Papers

AI reliability concerns whether an AI system performs its intended function dependably over a stated period and under stated operating conditions, with clear evidence. As these systems become more autonomous, this function should include more than only a correct output. Retrieval, memory, tool use, permissions, human oversight, and interactions among systems must operate consistently and safely, and, for a generative AI, so must the reasoning process that produces the output. Average benchmark accuracy only measures system capability; it does not quantify this broader reliability claim. This paper adapts established reliability engineering methods, from failure definitions and operational envelopes to FMEA, accelerated testing, field monitoring, and reliability growth, to AI systems. A four-level diagnostic framework classifies failures as component, operational-loop, agentic-conduct, or network and governance failures. Test, evaluation, verification, and validation (TEVV), sequential monitoring, and FRACAS create and refresh evidence. The SMART framework provides statistical guidance for measurement, analysis, assessment, and test planning, and the NIST's AI risk management framework provides organizational guidance for governance, evaluation, monitoring, and mitigation. Three cases are discussed to illustrate these programs. It is shown that while the established reliability engineering principles can still provide a usable foundation, new measurements and safety guardrails are urgently needed as these systems are self-evolving at a fast pace.

Large Language Model (LLM)-based agents promise to automate scientific discovery, yet exploring the vast hypothesis space remains costly. Existing principle-evolution methods accelerate this loop, but operate sequentially, which caps exploration breadth and wastes wall-clock time on challenging problems. To address this, we formulate collaborative scientific discovery as evidence transfer between parallel principle-evolution branches. We present COEVOLVE, which realizes this transfer through a coordination core over parallel branches. By integrating value-of-information-gated routing and context-discounted likelihood injection, COEVOLVE enables branches to collaborate through shared measurements while keeping their principle posteriors separate. Across six scientific-discovery tasks under a matched evaluation budget, COEVOLVE attains a mean solution quality of 66.5% versus 57.0% for single-branch principle evolution, with a 1.80x mean wall-clock speedup on the GPT-5.6-Terra backbone; on five auto-research tasks delegated to an autonomous research harness, it is the only arm whose mean stays above the published SOTA anchor on every task. These results establish when evidence sharing accelerates parallel discovery and when transfer safeguards are necessary to limit negative or inert transfers

Neural residuals can improve satellite evapotranspiration (ET) estimates, but selectors must predict when a correction helps and reject unsupported inputs. We evaluate ten-member models on 16,366 flux-tower observations from 151 stations paired with OpenET, across nine rolling years and five spatial folds. At one held-out station, Gain accepted corrections on all 32 physically invalid records: it predicted a mean benefit of 0.83 mm/day, but the corrections increased mean absolute error by 21.6 mm/day versus OpenET. On spatially held-out unit errors, SupportGain reduced station-macro MAE versus Gain by 0.148 mm/day under wind x3.6 (simultaneous 95% interval, 0.070 to 0.226), with 9.3% acceptance versus Gain's 51.8%; on clean inputs, its 0.006 mm/day advantage had an interval that includes zero. These fault analyses are exploratory; none of 40 preplanned temporal comparisons passed Holm correction, while a separate predeclared cropland contrast found 0.041 mm/day lower station-macro MAE with crop-only training (95% interval, 0.009 to 0.079).

We deform the complex structure of compact simply-connected Bismut flat manifolds, obtaining new complex manifolds that admit pluriclosed metrics but do not admit Bismut flat metrics. This produces genuinely new examples of pluriclosed manifolds while enriching our understanding of the deformation theory of pluriclosed metrics. Moreover, our main result is relevant for the search of pluriclosed solitons and non-trivial Bismut Hermitian Einstein metrics in all dimensions.

