore favor different reset configurations. For the same 300 ns analysis window, an effective rate model built from the f-ket and e-ket decay rates reported by Zhou et al. gives Lf is 108.0 ns, while the complete simulated configuration gives 0.783 ns. We use this only as a kinetic reference and not as a reproduction of the experiment. In the wider set of simulations, thermal occupation is the main source of degradation, while the frequency disorder considered here has a much weaker effect. The results suggest that time-integrated leakage is a useful quantity to report alongside endpoint residuals, especially when reset is part of repeated
quantum-error-correction cycles.
Read the full abstract ▾
Information Retrieval
TAGGRAPH: Tag-Augmented Graphs for Graph Retrieval of Agent Persistent Histories
Yu-Shu Chen, Yu-Jung Liang and Pengtao Xie
Long-term
memory lets LLM
agents recall past interactions and remain consistent across sessions, but
memory systems are hard to compare because they often vary in representation, indexing,
retrieval, and evaluation. We present a controlled evaluation framework based on shared 5W-style conversational
memories. Localized graph configurations traverse a common base graph; AdaptiveGraph adds chronological edges and Personalized PageRank
diffusion. We also evaluate BM25 over the same extracted notes and OpenClaw as a raw-input external reference.
Retrieval rankings vary across
memory settings. On LongMemEval-S, AdaptiveGraph is the strongest graph configuration at 0.844 MRR, but BM25 reaches 0.867 and OpenClaw 0.880. On ATANT Core, localized graph traversal outperforms
diffusion and BM25, whereas BM25 leads the stress rounds. Reducing LongMemEval-S within the tested range does not reproduce the ATANT
diffusion penalty, but the smallest tested store remains larger than ATANT Core, so store size cannot be ruled out. The penalty also persists under a permissive content-match criterion. Vocabulary normalization and extraction quality substantially affect graph
retrieval, and missing extraction tags are common among top-five misses.
Retrieval strategies should therefore be evaluated jointly with the
memory setting and against strong lexical baselines.
Read the full abstract ▾
Quantum Physics
Interferometric Readout of Momentum-Space Topology in a Programmable Dissipative Photonic Circuit
Andrea Cataldo, Emil J. Bergholtz, Daniel Leykam et al. (+2 more)
Topology under non-Hermitian dynamics is encoded in the phase of the bulk
evolution, yet strong dissipation suppresses the amplitudes carrying it. We resolve this using a programmable photonic integrated
circuit that implements such dynamics in synthetic momentum space through
unitary dilation, with the phase recovered by phase-shifted interferometry. For the non-Hermitian Su-Schrieffer-Heeger model, the method distinguishes trivial and non-trivial Zak phases and yields a coherence winding $q=\pm 1$ induced by an exceptional point. Extending to a synthetic
torus via a Rice-Mele pump gives the first Chern number $\mathrm{Ch}_1=0,1$ for trivial and non-trivial cycles, showing a programmable route to momentum-space
topology under strongly dissipative non-Hermitian dynamics.
Read the full abstract ▾
All authors Andrea Cataldo, Emil J. Bergholtz, Daniel Leykam, Jun Gao, Ali W. Elshaari
Fluid Dynamics
Three-Dimensional Simulation of Flood Propagation at Syabrubesi, Nepal
Rainald Löhner and Harbir Antil
Three-dimensional flood propagation through the Bhote Koshi–Langtang Khola junction at Syabrubesi, Nepal, is studied using an incompressible free-surface model. The calculation is motivated by the August 2026 flood and examines the passage of an imposed inflow through the local bends and changes in channel width. The Navier–Stokes equations and an interface transport equation are solved on unstructured tetrahedral meshes constructed from pre-event terrain data. For an inlet speed of $10\,\mathrm{m\,s^{-1}}$, the solution indicates surface speeds above $40\,\mathrm{m\,s^{-1}}$. The advancing front spreads across the junction; as the flow develops, water deepens and slows around the bend. Higher speeds persist in the confined upstream and downstream regions.
Read the full abstract ▾
Astrophysics of Galaxies
A New Hope for AGN SED Fitting: X-Ray Spectral Analysis of 12MGS AGN with the C2PO-Torus Model
Carys J. E. Gilbert, Lucia Marchetti, Luigi Barchiesi et al. (+8 more)
In this paper, we present the results of comprehensive, broadband X-ray spectral analysis of a subsample of 43 mid-Infrared (IR) selected
star-forming active
galactic nuclei (AGN). We analysed archival NuSTAR, Chandra, and XMM-Newton data, and present the first pointed X-ray observation of IRASF03450+0055. We introduce a new physically-based X-ray spectral model, C2PO-
Torus, based on a two-phase clumpy
torus geometry. It is designed to provide constraints on physical parameters of the
torus that can be directly linked to the parameters of the AGN model SKIRTOR. This model will allow us to exploit the synergies between the X-ray and IR regimes, place stronger constraints on the
torus properties, and improve spectral energy distribution (SED) fitting by providing priors for the SKIRTOR parameters. We compare the results from the C2PO-
Torus fitting to those obtained from traditional models, finding that it is an extremely effective tool for fitting a wide variety of X-ray spectra. We observe that our sample closely follows the established L_[OIII] - L_X AGN relation, while presenting an IR and
Radio excess with respect to the L_X - L_12micron and L_X - L_1.4GHz AGN relations at low luminosities. We believe that this excess is due to contamination from
star formation related emission. We anticipate that the C2PO-
Torus model will aid substantially in completing full-spectrum SED fitting of this sample, allowing us to properly disentangle the different emission components of these objects, which are local analogues of AGN at
Cosmic Noon.
