arXiv:2509.09658v2 Announce Type: replace
Abstract: Hallucinations in multimodal large language models (MLLMs) -- where the model generates content inconsistent with the input image -- pose significant risks in real-world applications, from misinformation in visual question answering to unsafe errors in decision-making. Existing benchmarks primarily test recognition accuracy, i.e., evaluating whether models can select the correct answer among distractors. This overlooks another important capability for trustworthy AI: recognizing when none of the provided options is supported by the image and abstaining from committing to a false choice, a humility-related behavior. We present HumbleBench, a new hallucination benchmark designed to evaluate false-option rejection in MLLMs under a forced-choice multiple-choice setting with a ``None of the above'' option. Built from a panoptic scene graph dataset, we leverage fine-grained scene graph annotations for objects and relations, use candidate attribute cues, and prompt GPT-4-Turbo to generate multiple-choice questions, followed by a rigorous manual filtering process. Each question includes a ``None of the above'' option, requiring models not only to recognize correct visual information but also to identify when no provided answer is valid. We evaluate a variety of state-of-the-art MLLMs -- including general-purpose, specialized reasoning, and proprietary models -- on HumbleBench and report empirical findings for the community. By incorporating explicit false-option rejection, HumbleBench fills a key gap in current evaluation suites by assessing a narrower but important abstention-oriented behavior that is relevant to trustworthy multimodal reasoning. Our code and dataset are released publicly and can be accessed at \href{https://github.com/maifoundations/HumbleBench}{https://github.com/maifoundations/HumbleBench}.
Science Journals
arXiv:2509.10452v2 Announce Type: replace
Abstract: Pretrained automatic speech recognition (ASR) models such as Whisper perform well but still need domain adaptation to handle unseen parlance. In many real-world settings, collecting speech data is impractical, necessitating text-only adaptation. We propose WhisTLE, a deeply supervised, text-only adaptation method for pretrained encoder-decoder ASR models. WhisTLE trains a variational autoencoder (VAE) to model encoder outputs from text and fine-tunes the decoder using the learned text-to-latent encoder, optionally combined with text-to-speech (TTS) adaptation. At inference, the original encoder is restored, incurring no extra runtime cost. Across four datasets and four ASR models, WhisTLE with TTS reduces word error rate (WER) by a relative 49.0% and outperforms all non-WhisTLE baselines in 100 of 112 scenarios. We also find that WhisTLE additively complements any combination of other domain adaptation approaches; we thus recommend the inclusion of WhisTLE during standard processes for adapting encoder-decoder ASR models.
arXiv:2509.12142v3 Announce Type: replace
Abstract: This paper investigates an information-theoretic model of secure semantic-aware communication. For this purpose, we consider the lossy joint source-channel coding (JSCC) of a memoryless semantic source transmitted over a memoryless wiretap channel. The source consists of two correlated parts that represent semantic and observed aspects of the information. Our model assumes separate fidelity and secrecy constraints on each source component and, in addition, encompasses two cases for the source output, in order to evaluate the performance gains if the encoder has an extended access to the source. Specifically, in Case 1, the encoder has direct access only to the samples from a single (observed) source component, while in Case 2 it has additional direct access to the samples of the underlying semantic information. We derive single-letter converse and achievability bounds on the rate-distortion-equivocation region. The converse bound explicitly contains rate-distortion functions, making it easy to evaluate, especially for some common distributions. The proposed achievability coding scheme involves novel stochastic superposition coding with two private parts to enable analysis of the equivocation for each source component, separately. Our results generalise some of the previously established source and source-channel coding problems. The general results are further specialised to Gaussian and Bernoulli sources transmitted over Gaussian and binary wiretap channels, respectively. The numerical evaluations illustrate the derived bounds for these distributions.
