Papers

This work presents the design, implementation, and evaluation of a system for generating personalized reading content using Large Language Models (LLMs) combined with Retrieval-Augmented Generation (RAG). The proposed architecture consists of four modules: Input, RAG, Generation, and Judging and enables users to specify both a question and a target reading content complexity. RAG is employed to retrieve relevant information from the Internet, enriching and grounding the content produced by three modern LLMs: Meta LLaMA 4 Scout, LLaMA 3.1 8B Instant, and Google Gemma2 9B. Reading materials are generated using three prompting strategies (Chain-of-Thought, zero-shot, and few-shot), and the LLM-as-a-Judge module automatically evaluates answer quality and alignment with the desired readability level. Experimental results show that RAG consistently improves system performance across all models and prompting techniques, increasing relevance and particularly groundedness by up to 26-35 percentage points. Overall, the findings demonstrate that the RAG-augmented architecture effectively produces reading content tailored to user queries and desired textual complexity.

This paper presents a structured analysis of security challenges in long-horizon agentic AI systems. The study reviews existing threats, evaluation approaches, attack propagation mechanisms, and security frameworks. A taxonomy of security threats and a framework for analyzing attack propagation are proposed to support future research in agentic AI security

The Materials Science and Engineering ontology landscape is fragmented along multiple axes simultaneously. Horizontally: a recent survey identified 94 ontologies of which over 40 are structurally incompatible; each new application domain – ceramics, polymers, batteries, smart materials – typically restarts ontology design from scratch. Vertically: EU regulation (CSRD, CSDDD, PPWR, CBAM, R2R, AI Act, ESPR) forces material, manufacturing, supply-chain, and lifecycle data into integrated digital product passports, leaving ontologies that only address horizontal fragmentation incomplete for any contemporary consumer. And mechanistically: a vocabulary that records that BNT-BT has $d_{33} \approx 580$ pC/N stores a fact but cannot surface why – Bi-6s$^2$ lone-pair stereo-activity, anomalous Born effective charges, soft modes, defect chemistry – without a systematic explanation skeleton. We propose a multi-level modular architecture with two independent classification axes – level of abstraction (L0 bridges, L1 material-agnostic laboratory-notebook, L2 material-class-specific, L3 categorical reasoning) and consumer audience (material vs. compliance) – in which the material-specific level is internally organised by a seven-tier mechanistic-explanation skeleton (Symmetry, Energy/DFT, Thermo/CALPHAD, Kinetics, Microstructure, Defect chemistry, Bonding) applicable to any crystalline ionic oxide. The level-and-audience modularity dissolves the horizontal fragmentation, the compliance audience absorbs the vertical regulation pressure, and the seven-tier organisation of Level 2 delivers the mechanistic explanation depth. We instantiate the architecture as the OntoCrafter Ceramics Ontology (OCO v0.94): 5,196 classes across 44 modules; 167,348 OWL axioms (40,454 logical); 1,674 properties; 829 cross-ontology bridge mappings; 1,172 SHACL shapes; 163 published competency questions.

Jet tagging at the Large Hadron Collider increasingly relies on deep learning models trained on massive simulated datasets, leading to high computational costs and limited robustness to detector mismodeling. We introduce JetParticle-JEPA (JP-JEPA), a self-supervised Joint-Embedding Predictive Architecture that learns physically meaningful jet representations directly from continuous particle clouds without tokenization or reconstruction of raw inputs. Built on a Particle Transformer backbone, JP-JEPA predicts latent representations of masked particles while preserving fine-grained kinematic correlations. On the JetClass benchmark, JP-JEPA achieves performance comparable to fully supervised state-of-the-art methods on the full dataset, surpasses supervised baselines in low-label regimes, and significantly outperforms existing SSL approaches. On Top Quark and Quark-Gluon Tagging benchmarks, it remains on par with supervised methods. The learned representations also exhibit strong robustness to missing detector information and improved uncertainty behavior, highlighting JP-JEPA as a promising foundation-model framework for robust and data-efficient jet physics at the LHC.

