arXiv:2606.03723v1 Announce Type: new
Abstract: Low-rank adaptation (LoRA) enables parameter-efficient specialization of foundation models, but the proliferation of task-specific adapters fragments capabilities across many adapters, complicating reuse and deployment. We study the problem of merging $T$ LoRAs into a single rank-$r$ LoRA, thereby preserving the benefits of low-rank structure. Existing Merge-then-Compress pipelines treat the rank constraint as an afterthought: they merge adapters in the full parameter space, then compress the merged result to rank $r$ via truncated SVD. However, full-parameter merging may destroy the low-rank structure, making it difficult for subsequent compression to recover an effective rank-$r$ LoRA. We propose Compress-then-Merge (CtM), a reversed pipeline that enforces the rank-$r$ bottleneck before merging: CtM computes shared $r$-dimensional subspaces using only the LoRA weights to capture cross-adapter common structure, projects each adapter into the shared subspaces to obtain $r\times r$ coordinates, and then applies standard merging rules in this reduced space. CtM guarantees a rank-$r$ LoRA by construction, avoiding post-hoc truncation, and enables efficient computation in the core space spanned by concatenated LoRA factors. Experiments across multiple models and tasks show that CtM consistently outperforms existing single-LoRA-output baselines while narrowing the performance gap to full-parameter merging methods.
Science Journals
arXiv:2606.03717v1 Announce Type: new
Abstract: Data-driven approaches for learning power flow models suffer from weak generalization across varying network topologies and limited computational scalability. Existing methods typically rely on training over a large set of grid topologies, which becomes impractical for large networks. This paper proposes a scalable and computationally efficient framework for topology-adaptive learning of power flow solutions. We propose a modular architecture consisting of bus-level Gaussian Process (GP) models, where each GP collects local features based on bus-level \textit{egonet} definition. The localized bus-level feature includes first-order power and admittance sensitivities, nodal injections and node degree. In addition to the modular architecture, we propose using Random Fourier Features (RFF) for feature reduction, which further enhances the computational scalability. We evaluate the effectiveness of the proposed method by simulations across multiple benchmark networks under N-1, N-2, and N-3 contingencies. Results for the PEGASE 1354 bus system under N-3 contingencies demonstrate high predictive quality, with an $R^2$ score of 0.983 and a voltage-magnitude RMSE of 0.0023 p.u. The framework maintains recall rates exceeding 98\% for detecting voltage limit violations across all test cases. Furthermore, the approach exhibits scalability, completing training and testing for the PEGASE 1354 system in 116.47 seconds while outperforming existing benchmarks in zero-shot generalization without requiring additional training samples.
arXiv:2606.03715v1 Announce Type: new
Abstract: Text-to-image models rely on text prompts as their primary interface to human intent. Prompts are encoded by a text encoder into embeddings that condition the image generation process. Beyond individual token meanings, text embeddings encode contextual information across the full prompt, such as compositionality and attribute binding. However, whether image models actually exploit this richer information remains underexplored. Here, we address the question: Which aspects of text representation are essential for image generation? We show that text-to-image diffusion transformer-based models commonly rely only on two relatively straightforward aspects of text representations: (i) the merging of adjacent tokens into a word representation, for words spanning multiple tokens, and (ii) word order, which is imprinted by the positional embedding of the text-encoder. To show this, we construct a new text embedding that encodes only individual word meanings and order but lacks any contextual information about the full prompt. We find that this bag of position-tagged words representation is sufficient to successfully guide image generation, achieving visual quality and text fidelity that are on par with full text embedding-guided generation. This demonstrates that, contrary to common belief, text-to-image models often do not use the rich information encoded in the text embedding beyond individual word meanings and word order. Instead, the decoding of complex linguistic structures is performed by the image model itself. Project webpage: https://nsping13.github.io/contextless-TTI/
arXiv:2606.03206v1 Announce Type: new
Abstract: Frequency upconversion technique, where the infrared signal is nonlinearly translated into the visible band to leverage the silicon sensors, offers a promising alternation for the mid-infrared (MIR) imaging. However, the intrinsic field of view (FOV) is typically limited by the phase-matching condition, thus imposing a remaining challenge to promote subsequent applications. Here, we demonstrate a wide-field upconversion imaging based on the aperiodic quasi-phase-matching configuration. The acceptance angle is significantly expanded to about 30$^\circ$, over tenfold larger than that with the periodical poling crystal. The extended FOV is realized in one shot without the need of parameter scanning or post-processing. Consequently, a fast snapshot allows to facilitate high-speed imaging at a frame rate up to 216 kHz. Alternatively, single-photon imaging at room temperature is permitted due to the substantially suppressed background noise by the spectro-temporal filtering. Furthermore, we have implemented high-resolution time-of-flight 3D imaging based on the picosecond optical gating. These presented MIR imaging features with wide field, fast speed, and high sensitivity might stimulate immediate applications, such as non-destructive defect inspection, in-vivo biomedical examination, and high-speed volumetric tomography.