arXiv:2603.28128v2 Announce Type: replace
Abstract: Although Graph Neural Networks (GNNs) have shown promise for smart contract vulnerability detection, they still face significant limitations. Homogeneous graph models fail to capture the interplay between control flow and data dependencies, while heterogeneous graph approaches often lack deep semantic understanding, leaving them susceptible to adversarial attacks. Moreover, most black-box models fail to provide explainable evidence, hindering trust in professional audits. To address these challenges, we propose ORACAL (Observable RAG-enhanced Analysis with CausAL reasoning), a heterogeneous multimodal graph learning framework that integrates Control Flow Graph (CFG), Data Flow Graph (DFG), and Call Graph (CG). ORACAL selectively enriches critical subgraphs with expert-level security context from Retrieval-Augmented Generation (RAG) and Large Language Models (LLMs), and employs a causal attention mechanism to disentangle true vulnerability indicators from spurious correlations. For transparency, the framework adopts PGExplainer to generate subgraph-level explanations identifying vulnerability triggering paths. Experiments on large-scale datasets demonstrate that ORACAL achieves state-of-the-art performance, outperforming MANDO-HGT, MTVHunter, GNN-SC, and SCVHunter by up to 39.6 percentage points, with a peak Macro F1 of 91.28% on the primary benchmark. ORACAL maintains strong generalization on out-of-distribution datasets with 91.8% on CGT Weakness and 77.1% on DAppScan. In explainability evaluation, PGExplainer achieves 32.51% Mean Intersection over Union (MIoU) against manually annotated vulnerability triggering paths. Under adversarial attacks, ORACAL limits performance degradation to approximately 2.35% F1 decrease with an Attack Success Rate (ASR) of only 3%, surpassing SCVHunter and MANDO-HGT which exhibit ASRs ranging from 10.91% to 18.73%.
arXiv:2509.12196v2 Announce Type: replace
Abstract: Standard attention mechanisms in transformers employ static token representations that remain unchanged across all pair-wise computations in each layer. This limits their representational alignment with the potentially diverse relational dynamics of each token-pair interaction. While they excel in domains with relatively homogeneous relationships, standard attention's static relational learning struggles to capture the diverse, heterogeneous inter-channel dependencies of multivariate time series (MTS) data--where different channel-pair interactions within a single system may be governed by entirely different physical laws or temporal dynamics. To better align the attention mechanism for such domain phenomena, we propose attention with dynamic relational priming (prime attention). Unlike standard attention where each token presents an identical representation across all of its pair-wise interactions, prime attention tailors each token dynamically (or per interaction) through learnable modulations to best capture the unique relational dynamics of each token pair, optimizing each pair-wise interaction for that specific relationship. This representational plasticity of prime attention enables effective extraction of relationship-specific information in MTS while maintaining the same asymptotic computational complexity as standard attention. Our results demonstrate that prime attention consistently outperforms standard attention across benchmarks, achieving up to 6.5\% improvement in forecasting accuracy. In addition, we find that prime attention achieves comparable or superior performance using up to 40\% less sequence length compared to standard attention, further demonstrating its superior relational modeling capabilities.
arXiv:2509.12672v2 Announce Type: replace
Abstract: The volume of machine-generated content online has grown dramatically due to the widespread use of Large Language Models (LLMs), leading to new challenges for content moderation systems. Conventional content moderation classifiers, which are usually trained on text produced by humans, suffer from misclassifications due to LLM-generated text deviating from their training data and adversarial attacks that aim to avoid detection. Present-day defence tactics are reactive rather than proactive, since they rely on adversarial training or external detection models to identify attacks. In this work, we aim to identify the vulnerable components of toxicity classifiers that contribute to misclassification, proposing a novel strategy based on mechanistic interpretability techniques. Our study focuses on fine-tuned BERT and RoBERTa classifiers, testing on diverse datasets spanning a variety of minority groups. We use adversarial attacking techniques to identify vulnerable circuits. Finally, we suppress these vulnerable circuits, improving performance against adversarial attacks. We also provide demographic-level insights into these vulnerable circuits, exposing fairness and robustness gaps in model training. We find that models have distinct heads that are either crucial for performance or vulnerable to attack and suppressing the vulnerable heads improves performance on adversarial input. We also find that different heads are responsible for vulnerability across different demographic groups, which can inform more inclusive development of toxicity detection models.