Building trustworthy medical multimodal large language models (MLLMs) is critical for reliable clinical decision support. Existing medical hallucination benchmarks mainly focus on data collection, but often ignore where hallucinations originate within the reasoning process. We find that hallucination sources vary across samples: errors may arise from visual misrecognition, incorrect medical knowledge recall, or flawed reasoning integration. To enable source-level hallucination diagnosis, we introduce ClinHallu, a benchmark for stage-wise hallucination diagnosis in medical MLLM reasoning. ClinHallu contains 7,031 validated instances, where each instance is augmented with a structured reasoning trace decomposed into Visual Recognition, Knowledge Recall, and Reasoning Integration. We also use stage-replacement interventions to measure how correcting specific stages affects the final answer. Beyond evaluation, we show that trace-supervised fine-tuning reduces stage-wise hallucinations. ClinHallu provides a fine-grained hallucination testbed for diagnosing and mitigating reasoning failures in medical MLLMs. The benchmark is publicly available at https://github.com/alibaba-damo-academy/ClinHallu.

Cooperative multi-objective multi-agent reinforcement learning (MOMARL) models team decision making under multiple, potentially conflicting objectives. In this setting, conflicts arise not only across objectives but also across agents with different observations, roles, and contributions. We propose Preference Coordinated Multi-agent Policy Optimization (PCMA), which learns coordinated agent-specific preferences to enable complementary trade-offs among agents. Theoretically, we formulate cooperative MOMARL as a team-optimal equilibrium problem, and show that, under suitable conditions, preference diversity yields a first-order improvement in the team objective. Experiments on multiple cooperative MOMA environments and a practical traffic-control scenario show that PCMA improves both performance and trade-off coordination.

AI systems coupled to proof assistants now generate formal mathematics at scale, and the gap between what a checker can verify and what a mathematician would value has become the binding constraint. We model the generation of valuable mathematics as nested language generation in the limit: a verifiable formal language $F$, accessed through a membership oracle (the proof checker), contains an unknown valuable language $H \in \mathcal{H}$ revealed only through an adversarial enumeration of a core $C \subseteq H$ of exact density $\alpha$ (the literature). Every output is valuable ($\in H$), trivial ($\in F \setminus H$), or a hallucination ($\notin F$). We settle four questions. First, the verifier is not taste: the collections admitting generation with breadth are exactly those of the oracle-free model, characterized fiber-wise by Angluin's condition. Second, the verifier does buy sound coverage, covering all unseen valuable statements while asserting only valid ones: possible with it, impossible without it; it relocates unavoidable errors from false to trivial. Third, and centrally, a sharp dichotomy on the tight family: generators emitting finitely many trivia achieve optimal coverage $\alpha/2$, while any infinite trivia allowance, even at vanishing rate, jumps the optimum to $1-\alpha/2$ (both tight, for cores presented as the candidate intersection), and one generator attains both ends. The transition is in trivia count, not rate; the gap $1-\alpha$ is the unrecorded mass. Fourth, both regimes instantiate in a compression model of mathematics. A perfect verifier cannot substitute for taste: the unbounded stream of correct-but-worthless statements is not an engineering accident but a provable necessity, since covering unrecorded valuable mathematics requires an infinite, but asymptotically negligible, stream of certified trivia.

Globally, cotton is a highly economically beneficial crop, as the textile industry heavily depends on it. So, the precise identification and detection of cotton leaf disease is crucial for economic stability. The development goal of "CottonLeafVision" is to accurately classify and detect cotton leaf disease. With this goal, we have evaluated multiple pretrained Deep Convolutional Neural Networks, including DenseNet201, InceptionV3, and VGG19 on a publicly available cotton leaf disease image dataset. This image dataset includes seven classes, six disease classes, and one healthy class, collected under various field conditions reflecting real-world challenges. Among these pretrained models, with DenseNet201, we have achieved the highest classification accuracy of 98%. To enhance the model reliability and interpretability, we have implemented different techniques and methods such as Gradient-weighted Class Activation Mapping (Grad-CAM), occlusion sensitivity analysis and adversarial training to increase the noise resistance of the model. Finally, we have developed a prototype in order to utilize the model's capabilities on real life agriculture. This paper shows the deep learning model's capabilities to classify the disease in real-life cotton disease management situations.