Read the full abstract ▾
All authors Carys J. E. Gilbert, Lucia Marchetti, Luigi Barchiesi, Mattia Vaccari, Cristian Vignali, Stefano Marchesi, Francesca Pozzi, Francesco Salvestrini, Anna Feltre, Malebo Moloko, Carlotta Gruppioni
Machine Learning
MILO: Automated Harness Discovery via Orchestrated Multi-Agent Evolution
Prithwish Jana, Mononito Goswami, Hao Liu et al. (+9 more)
Modern agentic systems combine an AI model with a harness that controls execution and environmental interactions. Harness design strongly affects long-horizon performance, yet its combinatorial search space demands substantial human effort that must be repeated as models change. Existing automated methods explore this space narrowly, optimizing only components such as
prompts or skills or becoming trapped by fixed, exploitative search strategies. We introduce MILO (Meta-evolutionary Island Orchestration), a framework that co-evolves
agent harnesses and the strategy used to discover them. MILO combines: (i) hierarchical lineage
memory over island-based trees, using rejected mutations as negative evidence; (ii) per-island mutator
agents that rewrite complete harnesses using global search history and parent-specific feedback; and (iii) an orchestrator that adapts search through lineage grafting and speciation, mutator reassignment and curriculum revision. Across Terminal-Bench 2.1, PaperBench, and DeepSWE, MILO-discovered harnesses outperform eight state-of-the-art harnesses and six search methods using frontier (Opus 4.8) and open-weight (gpt-oss-120b) models. With Opus 4.8, MILO improves resolution over its initial harness by $+12.0\%$, $+28.3\%$, and $+10.3\%$, respectively, compared with best prior-search gains of $+4.5\%$, $+18.3\%$, and $0\%$. On Terminal-Bench 2.1, it achieves $86.1 \pm 2.0\%$, exceeding the official leaderboard's top entry ($83.8 \pm 2.3\%$) while using 26% fewer
tokens than its initial harness. On EinsteinArena open problems, MILO improves best-known upper bounds for Erdős minimum-overlap ($0.3808586 \to 0.3808568$) and the first and third autocorrelation inequalities ($1.50274365 \to 1.50274360$; $1.45081 \to 1.44889$).
Read the full abstract ▾
All authors Prithwish Jana, Mononito Goswami, Hao Liu, Xinyu Li, Langlin Huang, Zhehui Huang, Zhishen Huang, Patrick Blöbaum, Anoop Deoras, Purak Jain, Nikos Kanakaris, Sahika Genc
Machine Learning
Function-Space Transformer with Adaptive Anchors
Guorui Sang and Pedram Rooshenas
Many forms of data, including physical fields,
geometric shapes, and visual signals, are naturally described by functions over continuous domains but are observed through discrete samples. Representing these functions on fixed uniform grids imposes a trade-off between resolving localized variation and increasing computation across the domain. Neural operators address this mismatch by learning mappings between functions, while latent-
attention architectures provide flexible processing of sampled observations. We introduce the Function-Space
Transformer (FST), a framework for learning from functions through a spatially adaptive continuous latent representation. FST stores features at anchors whose locations are predicted from the input observations and recursively refines these anchor features through function-space interactions. This allows the representation to adapt its spatial organization to each input rather than inherit that of the observation grid, while supporting both spatially resolved and finite-dimensional outputs. On PDE solution prediction using PDEBench Burgers and Darcy flow, FST substantially outperforms the Perceiver IO baseline, whose latent representation lacks explicit spatial organization, and is highly competitive with the
Fourier Neural Operator. On ImageNet-1K, FST achieves higher classification accuracy than the
Vision Transformer baseline, with fewer parameters across these comparisons. Ablations further support the benefits of function-space updates and recursive refinement. Together, these results highlight the potential of adaptive continuous representations for both scientific prediction and visual recognition.