arXiv:2509.13608v2 Announce Type: replace
Abstract: As Large Multimodal Models (LMMs) become integral to daily digital life, understanding their safety architectures is a critical problem for AI Alignment. This paper presents a systematic analysis of OpenAI's GPT-4o mini, a globally deployed model, on the difficult task of multimodal hate speech detection. Using the Hateful Memes Challenge dataset, we conduct a multi-phase investigation on 500 samples to probe the model's reasoning and failure modes. Our central finding is the experimental identification of a "Unimodal Bottleneck," an architectural flaw where the model's advanced multimodal reasoning is systematically preempted by context-blind safety filters. A quantitative validation of 144 content policy refusals reveals that these overrides are triggered in equal measure by unimodal visual 50% and textual 50% content. We further demonstrate that this safety system is brittle, blocking not only high-risk imagery but also benign, common meme formats, leading to predictable false positives. These findings expose a fundamental tension between capability and safety in state-of-the-art LMMs, highlighting the need for more integrated, context-aware alignment strategies to ensure AI systems can be deployed both safely and effectively.
arXiv:2509.14250v2 Announce Type: replace
Abstract: This paper explores prompts and prompting in large language models (LLMs) as dynamic semiotic phenomena, drawing on Peirce's triadic model of signs, his nine sign types, and the Dynacom model of communication. The aim is to reconceptualize prompting not as a technical input mechanism but as a communicative and epistemic act involving an iterative process of sign formation, interpretation, and refinement. The theoretical foundation rests on Peirce's semiotics, particularly the interplay between representamen, object, and interpretant, and the typological richness of signs: qualisign, sinsign, legisign; icon, index, symbol; rheme, dicent, argument - alongside the interpretant triad captured in the Dynacom model. Analytically, the paper positions the LLM as a semiotic resource that generates interpretants in response to user prompts, thereby participating in meaning-making within shared universes of discourse. The findings suggest that prompting is a semiotic and communicative process that redefines how knowledge is organized, searched, interpreted, and co-constructed in digital environments. This perspective invites a reimagining of the theoretical and methodological foundations of knowledge organization and information seeking in the age of computational semiosis
arXiv:2509.14521v4 Announce Type: replace
Abstract: Distributed systems require fusing heterogeneous local probability distributions into a global summary over sparse and unreliable communication networks. Traditional consensus algorithms, which average distributions in Euclidean space, ignore their inherent geometric structure, leading to misleading results. Wasserstein barycenters offer a geometry-aware alternative by minimizing optimal transport costs, but their entropic approximations via the Sinkhorn algorithm typically require centralized coordination. This paper proposes a fully decentralized Sinkhorn algorithm that reformulates the centralized geometric mean as an arithmetic average in the log-domain, enabling approximation through local gossip protocols. Agents exchange log-messages with neighbors, interleaving consensus phases with local updates to mimic centralized iterations without a coordinator. To optimize bandwidth, we integrate event-triggered transmissions and b-bit quantization, providing tunable trade-offs between accuracy and communication while accommodating asynchrony and packet loss. Under mild assumptions, we prove convergence to a neighborhood of the centralized entropic barycenter, with bias linearly dependent on consensus tolerance, trigger threshold, and quantization error. Complexity scales near-linearly with network size. Simulations confirm near-centralized accuracy with significantly fewer messages, across various topologies and conditions.
arXiv:2509.15543v2 Announce Type: replace
Abstract: Existing decentralized stochastic optimization methods assume the lower-level loss function is strongly convex and the stochastic gradient noise has finite variance. These strong assumptions typically are not satisfied in real-world machine learning models. For example, learning on language data typically leads to heavy-tailed gradient. To address these limitations, we develop a novel decentralized stochastic bilevel optimization algorithm for the nonconvex bilevel optimization problem under heavy-tailed noise. Specifically, we develop a normalized stochastic variance-reduced bilevel gradient descent algorithm, which does not rely on any clipping operation. Moreover, we establish its convergence rate by innovatively bounding interdependent gradient sequences under heavy-tailed noise for nonconvex decentralized bilevel optimization problems. As far as we know, this is the first decentralized bilevel optimization algorithm with rigorous theoretical guarantees under heavy-tailed noise. The extensive experimental results confirm the effectiveness of our algorithm in handling heavy-tailed noise.
