arXiv:2607.16097v1 Announce Type: new
Abstract: Reinforcement learning (RL) has become central to improving large language models (LLMs) on complex reasoning tasks, yet RL post-training is largely studied in isolation from the pretraining that precedes it. As a result, two basic questions remain open: (1) how do pretraining choices (model size, data) shape the returns to RL compute, and (2) what does RL actually do to the model? These questions are difficult to study in the standard LLM setting: pretraining corpora are vast and uncontrolled, making it hard to attribute behaviors to pretraining versus RL, and systematic compute sweeps across both stages are prohibitively expensive. To address these challenges, we use chess as a controlled testbed for studying reasoning across the full pretraining-to-post-training pipeline. We follow the standard LLM training pipeline by pretraining language models from 5M to 1B parameters on human chess games, supervised fine-tuning on synthetic reasoning traces, and running RL on chess puzzles with verifiable rewards. Using this framework, we find that the post-RL performance at given RL compute level is well-predicted from the pretraining loss, and slope of the RL reward curves improves approximately linearly with the pretraining tokens. Beyond scaling, we find that RL does not simply sharpen the SFT policy: on easy puzzles it amplifies correct moves the SFT policy already preferred, while on hard puzzles it surfaces correct moves that were nearly absent under SFT. We further test whether our findings transfer beyond chess by training a 1B language model on math-domain text, where the same predictive pattern emerges: longer-pretrained checkpoints reach higher post-RL performance and improve faster under RL. In sum, we provide a quantitative account of the pretraining-to-RL interface and a controlled testbed for studying the science of reasoning across the full pretraining-to-post-training pipeline.
Science Journals
arXiv:2607.16117v1 Announce Type: new
Abstract: Language models encode text as subword tokens, raw bytes, or rendered pixels, but these encodings are usually compared under modeling constraints that expose different amounts of linguistic content to models across different languages. We instead ask what each encoding preserves when both the content and the downstream capacity are controlled. Using verified parallel sentences across thirteen languages and five scripts, we compare tokens, bytes, and pixels through a shared bottleneck whose width is swept to trace rate-utility frontiers. This separates three quantities that are often conflated: the number of input positions an encoding creates, the latent capacity available after encoding, and the task-relevant information that survives compression. We evaluate three utilities: surface form preservation, cross-lingual sentence alignment, and topic classification. No encoding dominates across tasks or capacity regimes. Pixels preserve surface form best, bytes preserve cross-lingual alignment best, especially in same-script multilingual settings, and tokens support topic prediction best. These performances are not explained by sequence length alone. Short inputs can discard useful meaning, while long inputs can preserve information that compresses well. Choosing an encoding is therefore not a fixed preference for tokens, bytes, or pixels, but a rate-utility tradeoff that depends on the task, language mix, capacity regime, and compute budget.
arXiv:2607.15717v1 Announce Type: new
Abstract: Text-to-motion models are competent at the action a prompt names but unreliable at when each stroke lands: four punches alternating left and right rarely return four separable strokes. We introduce typed temporal events called Action Units (AUs) that make the individual stroke -- its body track, action class, time window, and impact timing -- an explicit conditioning signal. We ground a frozen text-to-motion backbone on the AU set through a lightweight gated adapter injecting two streams (per-stroke tokens and a per-frame phase channel), and at inference close residual timing errors with a training-free classifier gradient from a frozen frame-level detector. We measure per-stroke control on StrokeBench, whose prompts specify count, ordering, track, and core-frame placement, paired with an audited stroke corpus. AU grounding markedly raises the rate of correctly placed single strokes over the strongest prior interface, at the best motion quality among text-, interval-, and frame-level baselines. The prompted core frame emerges as a further steerable axis.
arXiv:2607.15718v1 Announce Type: new
Abstract: Concept Design structures software systems around concepts: user-facing, self-contained units of functionality with a focused purpose. Concepts are composed into applications using synchronization rules called reactions, which specify how actions in one concept trigger actions in others. This paper first gives a formal semantics for concepts and reactions, enabling automatic verification of safety invariants in applications developed with this methodology. It then presents a CEGIS-style, LLM-driven synthesis procedure for generating reaction designs that satisfy such invariants. Because many different designs can satisfy the same invariant, we study two ways of steering synthesis toward the user's intended design: natural-language prompts and positive/negative scenarios. We also propose an LLM-driven scenario elicitation technique to support early design exploration. In an evaluation on three applications and twelve design variants using one LLM configuration, invariant-only synthesis reached verified designs quickly but often produced inconsistent designs across runs, some of which were implausible, showing that invariants alone underconstrain the design task. Scenario-guided synthesis recovered intended designs more consistently than natural-language prompting, although minimal scenarios can lead to overfitting. LLM-driven scenario elicitation, where the user classifies proposed scenarios rather than authoring them from scratch, recovered the intended designs in most variants when enough scenarios were elicited, but missed behaviors and non-determinism prevented reliable coverage in all cases.