Read the full abstract ▾
Computer Vision
TrackFish3D: Self-Supervised 3D Tracking of Schooling Fish from Multi-view Videos
Patt Phurtivilai, Zhiyang Dou, Yifan Wu et al. (+5 more)
Quantifying collective fish behavior requires accurate trajectories, yet multi-view 3D tracking remains challenging due to frequent occlusions, visually similar individuals, and the long-standing scarcity of identity annotations. We present TrackFish3D, a geometry-driven self-supervised framework for dense multi-camera 3D tracking of schooling fish. Instead of relying on appearance-based re-identification or manually annotated identities, TrackFish3D turns calibrated multi-view geometry into supervision: triangulation and reprojection consistency provide pseudo-associations, while a
geometric encoder and global association
transformer learn all-to-all cross-view correspondence within each frame. To make these associations identity-aware, TrackFish3D introduces a self-supervised contrastive objective that separates co-visible individuals in the
embedding space, together with a temporal predictor that preserves identities and bridges short occlusions across frames. The resulting model is trained once on unlabeled footage and applied directly to unseen test
videos, requiring no cross-view identity labels, temporal annotations, 3D ground truth, appearance features, or test-time optimization. On our
benchmark, TrackFish3D improves 3D Multi-Object Tracking Accuracy from 87.7% for the strongest baseline to 95.8%. On the 3D-ZeF zebrafish
benchmark, it achieves 81.1% MOTA, compared with 77.4% for the best
geometric baseline. TrackFish3D also generalizes beyond fish, achieving strong results on real-world bird tracking.
Read the full abstract ▾
All authors Patt Phurtivilai, Zhiyang Dou, Yifan Wu, Kinfung Chu, Yuan Liu, Lei Yang, Wenping Wang, Taku Komura
Artificial Intelligence
Examining Variation in How Guided AI Tutors Resolve Student Impasses
Bakhtawar Ahtisham, Kirk Vanacore, Alessandra Napoli et al. (+5 more)
When a student is stuck, a tutor faces the assistance dilemma: help given too early can hinder productive struggle, while help withheld too long leaves the student in a frustrating, persistent impasse (i.e., wheel spinning). Generative AI tutors increasingly use guardrails restricting answer-giving, yet little is known about how such tutors behave once an impasse persists. We analyze 20,462 student turns from 1,260 authentic sessions with a guided LLM chemistry tutor, identifying 6,630 impasse turns of three major types: conceptual errors, expressed uncertainty, or help-seeking. We then used these impasses to simulate three tutoring conditions to study variation in AI tutor guidance through impasses: baseline, no-direct-answer, and guided tutor. For a sample of 150 impasses,
prompt specificity changed pedagogy: a baseline tutor provided the answer directly in 50.7% of responses, a no-direct-answer tutor asked a follow-up question every time, and the guided tutor responded in a wide variety of ways depending on the context. We then analyzed impasse trajectories in authentic interactions, finding that each additional impasse turn lowered the odds of next-turn recovery by 12.7% (AOR = 0.873, p < .001), and early dropouts were caught in recursive concept elicitation before reaching execution. The benefit of questioning decayed as impasses persisted (scripted question x depth AOR = 0.78; follow-up x depth AOR = 0.83), whereas addressing the student's error grew more beneficial (AOR = 1.14); after a failed scripted question, repeating it was followed by recovery in 28.1% of cases, compared with 39.8% when the tutor addressed the error instead. For learning analytics, these findings identify impasse depth and type as observable, turn-level dialogue signals that analytics can use to trigger graduated, state-sensitive assistance in real time.
Read the full abstract ▾
All authors Bakhtawar Ahtisham, Kirk Vanacore, Alessandra Napoli, Josh Arens, Ksenia Ionova, Clayton Cohn, Shima Salehi, Rene Kizilcec
Software Engineering
OpenCollab: A Multi-Agent Coding Framework with Programmable Collaboration and Controllable Runtime
Chun-Wah Hsu, Kai Gong, Yu Wu et al. (+12 more)
Multi-
agent coding systems are designed to tackle complex software engineering tasks through collaboration. However, existing evaluations typically assume configured organizations are followed faithfully, whereas reality differs. This behavioral gap, combined with differences in underlying system components, prevents clear attribution of observed gains. To this end, we introduce OpenCollab, a multi-
agent coding framework that provides a unified infrastructure for programmable collaboration and controllable runtime. Specifically, OpenCollab unifies organization design, enforces experimental control on a shared runtime, and tracks execution through fine-grained event streams. On this basis, we define Adherence to quantify whether the declared organization is actually realized. Our experiments reveal that
agents collaborate very differently across configurations: changing any single dimension shifts Adherence, from 47.2% to as high as 97.2%. Furthermore, extensive agentic coding
benchmarks show that OpenCollab (Duo) establishes new SOTA performance compared to the mainstream harnesses such as Mini-SWE-
agent, OpenHands, Codex CLI, and Claude Code, showing that a well-designed organization can outperform strong existing harnesses, while OpenCollab's single-
agent configuration uses the fewest
tokens across all evaluated suites. OpenCollab establishes a unified multi-
agent infrastructure for easy programmable collaboration and controlled causal evaluation.