arXiv:2606.02883v1 Announce Type: new
Abstract: Recommender systems have grown from content-organization tools into sophisticated systems that shape daily behavior. By controlling what we see, they shape what we perceive, raising concerns about filter bubbles, radicalization, polarization, and social inequality. Large language models (LLMs) enable more powerful personalization, intensifying these dynamics. Yet most recommenders are tuned for engagement or limited accuracy metrics, with little attention to broader social implications, e.g. how personalization reshapes exposure in socially consequential domains. We investigate whether LLM-assisted reranking, while improving personalization, inadvertently amplifies exposure to ideologically extreme or conspiratorial political content, a risk theorized but not empirically characterized in news recommendation. Using real news-consumption histories, we rerank YouTube's sidebar candidates through zero-shot, instruction-based prompting. We compare a baseline prompt with a constrained variant that preserves topical relevance and broadens ideological exposure while reducing conspiratorial or extreme content. Without constraints, reranking strengthened personalization but increased exposure to conspiratorial and extremist material for users whose histories contained such content. Lightweight prompt-level regularization reduced promotion of extreme content and increased ideological diversity, with modest relevance loss. Synthetic experiments suggest that LLMs rerank via statistical regularities in language rather than semantic understanding of ideology, clarifying why naive prompts amplify these patterns and why regularization can reshape them. Together, our results highlight the power of LLMs to operationalize contextual nuance in high-stakes recommendation, and the need to evaluate LLM-assisted personalization beyond accuracy and treat prompt design as a value-laden rather than neutral default.
arXiv:2606.03587v1 Announce Type: new
Abstract: Many proof-of-stake protocols finance validator rewards from two sources: transaction fees and a finite reserve of tokens. This creates a dynamic hand-off problem. Early in the life of the system, fees may be too small to fund the target level of security; later, fees may become sufficient. The central question is whether the reserve provides enough runway for the protocol to remain secure until this fee-only region is reached. We study this problem in a discrete-time stochastic model of validator participation. Token price and transaction demand fluctuate over time, while validators choose participation strategically. We solve the validator entry game and derive an exact state-dependent reserve threshold, i.e., the minimal reserve stock necessary and sufficient to sustain a target security level. This threshold separates three regions: infeasibility, reserve-dependent security, and fee-only security. Security fails if the reserve first falls below the state-dependent threshold, and a successful hand-off occurs exactly if the fee-only region is reached before that failure time. We derive stress-test guarantees that convert lower confidence bands for token price and demand into reserve requirements, and obtain explicit failure-probability and expected hand-off-time bounds. Finally, we extend the model to forward-looking validators and derive the Markov participation condition that captures how current participation affects future reserve-funded rewards. The main implication is that reserve policy should not be evaluated by nominal depletion dates or steady-state reward ratios alone. A protocol can have a large nominal reserve and still be close to security failure after adverse price or demand shocks. Conversely, once demand crosses the fee-only threshold, the reserve becomes redundant for security. This paper provides a tractable equilibrium framework for stress-testing this transition.