Large Language Models (LLMs) perform well on reasoning benchmarks, but it remains unclear whether this reflects genuine contextual reasoning or reliance on facts memorized in their parameters. We investigate this by distinguishing two possibilities: a broad memorization bias, where familiar content improves reasoning performance, and the Strong Parametric Shortcut Hypothesis, where models skip reasoning entirely and recall stored answers. To test these effects, we introduce MemoReason, a human-curated benchmark that pairs factual reasoning tasks with structurally identical \fictitiousterm{} versions where real entities like people, companies, or dates are systematically replaced by \fictitiousterm{} ones of the same type. This \scorerevision{preserves task structure and specified reasoning operations} while varying the familiarity of the context, allowing controlled measurement of how the parametric memory affects reasoning. \revision{Our evaluation of recent LLMs reveals consistent and statistically significant performance drops of up to 15.7% in the fictitious setting, demonstrating a clear memorization bias.} However, a targeted analysis of \revision{questions failed in the fictitious setting} shows that models rarely respond with the corresponding factual answer, indicating that direct parametric shortcuts are not the dominant failure mode. These findings suggest that parametric memory influences reasoning through mechanisms more complex than simple factual recall. MemoReason provides a controlled framework for studying these mechanisms and for extending paired factual-fictitious{} evaluation to broader reasoning settings.

Action-conditioned video world models predict future robot interactions from multiple cameras, yet their outputs remain disparate video collections rather than a shared metric scene queryable across viewpoints and time. While existing 4D reconstruction methods offer a path to spatialize these predictions, independently reconstructing and merging each camera stream fails to enforce cross-view consistency. This limitation is particularly detrimental when combining moving robot-mounted cameras with fixed external views. To address this, we introduce RoGSW4RLD, a feed-forward framework that lifts synchronized multi-camera rollouts into a unified, time-queryable metric 4D Gaussian field. Rather than learning a separate geometric transition model, RoGSW4RLD directly reconstructs the visual future generated by existing world models. Its core innovation is a two-stage architecture: Stage 1 jointly forms the metric 4D field by fusing cross-view evidence with robot-specific articulated geometry and kinematics, while Stage 2 refines the field's geometry and appearance while strictly preserving the initial temporal displacements. Evaluated on 256 held-out DROID episodes, RoGSW4RLD significantly outperforms camera-wise reconstruction with calibrated merging, improving novel-view PSNR by 2.15 dB, reducing depth AbsRel by 47%, and lowering robot displacement error by 61%. These robust gains extend to action-conditioned Cosmos 3 rollouts, demonstrating that predicted video futures can be successfully translated into consistent, spatially queryable 4D metric representations.

NATS JetStream's at-least-once delivery guarantee is conditional: five common configuration mistakes silently violate it, causing duplicate message processing, data loss, or redelivery storms with no error logged anywhere. The standard Prometheus NATS exporter exposes only server-level throughput metrics and cannot detect any of these failures. We present nats-lens, a standalone monitor that reads from the JetStream management API and detects all five violation classes without requiring changes to monitored applications or client code. We formally characterize each class with a precise condition, prove that standard Prometheus NATS metrics are structurally incapable of detecting any of them, and implement five targeted detectors. In a controlled evaluation of 30 rounds per scenario on both single-node and 3-node JetStream clusters, nats-lens achieves 100% detection coverage across all five classes—versus 0% for the baseline—with zero false positives over 30 minutes of healthy operation. Detection latency ranges from 2,003 ms to 8,013 ms (within three poll cycles). We confirm language-agnostic detection empirically using consumers in Rust, Go, and Python. The tool is open source and exposes findings through four output channels: web dashboard, Prometheus metrics, REST API, and NATS health events.

Physical infrastructures designed for stable climate conditions support a significant portion of daily human life activities. However, climate change reshapes the distributions of extreme climate intensities. As a result, physical infrastructures that may face extreme events exceeding their design thresholds risk damage and require adaptation. A preliminary step to adaptation is to identify exposed infrastructures to climate hazards. This paper aims to structure the exposure modeling to climate change hazards by providing an overall methodology. This paper instantiates a framework using the Boolean Exposure Model to assess how physical systems may be exposed to climate hazards. The framework combines a Boolean System Model for infrastructures and a Boolean Hazard Model informed by extreme value statistics. An illustrative case study quantifies the exposure of road bridges in French administrative units to extreme temperatures. Results indicate increasing exposure across future climate scenarios and time periods, with spatial non-uniformity. This framework supports multi-level analysis and integrates non-stationary climate extremes, providing a structured exposure model to facilitate the adaptation of physical systems to climate change.