Large language models increasingly serve as execution engines for agentic systems, yet they still consume context through a sequential text interface. This creates a mismatch with modern structured agent workflows, in which independent branches explore subtasks, retrieve evidence, or generate candidate solutions before a final synthesis step. Existing systems typically merge these branches by concatenating their textual outputs, which discards the parallel structure and incurs redundant prefill computation. In this work, we introduce Parallel-Synthesis, a plug-and-play framework that enables a synthesizer to directly consume the KV caches produced by parallel worker agents. Parallel-Synthesis combines a cache mapper that calibrates independently generated branch caches with a fine-tuned synthesizer adapter that enables generation from this non-sequential cache interface. We train Parallel-Synthesis using data that exposes the synthesizer to parallel cache contexts, teaches aggregation across cached branches, and distills reasoning behavior from standard text-concatenation-based synthesis. Across nine downstream datasets spanning math, science QA, code generation, GAIA, and multi-agent database diagnosis, Parallel-Synthesis matches or outperforms text-based synthesis on seven datasets and remains close on the other two. It also reduces time-to-first-token by 2.5x-11x, suggesting that direct cache-based synthesis is a promising interface for more native and efficient synthesis over parallel agent branches.

Artificial Intelligence (AI) is increasingly used to automate a variety of real-world computer vision (CV) applications, such as autonomous vehicle control, facial recognition, and security cameras. Recent research has shown that acoustic vibration can induce real physical motion in cameras, interfering with their internal stabilization mechanisms. Because the motion falls outside the conditions the stabilization system was designed to handle, the system introduces artifacts into the frame, causing AI-based CV models to misclassify, miss targets, or hallucinate objects. Previous work used ultrasonic frequencies (>20 kHz) to perform short-range attacks, which limits them to short distances due to the attenuation exhibited by high frequencies. In this work, we investigate acoustic attacks using lower frequencies in the audible range (<20 kHz), and we further expand our analysis to include how various image and object features are affected by the attacks. Specifically, we performed physical experiments to demonstrate the viability of our attacks on an off-the-shelf object detection model (YOLO11) by resonating a commercially available camera with various frequencies. Based on our results, we provide insights into several factors that make an AI CV system more vulnerable to these attacks, which could help inform the development of future mitigation strategies.

Sequential or time-stamped interaction logs provide objective records of digital application usage, yet their granularity and noise often obscure meaningful insights into people's work. Such insights are essential for improving digital products in ways grounded in real-world user interactions. Prior research has applied deep learning models to cluster user actions into high-level activities, but these approaches are highly sensitive to noise and struggle to generalize across applications. To address this limitation, we introduce WorkflowView, a framework that uses large language models (LLMs) to abstract low-level action sequences into high-level activities. We establish the effectiveness and generality of our approach across three distinct, challenging sequential tasks and diverse domains: (a) zero-shot task description reconstruction from browser logs (achieving high semantic similarity, $\mu_{sim} = 0.91$), (b) few-shot student dropout prediction using MOOC interaction logs (reaching weighted $F_1 = 0.90$ with only five few-shot examples), and (c) anonymized, privacy-preserving analysis of AI tool integration within document workflows in Microsoft Word. Our work demonstrates that LLM-based abstraction is a robust and efficient path forward for transforming low-level behavioral data into high-level, interpretable, and actionable insights. We also discuss practical considerations for deploying LLM-based inferences within logging infrastructures, including computational efficiency and user privacy.

Transformer-based automatic speech recognition (ASR) models such as Whisper are highly accurate, but their predictions remain difficult to interpret. Existing explainable AI (XAI) methods often lack faithfulness and precise temporal grounding. We propose Listening with Entropy-guided Attention for Faithful explainability (LEAF-X), a model-intrinsic XAI framework for transformer-based ASR. LEAF-X combines entropy-guided attention weighting, multi-layer attention rollout, and optional causal ablations to identify low-entropy, high-impact heads and layers, producing sparse token-to-frame attributions. Unlike perturbation-based explainers or raw attention maps, LEAF-X exploits the internal structure of encoder-decoder and speech-augmented decoder-only models to generate explanations that better reflect model computation. Results show 32% improved faithfulness, 35-39% stronger locality/sparsity, and the most stable attributions, supporting more transparent and auditable ASR.