arXiv:2606.03581v1 Announce Type: new
Abstract: Unstructured scenes present unique challenges for autonomous driving, as irregular obstacles and sparse scene layouts undermine the effectiveness of traditional perception methods such as 3D object detection. 3D semantic occupancy prediction has emerged as a prominent focus due to its ability to provide dense spatial representations by assigning semantic labels to individual voxels in 3D space. However, directly applying 3D semantic occupancy prediction to unstructured scenes remains challenging because scene sparsity hinders effective cross-modal fusion and the more severe long-tail distribution in these scenarios further degrades prediction performance. To validate the effectiveness of our approach, we construct a dedicated dataset of unstructured scenes collected from open-pit mines. Based on this, we propose UnsOcc, a multi-modal 3D semantic occupancy prediction framework that improves robustness in unstructured environments. At its core, we introduce a rendering-based fusion module, RenderFusion, which enhances cross-modal feature alignment through bidirectional rendering supervision. Furthermore, we propose GSRefinement, a detail-aware auxiliary supervision method based on Gaussian Splatting that projects sparse 3D occupancy predictions into dense 2D semantic segmentation maps, enabling effective supervision for long-tail categories. Extensive experiments on both the open-pit mine dataset and the nuScenes dataset demonstrate that our method significantly outperforms existing state-of-the-art approaches.
arXiv:2606.03556v1 Announce Type: new
Abstract: Vision-language-action (VLA) models are gaining attention in robotics, yet their robustness to adversarial attacks remains largely unexplored. Existing work shows that adversarial patches can mislead VLA-based robots but assumes full access to the entire execution trajectory, an unrealistic requirement in practice. We address this limitation by formulating a partially observable threat model, where the adversary can exploit only a short prefix of the trajectory to generate a fixed patch applied to all subsequent frames. Under this setting, we propose a two-phase framework. First, we localize the patch using the model's attention maps to identify visually critical regions that correspond to the full instruction. Then, we optimize the patch to disrupt the semantic grounding of target objects and increase the curvature of action trajectories, thereby compounding failures in both perception and control. Extensive experiments in simulation and real-world robotic environments show that our method sustains adversarial effects under partial observability, inducing long-horizon disruptions and significantly reducing task success rates.
arXiv:2606.02900v1 Announce Type: new
Abstract: Transonic flutter is a detrimental aeroelastic instability that can generate large-amplitude structural oscillations, leading to severe vibration, fatigue damage, reduced operational limits, and potentially catastrophic structural failure. Incoming wake disturbances can further amplify this instability, making it critical to identify the underlying aerodynamic mechanisms responsible for predicting and controlling flutter onset. The underlying flow physics is complex with nonlinear interactions between the wake and the wing, shock motion, shock-induced flow separation, vortex shedding and the wing motion. In this study, we perform high-fidelity direct numerical simulations of a sinusoidally pitching NACA0012 airfoil with an underwing cylinder at various transonic Mach numbers and a Reynolds number of 10,000. Through energy maps, we identify that the addition of the cylinder significantly expands flutter boundaries compared to an airfoil-only system. We extend the force partitioning method to partition the power transferred between the flow and the airfoil for compressible flows. Application of this approach to distinct regions of the flow domain indicates that the gap flow between the wing and the cylinder is the dominant contributor to the energy transfer from flow to the wing. The blockage effects due to the cylinder cause flow acceleration on the wing which further enhances the tendency for flutter. We investigate cylinder placement relative to the airfoil to reveal that flutter is enhanced only when the cylinder is placed upstream of the pivot point on the airfoil. The current study highlights how such partitioning methods can parse force and energy transfer mechanisms in complex, unsteady high-speed flows.