Read the full abstract ▾
All authors Chun-Wah Hsu, Kai Gong, Yu Wu, Xianhe Chen, Mengyang Liu, Jie Li, Hanyu Li, Zhixuan Liu, Naisheng Tang, Jiaying Chi, Ziheng Fan, Xuning He, Xiaokang Yang, Xue Jiang, Yihong Dong
High Energy Astrophysical Phenomena
VLBI Astrometric Procedure and Performance with CHIME/FRB Outriggers
Shion Andrew, Mattias Lazda, Calvin Leung et al. (+15 more)
Precise localizations of fast
radio bursts (FRBs) to specific environments within their host
galaxies with very-long-baseline interferometry (VLBI) have provided valuable insights into the nature of their progenitors. To date, such localizations have relied primarily on traditional VLBI facilities, whose observational capabilities limit them to targeted follow-up on a small subset of repeating FRBs that are observable at frequencies above $\sim$1 GHz. CHIME/FRB Outriggers is a low-frequency (400–800 MHz), wide-field VLBI network designed to overcome these limitations, aiming to localize a substantial fraction of CHIME-detected FRBs to $\sim$50 mas precision. In this work, we detail the analysis procedure used to localize FRBs with the CHIME/FRB Outriggers VLBI array. With over a thousand test localizations consisting of continuum calibrators and
pulsars with known VLBI positions, we provide a comprehensive demonstration of the VLBI array's ability to robustly meet, and under some conditions exceed, the nominal astrometric specification required for the central science goals of CHIME/FRB Outriggers. By demonstrating that a wide-field survey instrument can deliver routine $\sim$50 mas astrometry on one-off FRBs, this work lays the foundation for the upcoming generation of wide-field VLBI FRB surveys and the population-scale FRB samples they will enable.
Read the full abstract ▾
All authors Shion Andrew, Mattias Lazda, Calvin Leung, Kiyoshi W. Masui, Danté M. Hewitt, Jacob Willis, Alyssa Atkinson, Kevin Bandura, Shami Chatterjee, Nina V. Gusinskaia, Adam E. Lanman, Juan Mena-Parra, Kenzie Nimmo, Aaron B. Pearlman, Ziggy Pleunis, Alexander W. Pollak, Gurman Sachdeva, Vishwangi Shah
Computer Vision
Learning Semantic Inpainting for Animatable Gaussian Head Avatars
Pilseo Park, Fizza Rubab and Yiying Tong
We present SInGA, a novel method for learning Semantic Inpainting for animatable Gaussian head Avatars from a single image. Existing avatar approaches often rely on multi-view observations and lack effective handling of unobserved regions in single-view settings, limiting their applicability in such scenarios. To address this, we propose a semantic inpainting framework defined in UV space for completing unobserved facial regions. Our key insight lies in the structured
topology of the UV representation, which provides consistent spatial correspondences and enables reliable completion of identity-specific features using the inherent
symmetry cues of human faces. We extract features from observed regions and use them to complete unobserved regions. The completed representation is then used to regress Gaussian attributes, effectively performing Gaussian inpainting. In addition, instead of relying on a single Gaussian at each surface or pixel location, we stack multiple Gaussians to enhance detail. The resulting avatar generalizes across identities without requiring per-identity optimization and can be animated with driving inputs. Experimental results show that our method generates high-quality head avatars with improved completeness and identity preservation, while supporting realistic animation and consistent rendering from unobserved views.
Read the full abstract ▾
Machine Learning
Activation-Conditioned Self-Distillation
Zhexi Lu, Subhajit Chaudhury, Tejaswini Pedapati et al. (+2 more)
On-policy self-
distillation uses a model as its own teacher to provide dense supervision for
reasoning, often through reference-solution conditioning. Providing privileged information does not by itself ensure effective
token-level supervision throughout long responses. We introduce Activation-Conditioned Self-
Distillation (ACSD), which extracts a steering vector by contrasting activations of self-generated trajectories that reach verified correct answers within a generation budget with those of all remaining trajectories. A frozen copy of the base model applies this vector at each prediction position, and the student learns from its next-
token distributions on student-generated prefixes. Outcome verification is used for direction construction and calibration;
distillation requires neither problem-specific reference text nor teacher parameter updates. The distilled student is used alone at inference. On each of five models, ACSD achieves the highest mean accuracy over four mathematical
benchmarks among the evaluated methods. On DeepSeek-R1-0528-Qwen3-8B, mean mathematical accuracy reaches 71.9% and LiveCodeBench v6 pass@12 reaches 70.9%, compared with 69.0% and 66.3% for the reference-conditioned OPSD baseline. Contrasts among correct trajectories also support
distillation, and extracted directions can be reused across mathematical training
datasets. On fixed student trajectories, ACSD maintains more stable late-position logit-update magnitudes than OPSD.
Read the full abstract ▾
All authors Zhexi Lu, Subhajit Chaudhury, Tejaswini Pedapati, Keerthiram Murugesan, Lei Yu
High Energy Astrophysical Phenomena
HESS J1507-622: A Plausible Young Galactic Kilonova Remnant
Prantik Sarmah, Xilu Wang, Myles Carley et al. (+3 more)
HESS J1507$-$622 is an unusual TeV gamma-ray source in the H.E.S.S.