arXiv:2601.06870v2 Announce Type: replace
Abstract: Multimodal large language models have demonstrated strong ability in capturing semantic representations for multimodal sentiment analysis. Their capacity to learn stable and generalizable multimodal features is limited, however, by the scarcity of high-quality training data. To address this, we propose QASA (Quality-Aware Semantic Augmentation), which uses diffusion models to generate augmented visual and auditory samples, thereby enlarging the training dataset and supporting multimodal learning. The generated samples can vary in quality and may exhibit cross-modal inconsistencies. To manage this, we introduce a decoupled quality-aware scoring module that assigns training weights based on the reliability of each augmented sample. This approach reduces the influence of low-quality data and contributes to more stable and robust model training. The framework combines the generative capabilities of diffusion models with the semantic reasoning of multimodal large models, providing an automated data augmentation strategy that does not require human annotation while improving generalization and robustness under limited high-quality data. Experiments on the CH-SIMS dataset show that QASA yields a relative increase of 18.0\% and 5.9\% in five-class accuracy (Acc5) and binary accuracy (Acc2), respectively, and it also outperforms existing methods on the CMU-MOSI and MUStARD benchmarks.
arXiv:2509.16139v5 Announce Type: replace
Abstract: Predicting the extreme hydrodynamic response of porous and architected lattice materials is a fundamental challenge in high energy density physics, where shock-induced pore collapse, baroclinic vorticity, and anomalous kinetic and thermodynamic states must be resolved across multiple scales. Traditional high-fidelity hydrocodes are computationally prohibitive for large-scale design exploration in applications like planetary defense and inertial confinement fusion. We present a multi-field spatio-temporal model (MSTM) designed to overcome the limitations of standard machine learning surrogates, which often fail to capture the sharp gradients and non-linear field couplings characteristic of shock propagation. By training on high-fidelity, multiscale multiphysics data, MSTM simultaneously evolves seven coupled thermodynamic and kinetic fields - including pressure, temperature, density, and velocity - across complex material architectures. Our framework demonstrates the ability to accurately predict anomalous responses, such as counterintuitive post-shock density reductions and localized hotspot formation, with mean root mean squared errors as low as 1.4%. Crucially, the model's multi-field formulation maintains physical consistency and interface stability over long autoregressive rollouts, outperforming single-field models by 94% in structural fidelity. This framework enables a 1000x reduction in time to solution, providing a practical pathway for the real-time analysis and optimization of energy dissipation and momentum transfer in meso-structured media.
arXiv:2509.16931v3 Announce Type: replace
Abstract: The pre-ranking stage in industrial recommendation systems faces a fundamental conflict between efficiency and effectiveness. While powerful models like Target Attention (TA) excel at capturing complex feature interactions in the ranking stage, their high computational cost makes them infeasible for pre-ranking, which often relies on simplistic vector-product models. This disparity creates a significant performance bottleneck for the entire system. To bridge this gap, we propose TARQ, a novel pre-ranking framework. Inspired by generative models, TARQ's key innovation is to equip pre-ranking with an architecture approximate to TA by Residual Quantization. This allows us to bring the modeling power of TA into the latency-critical pre-ranking stage for the first time, establishing a new state-of-the-art trade-off between accuracy and efficiency. Extensive offline experiments and large-scale online A/B tests at Taobao demonstrate TARQ's significant improvements in ranking performance. Consequently, our model has been fully deployed in production, serving tens of millions of daily active users and yielding substantial business improvements. The code and data are available at https://github.com/zyody/tarq_sigir2026.
arXiv:2605.25710v1 Announce Type: new
Abstract: Realistic physical systems are characterised by emergent interactions across multiple length and time scales, posing a significant challenge for predictive machine learning (ML) models. Most scientific ML models focus on a narrow range of interactions. While machine learning force fields (MLFFs) offer near-quantum accuracy, the ubiquitous message-passing layers miss long-range many-body effects. Here we introduce the Multiscale Structural Ensemble (MuSE), a hierarchical model that uses Soft Coarse-Graining Pooling to construct coarse representations from smooth fractional assignments of atoms to coarse nodes, enabling MLFF modules to operate across multiple scales. MuSE is architecture-agnostic and coupled with SO3krates, MACE, and PaiNN MLFFs for both molecules and materials. We demonstrate the power of MuSE through Hessian-based benchmarks, folding trajectories for biomolecules, and energy profiles in molecule-graphene nanostructures, where MuSE accurately captures quantum-mechanical interactions at relevant scales -- unlike other recent long-range ML models.