Large language models treat conversation history as unverified context, so false premises injected into earlier turns can be adopted as fact later, a failure mode we call session-level contamination. We introduce five contamination protocols along a source-authority gradient, each holding the false premise constant while varying how it is presented, from self-attributed falsehood to explicit instruction override. We evaluated three architecturally diverse models, GPT-5.4 Mini, Gemini-3.1 Flash-Lite, and GLM-4.5-Air, on ten cases spanning five knowledge fields at temperature zero over 22,500 turns, using a dual-track judge checked against a 120-turn human-annotated sample (Cohen's kappa = 1.000 for binary adoption, a ceiling on this sample, and 0.92 linear-weighted for collapse severity). The three models implemented structurally different epistemic policies. GPT-5.4 Mini recorded zero adoptions across all 500 sessions; token-level probing of the GPT-5.4 base model left the correct token as the argmax in eight of ten conditions. Gemini-3.1 Flash-Lite showed a steep framing-dependent policy: 0.1 percent for self-attributed falsehoods, 23.5 percent for user-cited sources, 68.2 percent for system-injected authority, and 94.0 percent for instruction override. GLM-4.5-Air showed a shallower gradient (15.8 and 84.2 percent for those two vectors). This 68-percentage-point dissociation is consistent with authority deference and instruction compliance engaging distinct mechanisms within a single architecture. Recovery also differed: GLM recovered in 94.5 percent of affected sessions, whereas 26.1 percent of affected Gemini sessions did not return to factual accuracy within the window (40.0 percent under instruction override). We release the evaluation framework and taxonomy for reproducible testing.

Let $\mathcal G$ be a split semiabelian variety of dimension $n \ge 2$ and let $\mathcal C \subset \mathcal G$ an irreducible curve, where everything is defined over $\overline{\mathbb{Q}}$. We prove that, under the natural hypothesis that $\mathcal C$ is not contained in any proper algebraic subgroup of $\mathcal G$, the set of points $P$ of $\mathcal C$ lying in an algebraic subgroup $H$ which intersects the curve tangentially at $P$ is finite. This generalizes a result for curves in 2-dimensional tori by Marché and Maurin. We apply these results to square-freeness of certain geometric divisibility sequences.

We investigate the relation between directly measured hot-dust reverberation scales and spectroscopically predicted broad-line region (BLR) scales in the isolated Seyfert galaxies IC~5287 and Mrk~845 (MCG+09-25-022), combining ZTF optical light curves, WISE/NEOWISE monitoring, and archival SDSS spectroscopy. Both nuclei show BWB optical variability, RWB MIR behaviour, and delayed MIR responses consistent with circumnuclear dust reprocessing. For IC~5287, interpolated cross-correlation function analysis with flux-randomisation/random-subset-selection uncertainties yields a rest-frame W1 delay of $106_{-39}^{+24}$~d, corresponding to $0.089_{-0.033}^{+0.020}$~pc. The median W2 response occurs at a longer delay, but the lag distributions overlap substantially. The SDSS decomposition reveals strong host dilution and BLR reddening, $E(B-V)_{\rm BLR}=0.612_{-0.066}^{+0.067}$~mag. Applying the same colour excess to the AGN continuum as an illustrative scenario increases the predicted H$\beta$ BLR lag from approximately 4.8 to 11.7~d and reduces the dust-to-predicted-BLR radius ratios from approximately 20–25 to 8–10. For Mrk~845, transfer-function modelling gives a rest-frame W1 delay of $157_{-8}^{+19}$~d, corresponding to $0.132_{-0.007}^{+0.016}$~pc. Its W2 response occurs at a longer delay, with $P(\tau_{\rm W2}>\tau_{\rm W1})=0.963$, supporting wavelength-dependent hot-dust structure. The SDSS spectrum shows more moderate BLR reddening, $E(B-V)_{\rm BLR}=0.197_{-0.039}^{+0.046}$~mag. The comparatively large dust-to-predicted-BLR radius ratios in both galaxies therefore primarily reflect the spectroscopically predicted BLR scales. These results emphasise the need to distinguish directly measured dust reverberation radii from BLR radii inferred from single-epoch spectroscopy when interpreting the BLR–dust hierarchy.