Recent advances in speech generation have significantly improved the naturalness of synthetic speech, making spoofing detection increasingly challenging. A key limitation of current anti-spoofing systems is their limited robustness to unseen synthesis methods. In this work, we transform a self-supervised speech representation model into a Mixture-of-Experts (MoE) architecture to improve generalization. Feed-forward blocks in selected encoder layers are replaced by multiple expert networks controlled by a layer-wise gating mechanism, allowing experts to capture complementary acoustic patterns while preserving the representations learned during self-supervised pretraining. We further analyze the architectural choices affecting the performance of this MoE conversion and investigate the activation behavior of the experts. The proposed approach is evaluated on 14 spoofing datasets and reduces the macro EER from 5.46% to 4.81%, corresponding to 11.9% relative improvement over the baseline.

Verifier reliability is not portable across tasks. A verifier-guided self-DPO pipeline with genuine held-out gains on MathVista (+9.6 points on self-training data, +8.0 held out) can be harmful on MMMU. The failure is invisible from the target-task self-training signal: over six learner-verifier configurations, MMMU self-training performance still improves by +3.52 points while held-out performance drops by 1.42 points. We call this silent negative transfer: a verifier validated as useful before can still show signatures of successful self-training after its induced update stops transferring to unseen data. Downstream failure alone cannot diagnose this: poor generalization may mean the verifier induces a wrong learning direction, or that the correct direction itself does not generalize beyond the self-training distribution. We separate these effects via gradient fidelity F, the alignment between verifier-induced and correct training directions, and gradient transferability T, the alignment between correct and held-out directions, giving a conservative Safe-Transfer Margin: positive first-order held-out alignment is certified when arccos(F) + arccos(T) < pi/2. We then introduce Asymmetric Acceptance Gating (AAG), a precision-first filter selecting preference pairs by the verifier's absolute confidence in the response receiving the positive update. On MMMU, AAG raises fidelity from 0.29 to 0.42. In a same-cell analysis, the correct direction stays positively aligned with the held-out direction (T = 0.31), yet raw verification rotates the induced update to negative held-out alignment (-0.13); AAG restores positive alignment (+0.22). Across all six MMMU configurations, AAG beats raw verifier-guided training. Verifier reliability thus means not that it worked before, but that it induces the right update on the task at hand and that this update stays useful beyond the data that generated it.

We show that the three movements of Beethoven's "Moonlight Sonata" (Op. 27 No. 2) instantiate three distinct machine learning architectures – not by analogy, but by structural correspondence. Through computational analysis of the score (entropy, Jensen-Shannon divergence, dissonance, hand distributional overlap, self-similarity matrices, temporal memory decay, and contextual pitch embeddings), we establish four counterintuitive findings: (1) perceived musical "temperature" is governed by throughput, not distributional width; (2) the lightest movement carries the highest dissonance; (3) the movements implement streaming, recurrent, and periodic positional encoding memory architectures; and (4) the same pitch class acquires different contextual identities across movements, analogous to contextual vs.static embeddings in NLP – and unsupervised clustering recovers the tonal structure without music-theoretic input. We construct a reverse sonification (decoding analytical features back into MIDI) and quantify the chirality of the encode-decode cycle: what distributions preserve and sequential ordering destroys. Prompted by a listener's observation that the decoded piece sounds like "mirror isomers that can't be superimposed," the chirality measurement reveals reconstruction loss increasing monotonically with n-gram order. Bootstrap baselines and subsample checks confirm all movements carry sequential information above noise, though raw values are confounded by sample size. Cross-domain comparison shows natural language has higher chirality than music, reflecting stronger sequential constraints.