arXiv:2509.17057v3 Announce Type: replace
Abstract: We present RoboManipBaselines, an open-source software framework for imitation learning research in robotic manipulation. The framework supports the entire imitation learning pipeline, including data collection, policy training, and rollout, across both simulation and real-world environments. Its design emphasizes integration through a consistent workflow, generality across diverse environments and robot platforms, extensibility for easily adding new robots, tasks, and policies, and reproducibility through evaluations using publicly available datasets. RoboManipBaselines systematically implements the core components of imitation learning: environment, dataset, and policy. Through a unified interface, the framework supports multiple simulators and real robot environments, as well as multimodal sensors and a wide variety of policy models. We further present benchmark evaluations in both simulation and real-world environments and introduce several research applications, including data augmentation, integration with tactile models, interactive robotic systems, 3D sensing evaluation, and hardware extensions. These results demonstrate that RoboManipBaselines provides a useful foundation for advancing research and experimental validation in robotic manipulation using imitation learning. https://isri-aist.github.io/RoboManipBaselines-ProjectPage
arXiv:2509.17878v2 Announce Type: replace
Abstract: As artificial intelligence (AI) becomes increasingly embedded in digital, social, and institutional infrastructures, and AI and platforms are merged into hybrid structures, systemic risk has emerged as a critical but undertheorized challenge. In this paper, we develop a rigorous framework for understanding systemic risk in AI, platform, and hybrid system governance, drawing on insights from finance, complex systems theory, climate change, and cybersecurity - domains where systemic risk has already shaped regulatory responses. We argue that recent legislation, including the EU's AI Act and Digital Services Act (DSA), invokes systemic risk but relies on narrow or ambiguous characterizations of this notion, sometimes reducing this risk to specific capabilities present in frontier AI models, or to harms occurring in economic market settings. The DSA, we show, actually does a better job at identifying systemic risk than the more recent AI Act. Our framework highlights novel risk pathways, including the possibility of systemic failures arising from the interaction of multiple AI agents. We identify four levels of AI-related systemic risk and emphasize that discrimination at scale and systematic hallucinations, despite their capacity to destabilize institutions and fundamental rights, may not fall under current legal definitions, given the AI Act's focus on frontier model capabilities. We then test the DSA, the AI Act, and our own framework on five key examples, and propose reforms that broaden systemic risk assessments, strengthen coordination between regulatory regimes, and explicitly incorporate collective harms.
arXiv:2509.21592v2 Announce Type: replace
Abstract: We consider the problem of forecasting motion from a single image, i.e., predicting how objects in the world are likely to move, without the ability to observe other parameters such as the object velocities or the forces applied to them. We formulate this task as conditional generation of dense trajectory grids with a model that closely follows the architecture of modern video generators but outputs motion trajectories instead of pixels. This approach captures scene-wide dynamics and uncertainty, yielding more accurate and diverse predictions than prior regressors and generators. We extensively evaluate our method on simulated data, demonstrate its effectiveness on downstream applications such as robotics, and show promising accuracy on real-world intuitive physics datasets. Although recent state-of-the-art video generators are often regarded as world models, we show that they struggle with forecasting motion from a single image, even in simple physical scenarios such as falling blocks or mechanical object interactions, despite fine-tuning on such data. We show that this limitation arises from the overhead of generating pixels rather than directly modeling motion.