The objective of this study is to simulate the propagation of the shock wave in water due to an explosion. The study is part of a global research program on the development of an alternative stimulation technique to conventional hydraulic fracturing in tight gas reservoirs aimed at inducing a distributed state of microcracking of rocks instead of localized fracture. We consider the possibility of increasing the permeability of rocks with dynamic blasts. The blast is a shock wave generated in water by pulsed arc electrohydraulic discharges. The amplitude of these shock waves is prescribed by the electrohydraulic discharges which generate high pressures of several kilobars within microseconds. A simplified method has been used to simulate the injected electrical energy as augmentation of enthalpy in water locally. The finite element code EUROPLEXUS is used to perform fluid fast dynamic computation. The predicted pressure is consistent with the experimental results. In addition, shock wave propagation characteristics predicted with simulation can be valuable reference for design of underwater structural elements and engineering of underwater explosion.

Multimodal knowledge graphs (KGs) integrate information from text, figures, tables, and other modalities into a unified structured representation, with the promise that richer evidence enables better inference. In GraphRAG systems built over such graphs, it is commonly assumed that retrieving evidence from more modalities at inference time improves downstream performance. Yet, redundant or overlapping multimodal evidence may distract language models in question answering (QA), and whether each modality contributes equally across questions, models, and tasks remains poorly understood. In this work, we study how modality-aware retrieval affects downstream inference in a multimodal GraphRAG pipeline, using document visual question answering (DocVQA) as a testbed. We extend an existing KG-based QA framework to be modality-aware, leveraging the graph structure to track which modality supports which facts and to selectively filter evidence at the edge level. This enables us to investigate whether providing all available multimodal evidence at inference time benefits QA, and to evaluate the contribution and cooperation of modalities across question, task, and model characteristics. Through a controlled analysis within a state-of-the-art multimodal GraphRAG pipeline, five multimodal LLMs and two DocVQA benchmarks, we find that tables and text provide the strongest contributions, and that combining modalities frequently produces redundancy rather than synergy, particularly for pairs involving textual information. Positive cooperation appears mainly between non-text modalities and depends on question intent and task type. Our findings argue for selective, modality-aware retrieval in the design of more effective GraphRAG systems, where modalities are filtered according to the downstream task rather than retrieved uniformly.

Reinforcement learning with verifiable rewards (RLVR) via Group-Relative Policy Optimization (GRPO) is widely used for multi-turn VLM agent training, yet it suffers from zero-gradient silence on uniform failures and coarse episode-level credit assignment. While On-Policy Distillation (OPD) and On-Policy Self-Distillation (OPSD) mitigate sparse rewards using hindsight information, their underlying mechanisms remain poorly understood. Through controlled counterfactual rollback probes across five multi-turn VLM agent benchmarks, we reveal that performance gains in OPSD/OPD are largely driven by physical state rollback at the pivot step, defined as the first unrecoverable action without remaining step budget. However, physical state rollbacks are computationally prohibitive and infeasible in real-world environments. To bridge this gap, we present Pivot-Aware Internalized Visual On-Policy Training (PIVOT), an RL framework that internalizes pivot localization and state restoration directly into token-level parameter updates, eliminating environment rollbacks during RL training and additional skill hints at test time. PIVOT unifies three functional roles within a single architecture: a failure Analyzer non-invasively localizes the pivot step and diagnoses failure modes from visual trajectory collages and action logs; a detached Teacher re-scores failed tokens under this privileged diagnostic context; and a Student optimizes joint GRPO and confidence-gated OPD objectives. At test time, both Teacher and Analyzer branches are stripped. Evaluated on five multi-turn VLM agent tasks across cognitive grid puzzles, 3D embodied control and navigation, and generative reasoning, PIVOT achieves 0.90 overall accuracy on Qwen2.5-VL-3B (+8% over SFT+GRPO baseline and +5% over previous SOTA) and scales to 0.92 on Qwen3-VL-2B (+12% over SFT+GRPO baseline).

As Large Language Models (LLMs) become central to how we access and share information, they play an increasingly powerful role in shaping global knowledge. However, as these models evolve, their outputs risk converging into a generative monoculture, where the diversity of perspectives they represent narrows over time. Studies at the model level often fail to pinpoint which specific topics or viewpoints are being marginalised or amplified in this process. In this paper, we introduce a method to measure shifts in topic saliency across model families, tracking what gains or loses prominence during post-training. Applying this approach to a case study of climate change discourse, we demonstrate how homogenisation affects the representation of diverse solutions across different models. We also test interventions to counter this trend, showing that specialised models can help preserve a broader range of perspectives. This underscores the importance of monitoring topic saliency to diagnose the risks of monoculture and to ensure AI systems reflect a pluralism of ideas. Data and Code are accessible \href{https://github.com/oriane/topic_saliency_shift}{here}.