arXiv:2607.15725v1 Announce Type: new
Abstract: Spectral-domain ghost imaging enables high-resolution spectroscopy with a single-pixel detector. The technique does not rely on spectrally resolved detectors, which makes it inherently robust against turbulence and particularly adapted to weak-light conditions. These features are very attractive for spectral imaging in the mid-infrared region which hosts numerous molecular absorption features but lacks highly sensitive detectors. The implementation of spectral ghost imaging in the mid-infrared has however been limited by the absence of suitable light sources and detectors capable of generating and measuring spectral fluctuations in real time. Here, we demonstrate spectral-domain computational ghost imaging in the mid-infrared based on a nonlinear frequency downconversion scheme. Pre-programmed spectral patterns imposed on broadband light at 1.5 mm using a programmable spectral filter are transferred into the mid-infrared through difference-frequency generation in a nonlinear crystal. This enables computational ghost spectroscopy with a spectral resolution of 0.62 cm-1 using a single-pixel mid-infrared detector. The method is flexible, broadly applicable and, as proof of concept, we demonstrate ghost spectroscopy in mid-wavelength infrared and long-wavelength infrared bands using the nonlinear frequency conversion in chirped-poling lithium niobate and ZnGeP2 crystals, respectively. Our approach provides a new avenue for mid-infrared spectroscopy, remote sensing and spectral imaging.
arXiv:2607.15928v1 Announce Type: new
Abstract: Adult and pediatric electrocardiogram (ECG) interpretation relies on age-sensitive criteria, and models pretrained mainly on adult ECGs often transfer poorly to pediatric populations when pediatric labels are scarce. Existing multimodal ECG--text methods typically align waveforms and text at the global sample level, entangling evidence from co-occurring diagnoses and limiting transfer under this gap. We propose Pediatric-Adult ECG Alignment via Cross-modal Enhancement (PEACE), a knowledge-guided framework pretrained on the largely adult MIMIC-IV ECG corpus. PEACE describes each diagnosis along rhythm, morphology, and ST--T axes and, per recording, composes only positive-label descriptors into three axis tokens and a fused embedding. A label query network (LQN) uses diagnostic labels as queries to cross-attend over ECG tokens and axis tokens, while label set aware bidirectional contrastive learning (LSBC) aligns pooled ECG features with the fused embedding when recordings share diagnoses. Curriculum adaptive fusion (CAF) gates alignment strength according to smoothed classification loss and training progress, limiting disruption during early optimization. The knowledge branch is used only for training supervision; inference uses ECG signals alone. On ZZU-pECG, PEACE reaches macro average AUCs of 59.39%, 81.74%, and 91.56% under zero-shot, 50-shot, and full fine-tuning, with the clearest gains over foundation and knowledge-pretraining baselines under limited supervision; versus domain adaptation initializations, zero-shot improves substantially while 50-shot AUC is comparable to DANN. After fine-tuning on PTB-XL, PEACE reaches 96.90% macro average AUC over nine harmonized labels. Ablations confirm that label-conditioned knowledge alignment, rather than global text fusion, is the key driver of pediatric transfer gains.
arXiv:2607.15819v1 Announce Type: new
Abstract: In-context learning is a remarkable property of transformers and has recently received a lot of interest. In many studies of in-context learning, it has been shown that transformers are capable of implementing solver for linear and non-linear regression problems, in which the most of them implement gradient descent algorithm. However, it is still unclear whether those implementations have actually been acquired through training. In this paper, we construct a transformer with linear self-attention, which in-context learns the least squares estimate in a simple regression task. The point here is that the closed form (analytical) solution is approximately obtained by using layer normalization rather than an approximate solution based on gradient descent algorithm. Then, we show an experimental example, in which our implementation is mainly used in the transformer trained with l1 regularization when the target output is the least squares estimate.