arXiv:2606.02933v1 Announce Type: new
Abstract: Generative AI (GenAI) is becoming a widely adopted learning support tool for both students and instructors, as it offers benefits such as personalized tutoring and scaffolded learning. However, recent research highlights potential drawbacks such as overreliance and metacognitive issues, especially in novice programmers. Most prior work focuses on introductory programming courses, and important questions remain about the underlying mechanisms behind the negative effects of GenAI and if findings can be generalized when students learn more advanced computer science concepts. To address this gap, we conducted a mixed-methods study comparing student interactions with GenAI to two traditional learning supports in a second-year algorithms course: algorithm visualization (AV) and human live tutoring (LT). Twelve students participated in three 90-minute study sessions focusing on sorting, tree, and graph algorithms. We recorded gaze and interaction data, and each session concluded with a test assessing their conceptual understanding of the topic. Our analysis classifies when during the problem-solving process participants sought help, and compares the interaction patterns across the three learning supports. Although GenAI produced a larger increase in self-efficacy compared to live tutoring, it was associated with noticeably lower results in learning outcomes. We found that participants did not use algorithm visualizations effectively, faced usage barriers when using GenAI to learn advanced topics, and that live tutoring yielded the highest learning outcomes.
arXiv:2606.03555v1 Announce Type: new
Abstract: A surface geometry consisting of parallel, meandering streamwise grooves has been experimentally studied as an alternative means of passive transverse forcing of turbulent boundary-layer flow. Contrary to the original expectation, the flow does not exhibit a spanwise-uniform undulation aligned with the grooves; instead, a converging-diverging flow pattern results. This flow pattern can be attributed to the spanwise periodicity of the lateral pressure gradient. The forcing effect is found to initially increase with the groove amplitude, but it saturates when the groove slope becomes too steep. The observed induced flow, referred to as a Passive Stokes Layer (PSL), can be considered as being composed of an inertial (pressure-driven) outer solution generated by the displacement effect of the non-smooth surface geometry, and a viscous inner solution to accommodate the no-slip condition at the wall. The mechanism of transverse flow generation is elucidated by an inviscid flow model that relates the forcing to the surface geometric properties, with predictions in good agreement with the experimental results. Although a reduction in the near-wall turbulence levels over the groove surfaces is observed, no direct evidence for (mean) drag reduction is evident from the data. Instead, an estimate of the frictional drag potential is based on establishing a tentative relation to an equivalent spatial Stokes layer (SSL) induced by active wall forcing. This theoretical comparison indicates that the induced passive forcing is sufficient to act on the (active) spanwise forcing mechanism, but produces at most a few per cent of frictional drag reduction. Any potential savings are likely offset by pressure drag and other losses, so that, similar to active forcing, its potential for net drag reduction in practical applications is limited.
arXiv:2606.02604v1 Announce Type: new
Abstract: ESG and climate risk data remain fragmented across heterogeneous Scope 1, Scope 2, and Scope 3 reporting environments, while conventional validation pipelines lack provenance aware auditability, hidden drift detection, and reproducibility oriented governance. This paper proposes a deterministic climate risk intelligence framework integrating single source of truth orchestration, temporal anomaly detection, imbalance aware ensemble learning, and explainability oriented governance for auditable ESG validation. To support open reproducibility, we construct and release a synthetic ESG validation benchmark calibrated against publicly reported characteristics of the GHG Protocol, PCAF, and ISSB standards. The methodology incorporates temporal drift analysis, SMOTE based rare event optimization, ensemble learning, provenance aware orchestration, and TreeSHAP based interpretability for governance inspection and audit reconstruction. We evaluate the framework against statistical classifiers, anomaly detection methods, temporal forecasting baselines, and a threshold based system using classification metrics (recall, F1, ROC AUC), calibration metrics (ECE, Brier score), and a governance oriented audit trace completeness metric measuring the fraction of flagged anomalies for which a deterministic source to escalation provenance chain can be reconstructed. Results are reported as mean and standard deviation across stratified five fold cross validation with paired significance testing. The framework reframes ESG reporting toward deterministic climate risk governance infrastructure supporting reproducibility, explainability, and operational auditability.