Survival prediction plays a central role for healthcare providers and clinical researchers. Accurate risk stratification enables early intervention and improved patient management. Most existing deep survival models learn one common feature representation for all patients, which may hide important differences between patient subgroups. In contrast, a Mixture-of-Experts (MoE) framework allows different parts of the model to focus on different patient patterns, leading to more individualized representations. Therefore, in this work, we propose a mixture-of-experts enhanced adaptive deep clustering survival framework (AdaCSM) for modeling such heterogeneous survival patterns. We introduce a routing-based expert mechanism that enables conditional specialization within a parametric survival modeling framework. The proposed architecture allocates patients to specialized risk predictors dynamically while preserving the patient survival and subtype clustering objectives. We compare our method with state-of-the-art survival and deep clustering models on multiple real-world longitudinal clinical cohorts spanning diverse disease domains. The proposed method demonstrates improved predictive performance and leads to interpretable results in survival analysis.

Wearable devices and smartphones generate rich behavioural time series that can support proactive health interventions, yet systematic comparisons of modern forecasting architectures for these data are lacking. In particular, it remains unclear how models generalise across populations, how different architectures respond to participant-level fine-tuning and how forecasting accuracy degrades across multi-day horizons. We benchmark six deep learning architectures, two zero-shot Foundation Models (FM) and statistical baselines on three public datasets encompassing over 800 participants, reporting per-feature metrics for step counts, screen time and sleep duration across 1-8 day horizons. We further conduct a per-feature personalisation study across all six architectures and assess FM transferability across dataset sizes and temporal granularities. Our key findings are: (i) no single architecture dominates, PatchTST leads among trained models while the three runners-up (TCN, MLP, Transformer) show no meaningful performance difference; (ii) the FM TimesFM matches or exceeds trained models zero-shot, especially in low-data regimes and (iii) participant-level fine-tuning reduces per-feature RMSE by 16-60%, with sleep benefiting most and step counts least. These results provide practical guidance on architecture selection, FM applicability and personalisation strategies for mobile health forecasting. To the best of our knowledge, this is the first study to jointly evaluate modern deep learning, FMs and personalisation for multi-horizon behavioural forecasting from wearables.

AI-assisted software development has moved from line-level autocomplete to agents that can plan changes, edit files, and submit pull requests with limited human supervision. Open-source software, however, evolves through a process designed for humans: contributor agreements, codes of conduct, and review norms all assume a legally accountable person who can attest to provenance and answer reviewer questions. Autonomous and semi-autonomous AI contributors strain those assumptions, and the 2025-2026 record of agent-driven incidents, AI-generated nuisance volume, and platform-level shutdowns shows that the gap is operationally consequential. Several open-source organisations have responded with contribution policies, but the result is fragmented, and its alignment with emerging AI governance frameworks (EU AI Act, NIST AI RMF with the UC Berkeley Agentic AI Profile, ISO/IEC 42001 and 23894) is unmapped at the contribution level. We compare policies across six organisations (SymPy, LLVM, matplotlib, OpenInfra, the Apache Software Foundation, and the Linux Foundation) using Most-Similar Systems Design with indicator-based coding and process tracing for SymPy and LLVM. From this we derive a six-dimensional taxonomy (disclosure, responsibility, human oversight, licensing, enforcement, maintainer workload), an ordinal Policy Maturity Score, and a mapping of documented agent incidents onto the dimensions each policy fails to govern. Aligning the dimensions with the regulatory frameworks above identifies overlapping gaps neither side currently closes, and we close by sketching the shape of a harmonised tiered framework and the empirical evaluation needed to calibrate it.