arXiv:2509.22299v3 Announce Type: replace
Abstract: Mixture-of-Experts (MoE) architectures in large language models (LLMs) deliver exceptional performance and reduced inference costs compared to dense LLMs. However, their large parameter counts result in prohibitive memory requirements, limiting practical deployment. While existing pruning methods primarily focus on expert-level pruning, this coarse granularity often leads to substantial accuracy degradation. In this work, we introduce HEAPr, a novel pruning algorithm that decomposes experts into smaller, indivisible atomic experts, enabling more precise and flexible atomic expert pruning. To measure the importance of each atomic expert, we leverage second-order information based on principles similar to the Optimal Brain Surgeon theory. To address the computational and storage challenges posed by second-order information, HEAPr exploits the inherent properties of atomic experts to transform the second-order information from expert parameters into that of atomic expert parameters, and further simplifies it to the second-order information of atomic expert outputs. This approach reduces the space complexity from $O(d^4)$, where $d$ is the model's dimensionality, to $O(d^2)$. HEAPr requires only two forward passes and one backward pass on a small calibration set to compute the importance of atomic experts. Extensive experiments on MoE models, including DeepSeek MoE and Qwen MoE family, demonstrate that HEAPr outperforms existing expert-level pruning methods across a wide range of pruning ratios and benchmarks. Specifically, HEAPr achieves nearly lossless compression at pruning ratios of 20% ~ 25% in most models, while also reducing FLOPs nearly by 20%. The code can be found at [https://github.com/LLIKKE/HEAPr](https://github.com/LLIKKE/HEAPr).
arXiv:2602.02605v2 Announce Type: replace
Abstract: Evaluating true metacognition in Large Language Models (LLMs) is difficult due to biases and heuristics. This paper presents a framework to measure and enhance LLM metacognition while controlling for these biases. A measurement method using the $d'_{\rm type2}$ metric is established to isolate metacognitive ability. The Evolution Strategy for Metacognitive Alignment (ESMA) is proposed, demonstrating robust generalization across unseen datasets, languages, and newly acquired knowledge. Finally, parameter analysis reveals that these improvements are driven by a sparse set of parameters, offering new pathways for targeted metacognitive optimization.
arXiv:2509.23413v2 Announce Type: replace
Abstract: Multi-task neural routing solvers have emerged as a promising paradigm for their ability to solve multiple vehicle routing problems (VRPs) using a single model. However, existing neural solvers typically rely on predefined problem constraints or require per-problem fine-tuning, which substantially limits their zero-shot generalization ability to unseen VRP variants. To address this critical bottleneck, we propose URS, a unified neural routing solver that achieves zero-shot generalization across a wide range of unseen VRPs with a single model. We propose a unified data representation (UDR) that replaces problem enumeration with data unification, thereby broadening the problem coverage and reducing reliance on domain expertise. In addition, we introduce a Mixed Bias Module (MBM) during encoding to improve node embeddings, which efficiently captures multiple priors inherent to various problems. On top of the UDR, we develop a problem-conditioned parameter generator to further improve zero-shot generalization. Extensive experiments show that URS consistently produces high-quality solutions for 110 VRP variants (including 99 unseen variants) while demonstrating impressive scalability to large-scale instances with up to 7000 nodes. To the best of our knowledge, URS is the first neural solver to handle over 100 VRP variants with a single model. Our code is available at https://github.com/CIAM-Group/URS.
arXiv:2509.23975v2 Announce Type: replace
Abstract: The control of high-dimensional distributed parameter systems (DPS) remains a challenge when explicit coarse-grained equations are unavailable. Classical equation-free (EF) approaches rely on fine-scale simulators treated as black-box timesteppers. However, repeated simulations for steady-state computation, linearization, and control design are often computationally prohibitive, or the microscopic timestepper may not even be available, leaving us with data as the only resource. We propose a data-driven alternative that uses local neural operators, trained on spatiotemporal microscopic/mesoscopic data, to obtain efficient short-time solution operators. These surrogates are employed within Krylov subspace methods to compute coarse stable and unstable steady states, while also providing Jacobian information in a matrix-free manner. Krylov-Arnoldi iterations then approximate the dominant eigenspectrum, yielding reduced models that capture the open-loop slow dynamics without explicit Jacobian assembly. Both discrete-time Linear Quadratic Regulator (dLQR) and pole-placement (PP) controllers are based on this reduced system and lifted back to the full nonlinear dynamics, thereby closing the feedback loop. The framework is validated by stabilizing an unstable steady-state of the Liouville-Bratu PDE, demonstrating consistent performance between the learned surrogate and the true system, with quantified degradation under plant-model mismatch.