arXiv:2606.03551v1 Announce Type: new
Abstract: Simulation has become a core infrastructure for robotics research. Unlike previous simulators, NVIDIA Isaac Sim leverages GPU acceleration to enable large-scale parallel training and physics-accurate modeling. Its synthetic data generation pipeline alleviates the scarcity of high-quality training data, supporting data-driven robot learning and large-scale simulation-centric experimentation. However, existing surveys often treat it as one simulator among many, without a systematic analysis of its architectural characteristics, usage patterns, and limitations. This survey reviews Isaac Sim from system and application perspectives, outlining its architecture and comparing it with widely used simulators. We analyze representative studies across five major domains and summarize common usage patterns, particularly in data generation and high-fidelity simulation. We also outline key future directions and challenges, including physics open-world learning, simulation-centric training and practical usability constraints.
arXiv:2606.01111v2 Announce Type: replace
Abstract: Modern industrial recommender systems rely on thousands of heterogeneous features -- ranging from low-dimensional scalars (e.g., statistical value) to high-dimensional embeddings (e.g., user-id embeddings, MLP representations) -- to achieve high-precision predictions. Given the immense computational costs associated with training, efficient feature selection is critical. However, existing methods encounter three primary bottlenecks: (1) they typically assume uniform feature dimensions or require costly mapping to a fixed size; (2) they struggle with extreme sparsity, where the majority of features (e.g., 99%+) remain at default values; and (3) traditional permutation-based approaches are computationally prohibitive in large-scale settings.
To address these challenges, we propose LeAP (Learnable Adaptive Permutation), a novel, model-agnostic plug-in module for feature selection. LeAP transforms the inefficient random permutation process into a learnable mechanism, significantly accelerating the evaluation of feature importance. In addition, we introduce an adaptive regularization strategy tailored for heterogeneous dimensions and extreme sparsity, enabling superior feature importance ranking results across asymmetric input spaces. Experiments on four public recommendation datasets demonstrate that LeAP achieves state-of-the-art performance. Furthermore, LeAP has been deployed in a large-scale industrial search ranking model with over a billion daily requests and a 2TB model parameter scale. In this real-world scenario involving 12,000+ total feature dimensions, LeAP successfully identified and removed over 3,600 redundant dimensions without performance degradation, which is 2 to 10 times the ability of compared baseline methods.
arXiv:2505.16014v5 Announce Type: replace
Abstract: Retrieval-Augmented Generation (RAG) systems deployed in sensitive domains must provide interpretable evidence selection and robust safeguards against data poisoning, yet current approaches rely on opaque similarity-based retrieval with arbitrary top-k cutoffs that offer no explanation for their selections and remain vulnerable to adversarial manipulation. METEORA replaces re-ranking with rationale-driven selection via three components: a DPO-tuned LLM that generates explicit retrieval rationales, an Evidence Chunk Selection Engine (ECSE) that uses those rationales with statistical elbow detection for adaptive cutoff determination, and a Verifier LLM that filters poisoned evidence using the same rationales. Across six datasets, METEORA achieves 13.41% higher recall, 21.05% higher precision (without expansion), an 80% reduction in evidence volume, a 33.34% improvement in answer accuracy, and a 4.4x improvement in adversarial robustness. Human evaluation confirms genuine interpretability (3.64/5 confidence; 86% ground-truth agreement), demonstrating that interpretability, efficiency, and robustness are synergistic rather than competing objectives. The code is available in the GitHub repository https://github.com/YashSaxena21/METEORA
arXiv:2606.02607v1 Announce Type: new
Abstract: Tabular synthesis is critical for privacy-preserving sharing and augmentation, yet diffusion models rely on implicit mechanisms to capture inter-column relationships. We introduce Geometry-Aware Tabular Diffusion (GATD), which augments tabular diffusion denoisers with pairwise angles and lengths computed from column value differences and used as inputs and auxiliary targets. Our MLP instantiation achieves state-of-the-art benchmark performance while using 3.5x fewer parameters on average (up to 25x for classification tasks): on ten datasets, it wins 8/10 Shape, 7/10 Trend, and 9/10 downstream utility (F1/RMSE), reducing Shape and Trend error by 27% and 20%. Default loss weights transfer to GNN and Transformer denoisers, improving Shape on 27/30 and Trend on 25/30 architecture-dataset cells. A matched ablation shows supervision (not extra inputs or capacity) drives the gain. This shows explicit relational supervision is a portable inductive bias for tabular diffusion.