arXiv:2607.15930v1 Announce Type: new
Abstract: Conventional electromagnetic (EM)-based integrated sensing and communication (ISAC) systems degrade in cluttered, obstructed, and radio-frequency-hostile environments, while macroscopic molecular communication (MC) remains largely unexplored as an ISAC medium. This article introduces VaporISAC, a molecular ISAC framework in which chemical vapor pulses simultaneously convey information and probe the propagation environment, enabling a one signal, two outputs paradigm. The same received waveform is jointly processed to recover transmitted information and infer environmental properties such as airflow, turbulence, smoke, and chemical conditions. Rather than replacing conventional EM-based ISAC, VaporISAC complements existing approaches in chemically dynamic, infrastructure-limited, and EM-challenged environments. The sensing principles, system architecture, proof-of-concept demonstrations, emerging applications, and open research challenges of VaporISAC are presented, positioning it as a promising new paradigm for resilient communication and environmental sensing.
arXiv:2607.16129v1 Announce Type: new
Abstract: We introduce quantum Monte Carlo (QMC)-trained multi-state machine-learned (ML) force fields for nonadiabatic excited-state dynamics, targeting photochemical processes in which the electronic character changes along the reaction path and a consistent correlated description is required. In this framework, variational Monte Carlo wave functions combine compact selected configuration-interaction expansions with a Jastrow factor that explicitly accounts for dynamical correlation, while neural networks convert the stochastic QMC data into smooth potential energy surfaces for large surface-hopping ensembles. We apply this approach to azomethane, a demanding test case involving torsional relaxation through conical-intersection regions and C--N bond dissociation on the hot ground state. Benchmark calculations support the accuracy of the QMC reference data and show robust force convergence across isomerization and dissociation geometries. The QMC-trained dynamics preserves the expected photoisomerization mechanism, strongly reduces the excessive C--N breaking obtained with complete active space self-consistent field, and predicts a small but non-negligible prompt dissociation component after internal conversion, with a timescale consistent with femtosecond-resolved mass-spectrometry experiments. These results establish QMC-ML as a practical route to nonadiabatic photochemical dynamics with accurate wave-function reference data.
arXiv:2607.16130v1 Announce Type: new
Abstract: AI governance increasingly requires judgments about whether an AI system remains adequately trustworthy over time, whether observed changes are tolerable, and how such judgments should be documented in a transparent and contestable way. Yet existing work on AI trustworthiness remains either too high-level to support lifecycle monitoring and reassessment or too narrowly metric-driven to connect with governance needs. We therefore propose a lightweight methodology for auditable trustworthiness levels in AI governance. The methodology has two components: a formal framework for representing and learning trustworthiness levels, and a lightweight AI lifecycle governance procedure for documenting, monitoring, and reassessing them over time. The formal framework models governance-relative trustworthiness through a context-sensitive protocol of measurable dimensions and learns trustworthiness levels as interpretable rules over trustworthiness profiles. Using decision trees as an interpretable proof-of-concept model class, the methodology yields explicit trustworthiness plateaus, readable level transitions, and two simple lifecycle diagnostics: boundary margins and profile drift. The governance procedure embeds these formal objects in a conformity-oriented workflow for design-time labeling, post-deployment monitoring, reassessment, and reporting. It also assigns human responsibilities and control gates for protocol design, validation, monitoring, and reassessment. We illustrate the methodology on synthetic AI lifecycle traces involving degradation, shocks, updates, heterogeneous monitoring cadences, and system comparison. Our methodology does not replace legal or other expert judgment: it supports conformity documentation and lifecycle monitoring by providing an evidential basis for documenting and tracking AI governance-relevant changes over time.
arXiv:2510.21971v3 Announce Type: replace
Abstract: Inertial spheres settling through sharp density interfaces can arrest, reverse direction, and resume descent, a phenomenon known as bouncing. Using synchronized particle image velocimetry and tracking in water-salt and water-glycerol stratifications, we demonstrate that bouncing is the dynamic response of a coupled sphere-fluid composite. As the sphere crosses the interface, it entrains a boundary layer of lighter fluid, creating a transient buoyant wake.