Galactic Plane Survey, distinguished by its off-plane location ($b \simeq -3.5^{\circ}$) and compact angular extent ($\sim 0.3^{\circ}$). Its location and morphology challenge conventional
pulsar wind nebula and
supernova remnant interpretations, and the absence of a detected shell or associated
pulsar leaves its origin unresolved. Here, we propose that HESS J1507$-$622 is a young
Galactic kilonova remnant with gamma-ray emission of leptonic origin. In this scenario, the source's off-plane location is naturally explained by the natal kick imparted to the progenitor binary
neutron star system. We demonstrate that a leptonic kilonova remnant model can successfully reproduce the observed GeV-TeV flux via inverse-Compton scattering from shock-accelerated electrons, while the corresponding synchrotron emission remains below current X-ray and
radio upper limits. Based on the source's energetics and angular size, we infer a distance of 3.8-14.3 kpc and an age of 0.2-3.0 kyr. Importantly, our analysis favors a kilonova remnant younger than 1 kyr. The lack of a historical naked-eye record can be explained by the transient's rapid fading, far-southern declination, and potential line-of-sight extinction. We further assess the detectability of unique multi-wavelength signatures that could confirm this hypothesis. We find promising prospects for detecting the unique MeV gamma rays from the decay of $r$-process radioisotopes using future observatories, alongside complementary low-energy signatures from thermal
dust emission and a possible optical light echo. These distinct signals offer definitive tests of the kilonova remnant interpretation and motivate targeted observations with current and future facilities.
Read the full abstract ▾
All authors Prantik Sarmah, Xilu Wang, Myles Carley, Shu-Xu Yi, Ming-Yu Ge, Rebecca Surman
Artificial Intelligence
CARAT: Do Materials LLMs Reason or Recite?
Jiajun Wu, Jian Yang, Zixiang Ni et al. (+2 more)
When a materials LLM answers a question about
crystal structure, does it reason from the structure or copy an answer already printed in its input? Accuracy cannot tell: a structural description often prints the very field it is scored against. CARAT holds question and gold answer fixed across eight matched views, names each structural relation separately in GraphSpace, and adds matched fine-tuning, answer masking, evidence injection, paired inference, and a rule that can withhold claims. First, on the
benchmark's hardest families the grounded view is worth 17.3 points over formula inputs. Second, we turn that scrutiny on ourselves. GraphSpace beats a plain periodic graph by 19.3 points, but that margin is two effects at once: where the plain rendering carries everything the question needs it is 1.96 points, and where it omits those fields entirely, 46.7 points. The headline mostly measures what the baseline lacked, not how evidence is presented. Third, we attack our own
benchmark. A rule that skips the link and reads the list directly answers four of seven hardened families, so we rebuilt it until eleven such shortcuts sat near chance. The frozen model quotes that link yet answers the same when we redirect it, on 95.6% of paired cases: it repeats the relation without using it. After matched supervision it reaches 99.8%, and deleting the link drops it to 23.4%, below the 27.0% the best shortcut reaches: both steps are learnable.
Read the full abstract ▾
All authors Jiajun Wu, Jian Yang, Zixiang Ni, Zhenzhu Li, Bin Chong
Cryptography and Security
Aegis: Generative Gradient Masking for Privacy-Preserving Medical Federated Learning
Chaoyu Zhang, Shanghao Shi, Heng Jin et al. (+3 more)
Federated learning (FL) has become a foundational paradigm for multi-institutional medical AI, allowing hospitals and research centers to jointly train diagnostic models without exchanging patient records. This
privacy promise, however, is increasingly contested: a malicious or honest-but-curious server can launch model inversion attacks (MIAs) that reconstruct private patient images directly from shared model updates, and recent scalable, closed-form attacks penetrate even secure aggregation at clinically realistic batch sizes. Existing defenses face an unsatisfactory dilemma. Gradient-perturbation methods such as differential
privacy and
pruning trade away the diagnostic accuracy on which clinical reliability depends, while cryptographic protocols add system complexity yet still leave updates exposed to these scalable attacks. We propose Aegis, a principled client-side defense that breaks this dilemma without perturbing patient data or modifying the FL protocol. Our key insight is that the success of every known MIA is fundamentally bounded by the local batch size relative to the model's leakage capacity; once this limit is exceeded, distinct samples collide and reconstructions collapse into indistinguishable mixtures. Aegis turns this universal bottleneck into a defense: each client superimposes onto its real update a masking gradient computed on locally synthesized, task-relevant data, deliberately pushing the effective batch beyond the attack's recovery capacity. We complement the design with theoretical convergence guarantees under standard convex assumptions and evaluate Aegis on MNIST, CIFAR-10, and three MedMNIST modalities (chest X-ray, abdominal CT, colon pathology). Aegis neutralizes three state-of-the-art MIAs while preserving model utility and incurring only modest overhead, offering a practical
privacy primitive for medical FL.