Recent advances in pretrained large audio-language models (LALMs) have demonstrated strong capabilities across speech, sound, and music. To adapt these models to downstream tasks without the cost of pretraining from scratch, post-training has become a widely adopted paradigm. However, the effectiveness of post-training depends critically on the quality of the training corpus. We observe that existing post-training corpora, often constructed by aggregating public audio datasets, suffer from substantial acoustic redundancy, as many of these datasets are sourced from overlapping media platforms. Such redundancy leads to repeated exposure to similar acoustic patterns, causing diminishing returns in performance despite increased data volume. address this issue, we propose a three-stage data construction pipeline that performs acoustic redundancy filtering, converts retained samples into a unified multiple-choice question-answering format with chain-of-thought generation, and finally applies quality verification and filtering. Using this pipeline, we construct AudioRE, a post-training dataset of approximately 286k instances spanning sound, speech, and music. Supervised fine-tuning on AudioRE consistently improves the performance of Qwen2-Audio-7B-Instruct across diverse audio understanding and reasoning benchmarks, outperforming models trained on the unfiltered raw corpus with substantially more instances. These results validate the effectiveness of our redundancy-aware data construction pipeline and the resulting AudioRE dataset, and further highlight the importance of minimizing acoustic redundancy in audio-language post-training. To facilitate future research, we will release both the AudioRE and the fine-tuned Qwen2-AudioRE checkpoint.

LLM agent systems increasingly run as long-lived autonomous runtimes: scheduling jobs, calling tools, maintaining memory, and pushing results to humans. We present a longitudinal study of silent failures in one such system: a personal-assistant agent runtime in continuous production since March 2026, with roughly 40 scheduled jobs, 8 LLM providers, a tool-governance proxy, and a knowledge-base memory plane, defended by 4,286 unit tests and 827 governance checks. Over eight weeks we documented 22 incidents with full root-cause postmortems, in which one meta-pattern – a failure whose error signal never reaches a human in actionable form – manifested at least 28 times. We derive a five-class, mechanism-oriented taxonomy: (A) environment and platform quirks, (B) design-assumption mismatches, (C) error swallowing and dilution, (D) chained hallucination and fabrication, (E) operational omission and forensic blind spots. Class D is unique to LLM systems and the most dangerous: the system does not merely fail to report an error – the LLM transforms it into fluent, plausible narrative delivered to the user. We term this fail-plausible: gray failure's differential observability escalated – the observer is not just blind, it is convincingly lied to by the failure itself. Three findings: about 70% of silent failures were caught by human user-view observation, not tests or audits; a retrospective audit of 15 incidents found 0% ex-ante prevention but 87% regression blocking – audits are regression engines, not prediction engines; incident latency (13 hours to 60 days) tracks failure mechanism, not code complexity – the longest-lived failures lived in the seams between components, where no test runs. We describe the resulting defense framework and distill design principles for agent systems whose failures are loud, attributable, and boring. All postmortems and artifacts are public.

Generative dynamics models enable planning in challenging systems, but safe deployment requires detecting policy-induced out-of-distribution (OOD) transitions. Existing methods typically treat learned dynamics as fixed and rely on post hoc support surrogates for OOD detection. This overlooks a critical failure mode: learned dynamics that are insensitive to control changes can map unsupported controls to latent predictions resembling demonstrated transitions, suppressing OOD signals despite large prediction errors. We introduce support-conditioned control-sensitivity regularization to preserve control-induced variation by promoting local responsiveness in well-supported training regions. Experiments in vision-based obstacle avoidance, manipulation, and real-robot navigation demonstrate improved OOD detection and safer closed-loop planning.

Efficient route optimization play a vital role in ensuring both safety and punctuality in railway operations. It is very crucial particularly in heterogeneous multi-gauge railway networks with varying train speed, stopping pattern, infrastructure compatibility constraints increase coordination complexity. In single-track systems these challenges are further intensify due to all trains to share the same track and requires frequent track switching.Stochastic disruptions events including blocked tracks, blocked trains, engine failure and speed slowdowns introduces additional unpredictability in operations and deviate the timetable. However, existing studies predominantly focuses on high-level timetabling, omitting operational details such as track switching coordination. As a result leaving decision to human operators, increasing safety risks into railway operations. This study proposes a framework based on temporal planning for dynamic route optimization and disruption management in heterogeneous railway systems. The framework formulates railway operations as a temporal planning problem using PDDL 2.1 with explicitly modeling gauge compatibility constraints and diverse disruption scenarios. It generates conflict-free timestamped operational plans specifying both optimized schedules and executable action sequences. To evaluate the proposed framework, we developed a benchmark problem set with 200 instances using up to 1,000 track points and 120 trains. Two state-of-the-art temporal planners and a plan validator were employed to assessed the framework. The experimental results demonstrate that the framework effectively generates temporal operational plans for heterogeneous railway systems and handles multi-gauge constraints, disruptions, and reduces dependence on manual decision making.