This study presents an industrial experiment conducted with a dairy partner to investigate the temperature evolution of heat-generating soft cheeses within pallets during export cold chain. In the production factory, four homogeneous pallets of soft cheese were instrumented with temperature probes placed at three levels (bottom, medium, and top). Temperature was monitored over a 13 days export cold chain from France to Northern Europe. The objective was to evaluate the influence of air ventilation and pallet wrapping film on product temperature. The results clearly highlight the effect of ventilation: pallets placed in a well-ventilated area showed temperatures approximately 3{\textdegree}C lower than those in less ventilated zones. The study also underlines the importance of maintaining good ventilation of pallets before loading them into refrigerated transport vehicles. However, the effect of wrapping film remains unclear and does not allow definitive conclusions. This field study provides practical recommendations for better conditioning of heat-generating products in the cold chain.

Can sustained, per-identity participation be engineered into a security resource? Most anti-Sybil defenses price identity creation rather than identity survival. Once admitted, an adversary may sustain many identities without paying a recurring cost. We introduce the Bounded Participation Channel (BPC), a formal primitive for repeatedly verifying participation window by window. BPC issues fresh, identity-bound challenges under a strict deadline and enforces four structural properties: identity binding, freshness, real-time response, and bounded per-channel throughput. Together, these yield a provable cost theorem: sustaining $s$ identities over $T$ windows requires $C(s,T) \geq sT/\tau_h$ participation channel-windows. The guarantee is solver-agnostic: a channel may be operated by a human, an AI system, or a hybrid. We give a hash-based construction with publicly verifiable participation proofs, characterize four admissible challenge families, and evaluate two against GPT-4o, Gemini 2.5 Flash, and Claude Sonnet 4.5 across 600 trials. Despite near-perfect accuracy (97–100%) on the perceptual tasks, the evaluated automated channels remain throughput-bounded under the tested deployment conditions. The results illustrate a key distinction: solvability does not imply unlimited throughput. By requiring participation to be re-earned by every identity in every time window, BPC turns sustained participation into a measurable security resource with a linear structural cost floor, independent of whether the participation is supplied by humans, AI systems, or hybrids.

While the type-I seesaw model provides an attractive framework for simultaneously explaining the origin of neutrino masses and the baryon asymmetry of the Universe, flavor symmetries offer a natural approach to understanding the observed neutrino mixing pattern. However, the highly constrained neutrino mass structures predicted by flavor symmetries may also prevent conventional leptogenesis to work. In particular, when the Dirac neutrino mass matrix $M^{}_{\rm D}$ or the Majorana right-handed-neutrino (RHN) mass matrix $M^{}_{\rm R}$ is exactly degenerate (namely, they are proportional to the identity matrix $I$), the CP asymmetry of RHN decays will vanish either due to the orthogonality of different neutrino Yukawa coupling columns or the exact degeneracy of RHN masses. In this work, we propose that the flavon fields (which are inherently present in flavor-symmetry neutrino mass models to be responsible for generating the nontrivial neutrino flavor structure) can naturally overcome these obstacles. For the case of $M^{}_{\rm D} \propto I$, flavon–RHN interactions induce additional decay channels $N_I \to N_J \phi$ and generate new CP-violating contributions to RHN decays. For the case of $M^{}_{\rm R} \propto I$, three-body decays $N_I\rightarrow L^{}_\alpha H\phi$ mediated by heavy vectorlike fermions provide a CP asymmetry source even when the RHN masses are exactly degenerate. Our results demonstrate that the flavon sector provides a natural connection between neutrino flavor structure and leptogenesis, offering a new mechanism for leptogenesis in flavor-symmetry-based neutrino mass models.