arXiv:2606.02852v1 Announce Type: new
Abstract: Accurate short-term forecasting of residential energy load and indoor temperature is essential for home energy management systems, grid-level demand response, and community energy efficiency efforts. Domain adaptation and transfer learning have shown promise for improving forecasting accuracy under data heterogeneity and scarcity commonly seen in residential settings. However, progress is limited by the lack of comprehensive residential datasets: existing benchmarks are narrow in target coverage and rarely support structured cross-domain evaluation. We introduce RESCAST-100K, a large-scale residential forecasting benchmark for studying cross-domain generalization. It provides a configuration-driven interface that instantiates source and target domains along interpretable axes, including geography, climate zone, wall construction, and heating equipment, enabling systematic evaluation of transfer learning, domain adaptation, and zero-shot generalization under controlled domain shifts. The benchmark covers approximately 100,000 EnergyPlus-simulated U.S. homes derived from ResStock, with 15-minute time series for three coupled targets per home: total load, HVAC load, and indoor temperature. These are paired with weather channels, HVAC setpoints, and over 40 static building covariates. RESCAST-100K also integrates five real-world residential datasets under a unified schema, supporting sim-to-real evaluation on the same tasks. We benchmark recurrent, attention-based, and MLP-mixer architectures for zero-shot performance across domains, missing-input conditions, and forecasting tasks. Cross-attention and MLP-mixer models consistently outperform recurrent and classical transformer baselines under domain shift. RESCAST-100K is intended to aid the machine learning and building analytics communities advance cross-domain residential forecasting at home, community, and grid scale.
arXiv:2606.02610v1 Announce Type: new
Abstract: Ocean general circulation models (OGCMs) are essential to climate science but computationally expensive, limiting ensemble size and forcing scenarios. Neural emulators promise orders-of-magnitude speedups, yet existing ocean emulators have not combined fine spatial resolution with multi-year autoregressive rollouts. Samudra, the first autoregressive neural ocean emulator to produce multi-decade global rollouts, is limited to $1^\circ$ resolution and exhibits two long-horizon failure modes: \emph{variance collapse}, the loss of temporal variability, and \emph{imprinting artifacts}, in which velocity patterns leak into deep-ocean fields. We present Samudra 2, which introduces a wider U-Net backbone with modified ConvNeXt-style blocks and a reduced block-internal expansion factor, together with a dynamic loss that reweights output channels according to their prediction errors, strengthening gradients for slow-evolving deep-ocean fields. At $1^\circ$, Samudra 2 increases upper-ocean global-mean temperature $R^2$ from 0.56 to 0.87 and reduces deep-ocean temperature error by roughly sevenfold. The same architecture scales to $1/2^\circ$ and $1/4^\circ$ over approximately 8-year autoregressive rollouts, recovering mesoscale eddies and sharp western boundary currents. Running on a single GPU, Samudra 2 enables larger ensembles for sea-level projections, ocean heat uptake, and climate variability studies. We provide code, documentation, and benchmark resources at https://openathena.ai/Ocean_Emulator/.