Read the full abstract ▾
All authors Chaoyu Zhang, Shanghao Shi, Heng Jin, Ning Wang, Y. Thomas Hou, Wenjing Lou
Nuclear Theory
Neutron-nucleus scattering in an $L^2$-integrable basis
Bui Minh Loc and Calvin W. Johnson
First-principles calculations of scattering off
quantum many-body systems such as atoms and
nuclei is important but nontrivial, requiring large-dimension ($10^3$ to $10^8$ or more) bases. For the first time we apply to many-body targets a recently developed approach to find phase shifts in $L^2$-integrable bases, computing
neutron-
nucleus scattering using the same interactions as ab initio models of
nuclei, in very small (dimension 3-6) bases. Our work shows the mean field dominates low-energy
nuclear scattering; it also provides a foundation for more efficient future methodologies.
Read the full abstract ▾
High Energy Physics (Phenomenology)
A Partial Wave Formalism for High-Energy Meson Electroproduction
Dillon Leahy, Derek Glazier, Vincent Mathieu et al. (+7 more)
We present a comprehensive partial-wave analysis formalism for
meson electroproduction based on reflectivity amplitudes. By applying parity-based
symmetries, we construct a fully diagonalized framework that relates complex production amplitudes to measurable spherical harmonic moments and
Spin Density
Matrix Elements. We demonstrate that, under the assumption of high-energy Regge factorization, longitudinal and transverse cross sections can be natively separated without requiring beam-energy-varying Rosenbluth separations. We validate this inversion methodology through numerical closure tests and apply it to existing unpolarized $\rho^0$ and $\omega$ electroproduction data from HERMES and COMPASS. The extracted amplitudes align with $s$-channel helicity conservation expectations for the $\rho^0$, while clearly isolating the large unnatural-parity pion exchange contribution in $\omega$ production without model-dependent background approximations. Finally, we systematically generalize the formalism to include initial target and
recoil baryon polarization, establishing the rigorous observables required for complete amplitude extraction, no longer requiring any factorisation assumptions. This unified framework directly supports upcoming spectroscopy and nucleon structure programs at CLAS12, GlueX, and the future Electron-Ion
Collider.
Read the full abstract ▾
All authors Dillon Leahy, Derek Glazier, Vincent Mathieu, Marco Filippini, David Ireland, Glòria Montaña, Alessandro Pilloni, Bhawani Singh, Charlie Velasquez, Daniel Winney
High Energy Physics (Theory)
Scaling similarity and generalised conformal symmetry in D2-brane holography
Andrea Conti, Adolfo Guarino and Ricardo Stuardo
We study D2-brane holography within the purely dilatonic sector of four-dimensional maximal ISO(7)-gauged supergravity, obtained as a consistent truncation of type IIA supergravity on S$^6$. In the dual frame, the four-dimensional Lagrangian resulting from the reduction of the massless IIA theory exhibits scaling similarity, thus reflecting the generalised conformal
symmetry of the dual three-dimensional maximally supersymmetric Yang-Mills theory. Using an auxiliary spacetime construction, we show that the
holographic CFT$_3$ fixed points and RG-flows known in the massive IIA theory have natural counterparts in the massless IIA theory: QFT$_3$s with generalised conformal
symmetry connected by generalised RG-flows, which we construct numerically. We also revisit the Coulomb branch solutions and propose an extension of Gubser's singularity criterion to supergravity theories exhibiting scaling similarity.
Read the full abstract ▾
Software Engineering
E2E-SWE: Benchmarking LLMs on Building Working Codebases from Scratch
Hantian Ding, Chloe Bi, Jiacheng Zhu et al. (+5 more)
Coding
agents powered by large
language models (LLMs) are evolving from making localized code changes to developing complete software repositories. However, evaluating repository-scale generation remains challenging: tasks must demand system-level
reasoning while ensuring that all evaluated behaviors are precisely specified and independent of any particular implementation. We introduce E2E-SWE, a
benchmark for evaluating whether coding
agents can build complete, functional software repositories end to end. E2E-SWE contains 186 whole-repository generation tasks spanning 11 programming
languages. Given only a natural-
language specification and an empty workspace, an
agent must implement a complete, installable project that satisfies a comprehensive suite of hidden tests. Each task is constructed by a software engineer in collaboration with an LLM; together, they develop the test suite and a corresponding implementation-independent specification. To ensure that tasks are well specified and practically solvable, we further subject them to an iterative verification process in which autonomous
agents audit and repair task defects using static inspection and failures observed from real model rollouts. Evaluating 13 frontier models, we find substantial variation in end-to-end repository generation ability, with pass@1 ranging from 11.7% to 67.7%, providing strong model differentiation while leaving considerable headroom for future progress. Analysis of
agent trajectories further reveals long, front-loaded
reasoning patterns, highlighting the planning and system-level
reasoning required to construct working codebases from scratch.