High-throughput experimentation can evaluate many reaction conditions, yet combinatorial condition spaces still exceed the available experiment budget. This makes experiment selection a sequential decision problem: each new condition must be chosen from limited observations before its outcome is known. LLMs can express task-specific selection logic. A direct recommendation, however, is neither a persistent executable object that can be validated and revised nor an independently auditable decision rule. We introduce CARE, a reference-conditioned controller that separates program synthesis from experiment selection. An LLM writes an executable scoring program that ranks the remaining conditions, while a non-LLM reference policy supplies a numerical candidate and support summary. CARE forms an optional alternative from the program, applies a reference-conditioned intervention gate to compare it with the reference, and records the decision before the selected outcome is revealed. Each new outcome updates controller state and can trigger retention, revision, or regeneration of the active program. This outcome-guided program evolution changes the scoring logic without updating the LLM parameters. In matched offline replay with 30 seeds on eight reaction-optimization tasks, CARE attains the lowest normalized regret, the highest normalized best-so-far AUC, and the highest Top-1% Success@15 among the evaluated methods. These results support using a scoring program written by an LLM as one component of a reference-conditioned optimizer rather than as a standalone experiment selector.

When applying Group Relative Policy Optimization (GRPO) for GUI Grounding, rollouts are sampled from a single screenshot view; groups often become either all failures on difficult instances or all successes on easy ones, yielding no useful relative advantage. We propose VISTA (View-Consistent Self-Verified Training), a GRPO-based training framework that constructs each comparison group from multiple target-preserving views of the same GUI instance.Each view is generated by a crop that keeps the target element visible and remaps its box exactly, so model rollouts are compared across semantically equivalent but geometrically different inputs. To stabilize short coordinate generation without turning reinforcement learning into unconditional imitation, VISTA further adds a self-verified cross-view anchor: an oracle answer optimized with an advantage-weighted loss, excluded from the group baseline and activated only when the model has produced a maximum-reward rollout. Across five GUI-grounding benchmarks and multiple Qwen backbones, VISTA consistently improves grounding accuracy.On ScreenSpot-Pro, it raises Qwen3-VL 4B/8B/30B-A3B from 55.5/52.7/53.7 to 63.4/65.8/67.0. Robustness analyses further show higher worst-view accuracy and lower prediction flip rates.

Large language models (LLMs) are increasingly deployed as planners for autonomous agents in household environments. While existing benchmarks evaluate whether LLM-generated plans execute successfully, they overlook a critical type of failure: latent failures. Unlike immediate failures that trigger instant feedback at execution time and enable timely correction, latent failures do not immediately halt plan execution but silently compromise goal achievement. In severe cases, they cause irreversible harm. To address this gap, we introduce SIMMER, a benchmark for evaluating latent failures in LLM planning through a human-curated symbolic world model grounded in the kitchen domain. SIMMER defines a world model comprising 77 actions, 262 unique objects, and approximately 46,800 possible interactions that are semantically realistic, derived from real-world cooking scripts. It then leverages a state machine executor that validates plans against the world model and detects immediate precondition violations, latent hazards, and irreversible failures. Experiments across six LLMs show that even frontier models achieve at most 17% error-free plans. Moreover, up to 56% of plans contain latent failures, the majority of which lead to irreversible consequences. We further demonstrate that explicit state reasoning via counterfactual foresight simulation can reduce latent failures by up to 72% and irreversible cases by up to 75%, suggesting a promising direction for more robust LLM planners.