arXiv:2606.02613v1 Announce Type: new
Abstract: The discrimination of neutron and gamma pulse shapes is a key technology in fields such as nuclear safety monitoring and radiation assessment. An enhanced recursive gated cyclic residual-sparse linear attention network is developed on the CLYC detector experimental platform to overcome weak noise resistance, limited feature extraction and inferior real-time performance of conventional algorithms. The experimental dataset comprises 19,971 samples, which were pre-processed and stratified for model training and testing. Results indicate that the proposed algorithm achieves a quality factor of 2.2, with a classification accuracy of 98.7% and a recall rate of 99.4%. It achieves an accuracy of 95.1% under the 20 dB low signal-to-noise ratio condition, exhibiting excellent anti-noise ability.With around 2.8 million parameters, the model takes merely 0.05 ms to process a single pulse on GPU, satisfying real-time monitoring and embedded deployment demands.
arXiv:2606.02623v1 Announce Type: new
Abstract: Solving time-dependent partial differential equations (PDEs) is an important problem in computational science and engineering. Physics-informed neural networks (PINNs) learn PDE solutions from governing equations. However, accurately capturing temporal evolution remains challenging. Recent sequence-model-based approaches parameterize time evolution using general-purpose sequence models, which capture temporal dependencies but do not explicitly encode the structured dynamics of PDE solutions. In addition, their memory requirements can scale unfavorably with sequence length and resolution, limiting applicability in large-scale or high-dimensional settings. This work introduces a PINN approach that incorporates oscillatory state-space dynamics to represent the modal structure of PDE solutions. The proposed method leverages a linear-oscillator-based temporal evolution, together with a PDE-aware spectral basis in space. This design enables closed-form spatial differentiation and consistent enforcement of boundary conditions. The method is evaluated on forward, inverse, and high-dimensional PDE problems, including cases up to 100 spatial dimensions. The results show improved accuracy and reduced memory usage compared to recent sequence-model-based PINN approaches. Overall, this work highlights the benefits of incorporating structured dynamical priors into the temporal evolution of neural PDE solvers and suggests designing more physics-aligned and computationally efficient PINN architectures.
arXiv:2606.02628v1 Announce Type: new
Abstract: We investigate whether open-source LLMs encode a linearly separable truthfulness signal in their hidden states, and at which network depth this signal is strongest. Across three $7$B--$8$B instruction-tuned models (Llama-3.1-8B, Mistral-7B, Qwen2.5-7B) loaded in $4$-bit NF4 quantization, we extract per-layer hidden states on four hallucination benchmarks (TruthfulQA, HaluEval-QA, FEVER, and a controlled synthetic set) and compare four detection approaches: linear and MLP probes, INSIDE EigenScore, self-consistency, and attention entropy. A linear probe on a single mid-network layer achieves $0.904$--$1.000$ AUROC on held-out splits, while sampling-based detectors do not exceed $0.541$ AUROC under the same protocol. The truthfulness signal is approximately linear: MLP probes rarely surpass linear probes by more than $0.01$ AUROC. Peak probing layers fall in a consistent band across model families on natural-language benchmarks -- blocks~$13$--$18$ of~$32$ for Llama and Mistral, and blocks~$19$--$25$ of~$28$ for Qwen. First-block attention entropy provides a complementary signal in knowledge-grounded settings ($0.866$--$0.941$ AUROC on HaluEval-QA) at no additional inference cost. The low discriminability of sampling methods under this protocol reflects a structural mismatch between paired-label evaluation and the information these methods access, rather than an inherent limitation of those methods. Code and data are released for full reproducibility on a single $8$\,GB GPU.