We formalize this mechanism into a phenomenological dynamic model that couples the momentum of the sphere with the entrainment and detachment of the wake. Evaluating the stationary points of this system yields a criterion that classifies trajectory archetypes (smooth crossing, deep minima, and bouncing) across different fluid regimes. We identify a dual role of viscosity, which is often overlooked by density-only models: it acts kinematically to thicken the boundary layer and increase the entrained wake volume, and dynamically to alter the drag-to-weight balance.
Furthermore, we describe the spatial dynamics of the crossing: inertia-dominated spheres penetrate further into the lower fluid before arresting due to a longer wake-detachment length, whereas buoyancy-dominated spheres arrest closer to the interface. Finally, we show that the retention time is governed by the buoyancy-driven detachment of the entrained film. By normalizing the measured retention times with a characteristic detachment timescale, we collapse the data from different viscosity regimes onto a single curve. These physical insights allow the prediction of trajectory archetype, deceleration depth, and retention time from bulk properties.
arXiv:2510.24819v2 Announce Type: replace
Abstract: Large Language Models (LLMs) are increasingly embedded in software systems ( GenAIware), enabling new forms of automation and interaction However, their probabilistic nature and reliance on prompt programming challenge reliability, robustness, and maintainability In current practice, prompt-related concerns (e.g., context management, interaction logic, output validation) are embedded in general-purpose code, leading to implicit, hard-to-analyze systems We argue that prompt programming should be treated as a first-class Software Engineering (SE ) concern and propose LLM Scripting Language (LSL ), a Domain Specific Language ( DSL) for structuring LLM interactions as analyzable programs LSL introduces abstractions for interaction blocks, context scopes, output constraints, and control flow, separating deterministic logic from probabilistic model behavior while ensuring syntactic compliance From an SE perspective, LSL supports disciplined development by making interaction logic explicit, analyzable, and amenable to verification and validation It also acts as cognitive scaffolding, externalizing prompt design into programmable artifacts that reduce implicit reasoning and support systematic debugging, evolution, and reuse We illustrate these properties in a structured generation scenario, showing improved failure localization and interaction transparency While LSL does not guarantee semantic correctness or factual accuracy, it provides a principled foundation for more analyzable and maintainable prompt-based systems.
Global Kilometer-Scale Simulations with ARP-GEM2: Effect of Parameterized Convection and Calibration
arXiv:2511.00829v2 Announce Type: replace
Abstract: The objective of this paper is twofold. First, it documents the second version of the global atmospheric model ARP-GEM and its calibration at kilometer-scale resolution. The model is currently able to run simulations at a resolution of up to 1.3 km. Second, this paper focus on multi-year global atmospheric simulations at a 2.6 km resolution with and without parameterized convection and associated calibration. Simulations without deep convection tend to be similar to those with infinite, or at least large, entrainment values. Consistently, entrainment and detrainment are used as primary drivers for the gradual reduction of convection as resolution increases. The results indicate that, with this hydrostatic model, parameterized convection still plays a significant role in the correct representation of the mean state at the kilometer scale. Additionally, they suggest some added value of high resolution in representing climate variability. However, a compromise between the adequate representation of the mean state and variability is necessary, as both are differently favored by the degree of parameterized convection. Finally, it is likely that even higher resolutions are necessary to achieve an unequivocal added value.
arXiv:2511.12974v3 Announce Type: replace
Abstract: We introduce Controlled Stochastic Activity Networks (Controlled SANs), a formal extension of classical Stochastic Activity Networks that integrates explicit control actions into a unified semantic framework for modeling distributed real-time systems. Controlled SANs systematically capture dynamic behavior involving nondeterminism, probabilistic branching, and stochastic timing, while enabling policy-driven decision-making within a rigorous mathematical framework.
We develop a hierarchical, automata-theoretic semantics for Controlled SANs that encompasses nondeterministic, probabilistic, and stochastic models in a uniform manner. A structured taxonomy of control policies, ranging from memoryless and finite-memory strategies to computationally augmented policies, is formalized, and their expressive power is characterized through associated language classes. To support model abstraction and compositional reasoning, we introduce behavioral equivalences, including bisimulation and stochastic isomorphism.