Read the full abstract ▾
All authors Hantian Ding, Chloe Bi, Jiacheng Zhu, John Yang, Matt Deitke, Pengcheng Yin, Zijian Wang, Rui Hou
Computation and Language
EVOKE: Eliciting World Knowledge in Agents for Transferable Decision-Making
Yuhan Guo, Jinming Liu, Liang Xu et al. (+8 more)
Large
language models (LLMs) are increasingly deployed as
agents for multi-step decision-making, yet transfer poorly to unseen environments. World-model methods address this by training
agents to predict future observations, at the cost of additional training and errors that compound when predictions are used for planning. However, for LLM
agents operating in digital environments, much of this world knowledge is already internalized during pretraining, which shifts the problem from acquiring it to eliciting it. We argue that typical post-training provides little pressure for such elicitation, since supervision under a single goal at each visited state inadvertently drives policies to rely on superficial contextual habits. We introduce EVOKE, a post-training method that supplies this pressure through goal diversity at fixed states. Motivated by theory showing that an
agent competent across diverse goals must encode a world model recoverable from its action preferences, EVOKE holds the environment state and interaction history fixed and ranks the same candidate actions under alternative goals, forcing action preferences to change, so that a policy relying on contextual habits or single-goal correlations cannot order them correctly. This implicitly elicits the policy's pretrained world knowledge to inform decisions. We evaluate EVOKE across diverse tasks in three backbones, demonstrating improved task performance, unseen environment generalization, and data efficiency. We further conduct controlled analyses to better understand what drives these gains. These findings offer a new perspective on eliciting internalized world knowledge for transferable action through direct decision supervision.
Read the full abstract ▾
All authors Yuhan Guo, Jinming Liu, Liang Xu, Ziqiang Li, Jianguo Huang, Zhicheng Wang, Hu Zhu, Qiuyu Chen, Yuntao Wei, Xin Jin, Wenjun Zeng
Quantum Physics
Collective Dynamics in Spin Chains with Constrained Dissipation: Classically Fragile but Quantum Robust
Pietro Brighi, Gianluca Frazzei, Igor Lesanovsky et al. (+2 more)
Collective non-
equilibrium phenomena in many-body systems are strongly influenced by fluctuations, particularly in the vicinity of phase transitions, where the interplay between classical and
quantum effects can alter cooperative behavior. Here, we investigate this classical-
quantum competition in a kinetically constrained
spin chain, the dissipative far-East model. In absence of fluctuations, the system exhibits emergent collective dynamics, resulting in phase coexistence. We find that this phenomenon is robust to
quantum fluctuations, which further stabilize it, whereas classical fluctuations suppress phase coexistence at long times. We map out the steady-state phase diagram using a cluster mean-field approach and characterize the dynamics with
stochastic tensor-network methods. Within the bistable phase we identify signatures of dynamical heterogeneity in both the local magnetization and the
entanglement dynamics. This highlights how competing
quantum fluctuations and classical kinetic constraints shape the relaxation dynamics and the phase structure of many-body systems.
Read the full abstract ▾
All authors Pietro Brighi, Gianluca Frazzei, Igor Lesanovsky, Juan P. Garrahan, Alberto Biella
Machine Learning
Hermes: Learning Contextual Reasoning Unlocks Test-Time Scaling
Xinyu Li, Mononito Goswami, Hao Liu et al. (+5 more)
Test-time scaling improves model performance by allocating additional compute during inference. Using this compute effectively across multiple context windows requires deciding how to allocate fresh contexts and what information to carry between them. We call a model's ability to make these decisions contextual
reasoning. Existing approaches largely prescribe these decisions through their harness; we instead shift them to the model. We introduce 1) Hermes, a family of simple, configurable harnesses that progressively varies model control over context allocation and reuse, and 2) Hermes-Learn, a two-stage framework for learning these capabilities. We find that capable models can exploit this flexibility to scale with additional inference-time compute, while smaller open-source models initially struggle to do so. Training with Hermes-Learn closes this gap, inducing adaptive contextual
reasoning strategies that vary with both the problem and the progress of
reasoning. These gains generalize across
benchmarks and models, extrapolate beyond the inference-time compute seen during training, and transfer to complementary test-time scaling methods beyond Hermes.
Read the full abstract ▾
All authors Xinyu Li, Mononito Goswami, Hao Liu, Nikos Kanakaris, Langlin Huang, Prithwish Jana, Patrick Blöbaum, Purak Jain
General Relativity and Quantum Cosmology
Towards solving General Relativity with Physics-Informed Neural Networks
Juan A. Carretero, Jorge F. Urbán, Fernando Abalos et al. (+4 more)
Physics-Informed Neural Networks (PINNs) are a machine-learning framework for approximating solutions to systems of partial differential equations by constraining neural networks to satisfy the underlying physical laws. The resulting continuous representation does not require a predefined computational mesh and can be evaluated at arbitrary points within the training domain. In this work, we investigate the application of PINNs to Einstein's equations of General Relativity. We discuss the mathematical formulation, network architectures, loss functions, and training strategies used to obtain accurate spacetime solutions, and demonstrate the approach through a series of standard
benchmarks from Numerical Relativity. We then extend the method to a more astrophysical setting by considering the dynamical
evolution of an isolated solitonic
boson star. Our simulations show that a PINN operating directly in three-dimensional Cartesian coordinates can reproduce the
equilibrium structure of a highly compact
boson star and recover its characteristic radial oscillation frequencies. These results establish PINNs as a viable complementary framework for solving coupled matter-Einstein equations and exploring alternative formulations in numerical relativity.