arXiv:2606.02638v1 Announce Type: new
Abstract: Recent advances in neural song generation have enabled high-quality synthesis from lyrics and global textual prompts. However, most systems fail to model temporally varying attributes of songs, severely limiting fine-grained control over musical structure and dynamics. To address this, we propose SegTune, a Diffusion Transformer-based framework enabling structured and fine-grained controllability by allowing users or large language models (LLMs) to specify local musical descriptions aligned to song segments. These segment prompts are temporally broadcast to corresponding time windows, while global prompts ensure stylistic coherence. To support precise lyric-to-music alignment, we introduce an LLM-based duration predictor that autoregressively generates sentence-level timestamps in LyRiCs format. We further construct a large-scale data pipeline for high-quality song collection with aligned lyrics and prompts, and propose new metrics to evaluate segment alignment and vocal consistency. Experiments demonstrate that SegTune outperforms existing baselines in both musicality and controllability. Visit our project page (https://github.com/KlingAIResearch/SegTune) for codes and more generated songs.
arXiv:2606.02630v1 Announce Type: new
Abstract: Patient-facing medical chatbots are commonly evaluated on single-turn prompts, yet real users push back after refusals, add urgency, and invoke authority. We introduce MultiTurnPSB, a four-turn adversarial extension of PatientSafetyBench, and evaluate GPT-4.1-mini under fixed template, template-adaptive, and live adversarial attacks. Unsafe responses rise from 35% to nearly 80% by Turn 4 under live attack. Under the same adversary, GPT-4.1-mini and Claude Sonnet 4.5 are statistically indistinguishable at baseline but diverge to a 19x gap by Turn 4, a difference invisible to single-turn evaluation. We characterize four degradation trajectory signatures and identify a two-element attack formula responsible for most catastrophic failures. A lightweight input-side classifier reduces Turn 4 unsafe responses by 52 percentage points despite severe accuracy degradation, but the 45% false alarm rate on benign queries is the primary deployment constraint. A methodological finding also emerges: Claude Sonnet refused to generate adversarial messages in over half of late-turn conversations despite explicit red team framing, suggesting safety training may generalize to the attacker role.
arXiv:2606.02663v1 Announce Type: new
Abstract: Recent advances in machine learning have produced probabilistic weather forecasting models comparable to state-of-the-art numerical weather predictors. But no model consistently dominates spatio-temporally, and relative performance is highly context-dependent. This motivates adaptive methods for combining multiple forecasts to obtain improvements and robustness. While combined forecasts have been proposed in the literature, these are achieved either through supervised learning or through prediction with expert advice methods. We introduce AdaWeather, an adaptive framework that combines many probabilistic forecasts using both machine learning as well as mixture of experts to arrive at a unified improved probabilistic forecast. While traditional expert methods develop the regret bounds with respect to the best single expert in hindsight, we extend the algorithm and analysis to show our method has logarithmic regret compared to the best static mixture of experts in hindsight. Empirically, we focus on forecasting temperature, and observe improvements over existing methods.
arXiv:2606.03539v1 Announce Type: new
Abstract: Spatio-Temporal Video Grounding aims to localize object tubes based on textual queries. While recent methods have achieved remarkable success, they mainly focus on high-quality(HQ) inputs, neglecting the widespread presence of low-quality(LQ) videos in real-world scenarios. Although tuning methods like LoRA can adapt to degraded inputs, they inevitably disrupt pre-trained knowledge. To address this, we propose Null-Space Tuning (NST). This framework exploits the geometric property that adding vectors within the null-space of frozen weights to the layer input does not affect the output. Leveraging this, NST injects learnable residuals into input features that can be selectively invisible to the pre-trained backbone. Specifically, NST combines the Quality-Adaptive Unit and Dual-Space Reparameterization to synthesize these residuals by confining components for HQ inputs to the null-space, while directing restoration components for LQ inputs to the non-null space. As the frozen weights eliminate null-space components, we effectively rectify degraded inputs while preserving pre-trained knowledge for HQ inputs. Extensive experiments show that NST outperforms state-of-the-art methods on our Mixed-Quality benchmark.