Controlled SANs generalize classical frameworks such as continuous-time Markov decision processes (CTMDPs), providing a rigorous foundation for the specification, verification, and synthesis of dependable systems operating under uncertainty. This framework enables both quantitative and qualitative analysis, advancing the design of safety-critical systems where control, timing, and stochasticity are tightly coupled.
arXiv:2511.14159v4 Announce Type: replace
Abstract: Evaluating the robustness of Large Vision-Language Models (LVLMs) is essential for their continued development and responsible deployment in real-world applications. However, existing robustness benchmarks typically focus on hallucination or misleading textual inputs, while largely overlooking the equally critical challenge posed by misleading visual inputs in assessing visual understanding. To fill this important gap, we introduce MVI-Bench, the first comprehensive benchmark specially designed for evaluating how Misleading Visual Inputs undermine the robustness of LVLMs. Grounded in fundamental visual primitives, the design of MVI-Bench centers on three hierarchical levels of misleading visual inputs: Visual Concept, Visual Attribute, and Visual Relationship. Using this taxonomy, we curate six representative categories and compile 1,248 expertly annotated VQA instances. To facilitate fine-grained robustness evaluation, we further introduce MVI-Sensitivity, a novel metric that characterizes LVLM robustness at a granular level. Empirical results across 18 state-of-the-art LVLMs uncover pronounced vulnerabilities to misleading visual inputs, and our in-depth analyses on MVI-Bench provide actionable insights that can guide the development of more reliable and robust LVLMs. The benchmark and codebase can be accessed at https://github.com/chenyil6/MVI-Bench.
arXiv:2511.18468v2 Announce Type: replace
Abstract: Continual Test-Time Adaptation (CTTA) is crucial for deploying models in real-world applications with unseen, evolving target domains. Existing CTTA methods, however, often rely on source data or prototypes, limiting their applicability in privacy-sensitive and resource-constrained settings. Although several methods attempt to mitigate catastrophic forgetting, they often fail to preserve long-term domain-specific knowledge across many domain shifts. Moreover, their relatively slow adaptation rates during domain transitions can cause error accumulation, allowing mistakes to propagate before effective adaptation occurs. To address these challenges, we propose SloMo-Fast, a source-free, dual-teacher CTTA framework designed for enhanced quick adaptability and generalization. It includes two complementary teachers: the Slow-Teacher, which exhibits slow forgetting and retains long-term knowledge of previously encountered domains to ensure robust generalization, and the Fast-Teacher rapidly adapts to new domains while accumulating and integrating knowledge across them. This framework preserves knowledge of past domains and adapts efficiently to new ones. Our extensive experiments show that SloMo-Fast consistently outperforms state-of-the-art methods across Cyclic Test-Time Adaptation (Cyclic-TTA), a CTTA benchmark that simulates recurring domain shifts, along with ten other CTTA settings, highlighting its ability to both adapt and generalize across evolving, revisited domains.
arXiv:2511.19356v4 Announce Type: replace
Abstract: Group Relative Policy Optimization (GRPO) enables stable and preference-oriented updates via group-wise comparisons for post-training video generation. However, GRPO directly optimizes reward-induced advantages. Under sustained optimization, the reward score can lose fidelity as a proxy for true video quality, consistent with the phenomenon described by Goodhart's Law. This leads to two recurring issues: (i) shortcut-driven optimization under composite objectives and (ii) reward saturation within prompt groups. To address these issues, we introduce TaRoS, a Target-Robust Reward Signaling framework for Video generation GRPO. TaRoS leverages component level performance assessment together with intra-group sparsity to organize multi-aspect rewards towards optimization objectives. In addition, it adaptively downweights components that exhibit saturation, thereby preserving effective optimization directions and mitigating redundancy. This maintains meaningful optimization directions and preserves within-group ranking separation, thereby preventing reward hacking and leading to more reliable policy updates. Extensive experiments show consistent improvements in visual fidelity, motion coherence, and text-video alignment over strong baselines.
arXiv:2511.20332v3 Announce Type: replace
Abstract: This article trained a network for perceiving three-dimensional motion information of binocular vision target, which can provide real-time three-dimensional coordinate, velocity, and acceleration, and has a basic spatiotemporal perception capability. Understood the ability of neural networks to fit nonlinear problems from the perspective of PID. Considered a single-layer neural network as using a second-order difference equation and a nonlinearity to describe a local problem. Multilayer networks gradually transform the raw representation to the desired representation through multiple such combinations. Analysed some reference principles for designing neural networks. Designed a relatively small PID convolutional neural network, with a total of 17 layers and 413 thousand parameters. Implemented a simple but practical feature reuse method by concatenation and pooling. The network was trained and tested using the simulated randomly moving ball datasets, and the experimental results showed that the prediction accuracy was close to the upper limit that the input image resolution can represent. Analysed the experimental results and errors, as well as the existing shortcomings and possible directions for improvement. Finally, discussed the advantages of high-dimensional convolution in improving computational efficiency and feature space utilization. As well as the potential advantages of using PID information to implement memory and attention mechanisms.