Most clinical prediction systems learn patient-variable-outcome associations; we investigate a training-free diagnostic paradigm mapping patient observations to explicit medical knowledge. CKG Reasoner integrates candidate-specific Evidence Feature Nodes, patient-reference matching, a bounded Information Gate, knowledge-weighted evidence accumulation, disease similarity, and decisive clinical rules. Missing-aware normalization and coverage auditing distinguish absent from unavailable evidence. Candidate ranking is separate from outcome-label-independent K-means clustering, which uses four derived evidence coordinates (evidence strength, relative magnitude, directional similarity, and evidence completeness), not raw predictors or targets, to derive cohort-level assignments. Across six retrospective cohorts - four dengue (N = 1000, 1523, 989, 1018), malaria (N = 2190), and influenza (N = 4569) - a uniform, label-free, cohort-fitted K = 2 protocol yielded positive-class F1 scores of 0.996, 0.634, 0.936, 0.917, 0.695, and 0.842, and all-record accuracies of 0.996, 0.558, 0.914, 0.893, 0.707, and 0.906, respectively, with full partition-decision coverage using the frozen package and disease-specific knowledge representations. Neither scoring nor clustering uses outcome labels. Logistic regression provides a supervised baseline. Influenza incorporates confirmatory molecular PCR and is not independent pre-test prediction. Results characterize knowledge-grounded evidence separation, auditability, and sensitivity, not prospective clinical validity or comparative superiority. FOL/LLM-based clinical explanation remains unevaluated.

Pretraining a language model takes enormous compute, and the right optimizer can save a good part of it. Muon's spectral step gives a stronger direction than a sign step, but it is expensive. Every step runs Newton-Schulz iterations on the full matrix and, in distributed training, an extra all-reduce. Sign steps, as in Lion and Signum, are cheap and stay local to each device. We propose LionMuon, which takes one Muon step every $P$ iterations and Lion steps in between, with a single dual-EMA momentum buffer shared by both. Muon's compute and communication are paid once per $P$ steps, and the optimizer state is half of AdamW's. A single-EMA variant, SignMuon, already improves on Muon. We prove complexity bounds under heavy-tailed noise in which the period sets an interpolation between Muon's and Lion's smoothness and noise constants, and which say when LionMuon is faster than both. On 124M and 355M models trained on FineWeb, LionMuon with $P=2$ and $P=5$ reaches a lower loss than Muon, AdamW, Lion and Signum at the same number of tokens. Under 4-GPU data-parallel training it reaches Muon's final loss with a third less wall-clock on PCIe, and it beats the communication-efficient Muon variants Dion and MuonBP on loss at no more exposed communication, while keeping the exact gradient. Code: https://github.com/brain-lab-research/lion-muon

Computational antibody design requires representations that capture the geometric patterns underlying antigen–antibody interactions, yet existing approaches often rely on scalar distances or surface-intrinsic features, leaving cross-molecular geometry largely implicit. We present AbGaze, an end-to-end antibody design framework based on attentive geometric representation learning, which encodes distance, spatial direction, and surface-normal orientation of antigen surfaces relative to antibody-residue local frames, and adaptively aggregates these geometric interactions according to their interfacial context. The learned interaction representation is shared across multi-CDR co-design, complex structure prediction, and affinity optimization, with local-frame geometric supervision further constraining the representation. AbGaze outperforms prior methods across all three tasks: relative to the second-best method, it improves amino-acid recovery by 7.1% and reduces structural error by 14.9% on average over the six CDRs, improves interface docking quality (DockQ) by 6.6%, and raises the affinity improvement rate (IMP) by 32.5%.

We address Task A of the 1st DAFx Parameter Estimation Challenge, which aims to retrieve the physical parameters of a plate model from an impulse response. To do so, we use the Simulation-Based Inference (SBI) framework, in which we train a neural network to estimate a density over plate parameters given an impulse response, using a dataset generated by the simulator. Inference for a new impulse response then requires only a forward pass through the network, without involving the simulator. For each test observation, we fine-tune a specific network: additional simulation rounds are performed by sampling parameters from the current estimated distribution, simulating the corresponding impulse responses, and fine-tuning to produce the specialized network.