Read the full abstract ▾
All authors Juan A. Carretero, Jorge F. Urbán, Fernando Abalos, Joan Massó, Oscar A. Reula, Jose A. Pons, Carlos Palenzuela
High Energy Astrophysical Phenomena
Cosmic-neighbor-associated distances to VHE and candidate VHE BL Lacs via SDSS photometric redshifts
Karri I. I. Koljonen and Elina Lindfors
BL Lac objects have distances that are difficult to estimate because their optical spectra rarely show identifiable emission or absorption lines. We present a method to constrain BL Lac distances by associating them with their
cosmic neighborhoods (local overdensities in the
cosmic web) identified in optical fields. Distances to
galaxies in these neighborhoods are estimated using the Sloan Digital Sky Survey (SDSS) photometric
redshifts. Using
cosmic-neighborhood associations previously established from multi-object spectroscopy and BL Lacs with spectroscopically measured
redshifts as control samples, we find that the
redshifts of the associated neighborhoods can be recovered with an accuracy of $\Delta z \sim 0.02-0.04$. When multiple candidate neighborhoods are present in the field, we rank the candidate associations using a central-location criterion. In 50% of cases, the BL Lac is associated with the correct neighborhood. We provide distance estimates for BL Lac objects without secure
redshift measurements, focusing on sources detected at very high energies (VHE; $E>100$,GeV) and on candidate VHE emitters. Combining intervening absorption systems, host-
galaxy imaging, $\gamma$-ray constraints, and our SDSS group-association method, we propose likely
redshifts for TXS 0141$+$268 ($z\sim0.93-0.95$), PKS 0735$+$178 ($z=0.64$), 3FHL J0905.5$+$1357 ($z=0.644$), B2 0912$+$29 ($z = 0.53$), 3FHL J1120.8$+$4212 ($z=0.363$), BZB J1243$+$3627 ($z=0.485$), 3FHL J1253.1$+$5300 ($z=0.664$), and RX J2156.0$+$1818 ($z=0.635$). Finally, we discuss how forthcoming narrow-band photometric-
redshift surveys can improve blazar distance estimates via group-association techniques.
Read the full abstract ▾
Computer Vision
ExploreNet: Learning Where to Explore in Diffusion GRPO
Shuyue Stella Li, Xiaochuang Han, Yulia Tsvetkov et al. (+1 more)
Group-relative RL methods such as Flow-GRPO post-train image generators by exploring with isotropic Gaussian noise added at every denoising step. This noise decides which rollouts the model learns from, yet it perturbs every channel and spatial position of the latent equally. In this paper, we instead show that latent elements differ in how much they change the generated image, so exploration should adapt to these differences. We introduce EXPLORENET to learn an adaptive exploration distribution. EXPLORENET is a policy that predicts a noise scale for every latent element from the current latent, the denoising step, and the
prompt, before any reward is observed; it is trained on the reward spread of each rollout group and discarded after training, leaving inference unchanged. On Stable
Diffusion 3.5 Medium, EXPLORENET improves held-out GenEval2 by 14% over Flow-GRPO, transfers to two independent compositional
benchmarks and five preference and image-quality models, and reaches a 67.2% human preference win-rate. Overall, across our group-relative
diffusion RL experiments, we find that exploration is learnable, the shape of the exploration distribution outweighs its magnitude, and rollout quality is more effective than rollout quantity.
Read the full abstract ▾
All authors Shuyue Stella Li, Xiaochuang Han, Yulia Tsvetkov, Luke Zettlemoyer
Quantum Physics
Exponential strong converse for blind quantum data compression
Kohdai Kuroiwa
We establish exponential strong converse
theorems for blind
quantum data compression of finite-dimensional mixed-state sources, with and without
entanglement assistance. Blind compression is a fundamental
quantum information processing task in which the sender aims to transmit
quantum states without direct access to the classical label. Strong converse is a notable aspect of data compression: the accuracy of the compression abruptly drops down to zero once the compression rate goes below a threshold. Strong converses for data compression have been established for pure-state sources; however, for mixed-state sources, the optimal compression rate is determined by a structural decomposition of
quantum states and can change substantially under small perturbations. This sensitivity makes it difficult to quantitatively analyze the trade-off between compression rate and error. Despite recent progress, whether a strong converse holds at the optimal rate has remained unresolved in general. In this work, we resolve this question affirmatively. For unassisted compression, we show that at every rate below the optimum, the accuracy decays exponentially with the block length for general mixed-state sources without restrictions on protocols. We also prove exponential strong converse for blind compression with unlimited
entanglement assistance at its optimal rate. Combining the two strong converse bounds yields exponential decay at every compression-
entanglement rate pair outside the complete achievable rate region. Our proof introduces a new overlap quantity based on the structure of
quantum states; this quantity connects preservation of the source information to the dimension of the transmitted system. We expect that this approach provides useful tools for converse analyses of other
quantum information-processing tasks involving mixed states.
Read the full abstract ▾