arXiv:2607.15732v1 Announce Type: new
Abstract: Visual grounding with multimodal large language models is commonly formulated as autoregressive coordinate generation, where a model outputs bounding-box coordinates as text given an image and a referring-expression prompt. While this interface is simple and compatible with instruction following, it introduces a mismatch between training and evaluation: training optimizes token-level likelihood over coordinate strings, whereas grounding quality is measured by geometric overlap. We propose IoUPD, an IoU-aware privileged distillation method for coordinate-generating multimodal large language models. IoUPD uses ground-truth boxes not only as coordinate targets, but also as privileged training-time guidance. During training, the student receives the original image and prompt, while a frozen teacher receives a box-marked image and an augmented prompt that indicates the marked region. The student is trained with a supervised fine-tuning anchor and a privileged distillation loss whose token weights reflect both geometric importance and teacher reliability. At inference time, IoUPD requires no box overlay, privileged hint, teacher branch, or additional prediction module. Experiments on standard referring-expression grounding benchmarks show consistent region-level improvements over strong coordinate-generating baselines, demonstrating that ground-truth boxes can provide useful privileged guidance beyond serving as coordinate labels.
arXiv:2511.20687v4 Announce Type: replace
Abstract: This paper presents a novel hybrid approach for coupling subdomain-local non-intrusive Operator Inference (OpInf) reduced order models (ROMs) with each other and with subdomain-local high-fidelity full order models (FOMs) with using the overlapping Schwarz alternating method (O-SAM). The proposed methodology addresses significant challenges in multiscale modeling and simulation, particularly the long runtime and complex mesh generation requirements associated with traditional high-fidelity simulations. By leveraging the flexibility of O-SAM, we enable the seamless integration of disparate models, meshes, and time integration schemes, enhancing computational efficiency while maintaining high accuracy. Our approach is demonstrated through a series of numerical experiments on complex three-dimensional (3D) solid dynamics problems, showcasing speedups of up to 106x compared to conventional FOM-FOM couplings. This work paves the way for more efficient simulation workflows in engineering applications, with potential extensions to a wide range of partial differential equations.
arXiv:2607.15816v1 Announce Type: new
Abstract: The rapid evolution of beyond-5G and emerging 6G networks is driving the need for flexible, reliable, and cost-efficient virtualized Radio Access Network (vRAN) architectures capable of supporting heterogeneous services such as enhanced Mobile Broadband (eMBB), Ultra-Reliable Low-Latency Communication (URLLC), and Massive Machine-Type Communication (mMTC). Future disaggregated RAN systems are expected to rely heavily on network slicing, functional split flexibility, and optical x-haul infrastructures to support stringent performance, scalability, and availability requirements. In this paper, we present an integrated framework for reliable, slice-aware, and functional split-aware Virtual Network Function (VNF) placement with lightpath provisioning in disaggregated vRAN environments. The proposed approach maximizes mobile network operators' profit by jointly optimizing function placement and optical resource allocation under latency, processing, bandwidth, and availability constraints. We formulate the problem as an Integer Linear Programming (ILP) model with two variants: one that employs unshared backups and another that uses a more cost-efficient shared backup scheme. To address ILP complexity, we develop a heuristic algorithm and a Genetic Algorithm (GA)-based metaheuristic that yields near-optimal solutions in real time. Extensive evaluations on topologies up to 128 nodes show that shared backup variants yield up to 18% higher profit, while maintaining up to 5-10% lower normalized CPU usage than unshared counterparts.