Object hallucination remains a major challenge for large vision-language models. While off-policy preference optimization proves to be an effective solution, on-policy reinforcement learning provides a more promising direction as it directly targets a model's current failure modes. However, we find that without fine-grained reward formulation and allocation, on-policy optimization often falls into an easy shortcut: reducing hallucinations merely by saying less—making fewer valid claims. To comprehensively resolve this, we propose a fine-grained alignment framework that couples dense reward signals at the data level with precise credit assignment at the algorithmic level. Specifically, we first construct the Dense Object Presence and Absence (DOPA) dataset to address sparse annotations that prevent valid object claims from being verified and rewarded. DOPA exhaustively annotates the deterministic presence and absence of every concept across an expanded vocabulary, significantly increasing the density of reliable reward signals during on-policy rollouts. Second, we propose Subsentence-level Credit Assignment for on-Policy Optimization (SCAPO) to prevent response-level shared advantages from allowing local hallucinations to compromise all other valid outputs within the same response. By assigning credit to each subsentence independently based on its object claims, SCAPO can precisely reinforce faithful generations and penalize hallucinations. Furthermore, we leverage the resulting faithful image descriptions as auxiliary context to transfer generative gains to discriminative tasks. Experiments demonstrate that our method produces highly informative, faithful descriptions in generative tasks while yielding clear performance gains on discriminative evaluation.

Aspect-based sentiment analysis (ABSA) has largely turned to text generation. We show that competitive dimensional ABSA does not need it. Using Jev, a frozen model that answers typed questions with rubric scores, label probabilities, and yes/no judgments, we decompose all three tasks of SemEval-2026 Task III Track A into such decisions and align them with the annotation scheme through 488 coefficients fitted on CPU, with no text generation and no backbone tuning. On valence-arousal regression over ten corpora in six languages, the system reaches 1.0645 RMSE, the lowest aggregate error of any participating system. On triplet and quadruplet extraction, it reaches 52.09 and 44.06 continuous F1, above fine-tuned Llama-3.3-70B and GPT-OSS-120B baselines. Analyses and ablations show where the accuracy comes from: supervised calibration roughly halves the raw regression error, exact valence-arousal would add only 4.5 F1 to extraction, and the learned combination of span-boundary evidence, not any single signal, carries the extraction systems.

Recent representation alignment (REPA) methods accelerate diffusion transformer training by aligning projections of the transformer's hidden states with representations from pretrained visual encoders. In this work, we explore a reverse and complementary direction to REPA: rather than projecting diffusion representations into the encoder's space, we inject encoder representations into the diffusion transformer, allowing them to actively participate in the denoising process. To this end, we introduce REPresentation Injection (REPI), a training framework based on a scaffold-to-internalization strategy, in which projected encoder representations initially serve as a temporary scaffold and are then progressively internalized by the diffusion transformer. REPI outperforms REPA across a wide range of backbones and is highly complementary to it: combining the two yields substantial gains over either alone. Notably, with only 160K training steps, REPI + REPA matches vanilla SiT trained for 7M steps, a speedup of over $43.5\times$. Code will be available at https://jeneveuxpas.github.io/REPI

Modern AI models are aligned through post-training to adapt them to downstream tasks. Recent work shows that fine-tuning language models on narrow tasks can induce emergent misalignment (EM), causing broadly harmful behaviors beyond the training task. However, EM has been studied almost entirely in text-only tasks, leaving its manifestation in multimodal models unclear. In this paper, we define and analyze EM in the context of vision-language models. We first induce EM via fine-tuning on narrow multimodal tasks targeting vulnerable code, careless household-object use, and conspiratorial interpretations of ordinary scenes. Across fifteen commercial and open-source models with different scales, we find that narrow multimodal fine-tuning can induce coherent and broadly misaligned behavior that transfers to unrelated tasks, including misaligned opinions, visual factual dishonesty, unsafe image generation, vulnerability to visual jailbreaks, and risky agentic actions. We further find that multimodal EM does not depend on the apparent harmfulness of training data but is sensitive to training-evaluation modality alignment. EM can arise under both supervised fine-tuning and preference optimization and can propagate through intermediate reasoning. Finally, we explore several mitigation strategies, including prompt inoculation, benign continued training, and activation-level steering, which can partially reduce EM. Overall, our findings suggest that multimodal EM reflects a behavioral shift rather than a general loss of capability, extending beyond text to the visual modality.