arXiv:2509.24050v4 Announce Type: replace
Abstract: Device-cloud collaboration holds promise for deploying large language models (LLMs), leveraging lightweight on-device models for efficiency while relying on powerful cloud models for superior reasoning. A central challenge in this setting is determining, for each incoming query, whether it should be processed locally or offloaded to the cloud. Existing approaches typically rely on external routers, which often struggle to determine difficulty from the prompt itself, especially for tasks involving complex reasoning. Motivated by this limitation, we propose enabling on-device LLMs to decide internally whether to invoke cloud assistance at inference time, with this capability instilled through reinforcement learning based post-training. Casting on-device LLM post-training as a reward maximization problem, we design hierarchical rewards to encourage local problem solving and judicious cloud offloading. To solve the resulting problem, we develop an algorithm featuring a group-level policy gradient that stabilizes optimization, together with adaptive prompt filtering that provides complementary learning signals to mitigate policy collapse (i.e., exclusive local execution or exclusive cloud offloading). Extensive experiments on on-device-scale LLaMA and Qwen models across multiple reasoning benchmarks show that our method consistently outperforms baselines and significantly narrows the gap to full cloud LLMs.
arXiv:2509.24621v3 Announce Type: replace
Abstract: Multimodal large language models (MLLMs) are emerging as versatile foundations for mixed-modality retrieval. Yet, they often require heavy post-hoc training to convert them into contrastive encoders for retrieval. This work asks: Can off-the-shelf MLLMs serve as powerful retrievers without additional training? We present FreeRet, a plug-and-play framework that turns any MLLM into a two-stage retriever. FreeRet first derives semantically grounded embeddings directly from the model for fast candidate search, and then exploits its reasoning ability for precise reranking. The framework contributes three advances: bypassing lexical alignment layers to obtain semantically faithful embeddings, conditioning representation generation with explicit priors, and mitigating framing effect in reranking via neutral choice framing. On the MMEB and MMEB-V2 benchmarks spanning 46 datasets, FreeRet substantially outperforms models trained on millions of pairs. Beyond benchmarks, FreeRet is model-agnostic and scales seamlessly across MLLM families and sizes, preserves their generative abilities, supports arbitrary modality combinations, and unifies retrieval, reranking, and generation into end-to-end RAG within a single model. Our findings demonstrate that pretrained MLLMs, when carefully harnessed, can serve as strong retrieval engines without training, closing a critical gap in their role as generalists.
arXiv:2509.26204v2 Announce Type: replace
Abstract: Automated test generation (ATG), which aims to reduce the cost of manual test suite development, has been investigated for decades and has produced countless techniques based on a variety of approaches: symbolic analysis, search-based, random and adaptive-random, learning-based, and, most recently, large-language-model-based approaches. However, despite this large body of research, there is still a gap in our understanding of the characteristics of developer-written tests and, consequently, our assessment of how well ATG techniques and tools can generate realistic and representative tests. To bridge this gap, we conducted an extensive empirical study of developer-written tests for Java applications, covering 1.7 million test cases from open-source repositories. Our study is the first of its kind to evaluate aspects of developer-written tests that are mostly neglected in the existing literature -- including test scope, test fixtures and assertions, types of inputs, and use of mocking -- and characterize tests accordingly. Based on this characterization, we then compare existing tests with those generated by two state-of-the-art ATG tools. Our results highlight that the vast majority of developer-written tests exhibit characteristics that are beyond the capabilities of current ATG tools. Finally, based on our findings, we identify promising research directions that can help develop more effective tool support for developer testing practices. We believe this work can set the stage for additional research and bring ATG tools closer to generating the types of tests developers write.
arXiv:2509.26545v3 Announce Type: replace
Abstract: This work demonstrates that the Lyapunov method can effectively identify the growth rate of a linear time-periodic system describing cold fresh water on top of hot salty water with a periodically time-varying background shear flow. We employ a time-dependent weighting matrix to construct a Lyapunov function candidate, and the resulting linear matrix inequalities are discretized in time using the forward Euler method. As the number of temporal discretization points increases, the growth rate predicted from the Lyapunov method or the Floquet theory will converge to the same value as that obtained from numerical simulations. Additionally, the Lyapunov method is used to analyze the most dangerous disturbance, and we also compare computational resource usage for the Lyapunov method, numerical simulations, and the Floquet theory.