arXiv:2607.15752v1 Announce Type: new
Abstract: Personalized Image Aesthetic Assessment (PIAA) aims to predict aesthetic ratings of images that vary across individuals. The aesthetic preferences manifest to different extents across distinct visual stimuli and exhibit cohort-specific patterns. Motivated by the above fact, this paper presents a Multimodal Large Language Model (MLLM)-based approach, which models individual aesthetic preferences by Preference-Rich sample mining and Aesthetically-resonant Cohort merging (PRAC). Specifically, PRAC first identifies preference-rich samples by analyzing both Collective Controversy and Personalized Deviation of images, maximizing the utility of limited user data. Based upon the preference-rich samples, cross-user preference similarities are measured by comparing preference embeddings. Then, a cohort-based model merging strategy, is proposed by aggregating preference patterns from aesthetically-resonant users, which further enhances the personalization for the target individual. Extensive experiments and comparisons on four benchmark PIAA databases demonstrate the superiority of the proposed PRAC model over the state-of-the-arts. The code and model will be public at https://github.com/yzc-ippl/PRAC.
arXiv:2607.15755v1 Announce Type: new
Abstract: Conversational Speech Synthesis (CSS) aims to synthesize speech with human-like emotional expression and contextual consistency in user-agent interactions. Existing CSS methods struggle to render authentic human emotions due to limited predefined emotion label spaces (e.g., seven emotion categories), while redundant multimodal tokens in multi-turn dialogue history interfere with context understanding. To address these issues, we propose AuEmoChat, a CSS framework for authentic emotion understanding and rendering. First, we develop AuEmoCodec, which learns a discrete authentic emotion token space from large-scale emotional speech via finite scalar quantization, enabling a more authentic emotion representation than limited basic emotion categories. We further propose AuEmoToMe, an authentic-emotion-guided token merging algorithm that merges redundant tokens in multimodal dialogue history while preserving emotion-relevant context. We integrate it into an autoregressive text-speech model to predict the target authentic emotion token and speech tokens. Finally, we propose Authentic Emotion Flow Matching, which renders speech by jointly conditioning on merged dialogue context, target authentic emotion, and acoustic priors. Extensive experiments on the NCSSD-EmCap dataset demonstrate that AuEmoChat outperforms state-of-the-art CSS baselines and generates more expressive and authentic emotional speech.
arXiv:2607.15901v1 Announce Type: new
Abstract: Despite strong capabilities in data understanding and decision-making, autonomous data science agents still heavily rely on trial-and-error workflows that involve expensive computation. This bottleneck motivates models that can anticipate the effects of data science operations before real execution. In this paper, we introduce the concept of Data Science World Model, which model the data science execution environment by predicting environment state transitions conditioned on current workflow states and candidate operations. We further propose DSWorld, a practical framework that combines structured state construction, cost-aware routing, lightweight real execution, and an LLM-based simulator for expensive operations. To support training, we construct an 8K-scale transition trajectory dataset and introduce Reflective World Model Optimization, an error-aware reinforcement learning strategy for improving transition prediction. Experiments show that DSWorld accelerates RL-based agent training by approximately $14\times$ and search-based inference by approximately $3$-$6\times$ while maintaining competitive performance, and outperforms the strongest LLM baseline by 35.6% on transition prediction tasks. The code is available at https://anonymous.4open.science/r/DSWorld.
arXiv:2607.15977v1 Announce Type: new
Abstract: Safety defenses for large language models (LLMs) have been extensively studied, with existing approaches focusing on attack detection and refusal mechanisms. Such fixed-form direct refusal strategies may introduce the risk of prefix injection attacks. Recent work has explored a new direction that leverages humor as an indirect refusal mechanism to mitigate over-refusal in jailbreak scenarios and reduce prefix injection risks. However, this approach implicitly assumes that humorous responses are safe. Whether humorization itself introduces safety risks remains unexplored. To address this issue, we conduct an exploratory study involving over 30,000 real-world agent interaction records and 45 stand-up comedians, revealing practical safety concerns in LLM-based content humorization. Motivated by these findings, we propose \textsc{HumorSafe}, a novel framework for evaluating latent safety risk propagation during humorization. \textsc{HumorSafe} enables LLMs to learn harmful humorization patterns and use them to transform benign content into humorous content with safety risks. Across five frontier LLMs, we find that LLMs can introduce stereotypes and toxicity during humorization. We further propose \textsc{HumorPIA}, a prompt injection attack that exploits latent risks in humor-based defenses. \textsc{HumorPIA} preserves the appearance of safe humorous refusal while covertly injecting harmful content, allowing latent risks to evade existing detection mechanisms. Experiments show that it increases toxicity by 3.14$\times$ while maintaining an apparent safety rate of 97.8\% even under defense settings. Our findings highlight a gap in existing LLM safety evaluations under humorized settings.