Recent work has explored improving agents by jointly evolving their harnesses and models, but often takes a ”potpourri” approach that bundles together new tools, new decision-making procedures, and model adaptation to the evolved harness under a single notion of agent improvement. In this paper, we instead investigate how agents can improve their decision-making procedures. In particular, we propose EvoIn, an agent fine-tuning framework that bridges evolution and internalization. EvoIn first analyzes agent execution traces to evolve and validate new decision-making procedures by temporarily instantiating them in the harness. The validated procedures guide the agent to generate improved reasoning traces. These traces are then rewritten into self-contained reasoning traces, removing explicit references to harness instructions while expressing the induced decision logic as the model's own reasoning. Finally, EvoIn fine-tunes the model on the rewritten traces, internalizing these procedures so that the improved decision-making persists without the evolved harness at inference time. We evaluate EvoIn on diverse benchmarks and find that it consistently enables agents to learn stronger decision-making procedures, raising the pass rate by 10.9 points in-domain and by 9.2 points out-of-domain. Results further show that the internalized decision procedures generalize to unseen tasks. Case studies show that agents can learn to decide how to solve a task before solving it, for example by checking a document's length to choose between reading it in full and searching it. EvoIn is also broadly applicable, showing consistent improvements on another model family.

Denoising Diffusion Probabilistic Models have shown remarkable performance in unconditional image generation. In order to generate images with desired semantics, recent works have restricted the solution space by using guidance constraints in the diffusion sampling process. However, for image enhancement across different domains, these methods struggle to balance two main requirements: looking realistic in the target domain (photorealistic images) and preserving relevant features of the source domain, e.g., low-quality renderings or art paintings. Here, small local changes can alter the fidelity of the image completely, while large changes in other regions might be insignificant. We introduce LocDiff, a locality-constrained guidance method for image enhancement, which serves as a zero-shot extension to pre-trained diffusion models, ensuring the preservation of critical features during domain adaptation. In this way, we retain important local features, while allowing less critical regions to remain unconstrained and not interfere with the guidance process for relevant regions. We evaluate our method on two different domain-shift tasks: For art-to-photo translation, we apply the method in a fully zero-shot setting, preserving facial identity from paintings while generating photorealistic details. For enhancing low-quality fetal ultrasound renderings, we demonstrate zero-shot inference with auxiliary prior alignment. Here, the objective is to artificially add high-resolution characteristics and produce photorealistic ultrasound renderings, a target domain for which no ground truth distribution exists. Our experimental results demonstrate that LocDiff achieves favorable realism-faithfulness trade-offs compared to state-of-the-art methods, enabling controllable cross-domain enhancement.

Joint-embedding self-supervised learning typically combines an invariance objective across augmented views with additional mechanisms to prevent representational collapse. These objectives are often applied after a projection head, while downstream tasks use the backbone representation before the projector. We find that this mismatch does not necessarily prevent dimensional collapse in the backbone, which can retain low effective rank and potentially limit downstream transfer. To address this, we introduce SACReg, a spectral anti-collapse regularizer motivated by an analysis of $\lambda$-balance, which captures the relative scale of weight matrices across layers. In a two-layer linear network, we show that (i) $\lambda$-balance prevents collapse, and (ii) our regularizer applied to the backbone induces $\lambda$-balance. In the nonlinear case, this regularizer leads to anti-collapse as well and, in realistic architectures on ImageNet100, it empirically increases the representations' ranks. We apply SACReg to JEPA and propose $\lambda$-JEPA, which improves over LeJEPA and VISReg on ImageNet-1k classification and in average linear-probe transfer performance across eight downstream image datasets. On video self-supervised learning, $\lambda$-JEPA improves over LeVJEPA and V-JEPA 2 on the Something-Something-v2 and Kinetics-400 benchmarks. Code is available at https://github.com/berkerdemirel/lambda-jepa.

Let $\mathcal{C}$ be a spherical fusion category with global dimension $D$. The Galois conjugates of the dimension character index a block of the $S$-matrix of the Drinfeld center, which we call the dimension block. We prove that for every character $\chi$ of $\mathrm{Gal}(\mathbb{Q}(D)/\mathbb{Q})$, the twisted sum $\sum_\sigma\chi(\sigma)/\sigma(D)$ has absolute value at most $e^{-1}\mathfrak{f}(\chi)^{-1/2}$, where $\mathfrak{f}(\chi)$ is the conductor of $\chi$ and $e$ is the order of the dimensional grading group of $\mathcal{C}$. As an application, we determine the possible global dimensions of spherical fusion categories below $\sqrt{5}$, extending the known classification up to $4\sqrt{3}/5$ of V.\ Ostrik and P.\ Etingof.