A central role of personal-agent memory is to turn stored information and prior interactions into future-oriented assistance. In daily use, useful cues come from what the agent observes and how the user interacts with the agent, and the agent must carry them forward from the current request to similar future tasks. Existing memory benchmarks usually test dialogue recall or task improvement in isolation, leaving the trajectory from streaming observations to later assistance largely untested. We introduce StreamMemBench, a streaming benchmark that constructs a two-step task sequence around each evidence anchor from EgoLife egocentric streams. The initial task tests evidence use, while the follow-up task tests whether feedback and interaction experience are reused. Four metrics diagnose evidence recall, initial evidence use, feedback incorporation, and follow-up reuse. Experiments with eight memory systems across two backbones show that current systems often fail to use observed evidence or turn feedback into reliable follow-up behavior, even when evidence is stored or feedback is incorporated locally. StreamMemBench is publicly available at https://github.com/landian60/StreamMemBench.

Emulators provide a cost-effective alternative to regional climate models (RCMs) by capturing their dynamical downscaling function. They link large-scale predictors simulated by global climate models (GCMs) to RCM-simulated high-resolution fields of the target variable, here precipitation. Machine learning methods, typically deep learning, are cheaper than running RCMs in computation time and energy. Among them, generative models are appealing because they can simulate ensembles of local high-resolution fields consistent with the predictors. This ensemble, which we call the uncertainty envelope, remains to be properly assessed for added value. Here, we make three contributions. First, we introduce ParamDiffusion, a new two-stage diffusion-based framework, and compare it with a state-of-the-art diffusion approach. Second, we expand standard validation through a comprehensive framework aligned with climate-science needs, examining specific precipitation events, including extremes. Third, within this framework, we assess the added value of diffusion approaches relative to deterministic methods. We intercompare four deep-learning models: a deterministic model designed to capture the precipitation tail; a parametric probabilistic model based on it; a recently proposed diffusion approach; and ParamDiffusion, which couples the parametric model with a diffusion model. Our results show that diffusion-based approaches reproduce climatological precipitation statistics with high skill, including distributional tails and spatially compounded extremes, while generating spatially detailed fields. However, none of the assessed models consistently accounts for the most extreme RCM-simulated events within its uncertainty envelope. Diffusion models are therefore promising for probabilistic RCM emulation, but progress is still required before they can reliably represent high-impact precipitation extremes.

Modern image classifiers widely adopt global average pooling (GAP) followed by a linear classification head. This linearity ensures that the image-level logits equal the average of logits obtained by applying the classification head pointwise to the feature grid prior to GAP. Consequently, standard classifiers may inherently retain spatial class evidence that remains recoverable even when the image-level prediction is incorrect. This structure naturally suggests a multiple-instance learning (MIL) interpretation, where an image is viewed as a bag of spatial instances. Within this formulation, we demonstrate that standard classifiers trained with a single label per image can still learn the intended classification task in multi-object scenes. We further exploit this property to decompose image-level logits into a prediction grid, providing a post-hoc diagnostic to extract spatial class evidence that GAP otherwise obscures. Our systematic evaluation reveals that off-the-shelf models consistently recover the ground-truth class within foreground regions. The MIL interpretation further suggests that common classifier failures reflect known limitations of mean aggregation.

Robots under autonomous operation may require decisions based on evidence that is no longer visible. We study delayed-evidence tasks, where an early cue disappears before a later decision point, so visually similar observations can require different actions. In these settings, the current observation is not a sufficient state for control. We introduce TRAjectory-routed Causal Evidence (TRACE), a memory framework for visuomotor imitation policies. TRACE stores task-relevant visual and robot-state evidence, such as object identity, target choice, or route-dependent state, in a fixed-size latent memory that remains bounded over long episodes. Instead of indexing memory by raw time or manually provided task labels, TRACE uses path signatures: compact, order-sensitive features of the executed robot-state trajectory. These signatures do not store the visual cue itself; rather, they provide trajectory-conditioned keys for writing and retrieving the evidence stored when the cue was visible. When the robot later reaches an ambiguous observation, the policy conditions on TRACE memory to recover the missing context and choose the correct branch. TRACE attaches through lightweight adapters to policies, without changing the policy backbone, action head, or imitation objective. Across real-world long-horizon manipulation tasks with visually ambiguous branch points, TRACE improves branch selection and task success over alternative baselines, including short-history and recurrent memory. Project page: https://jeong-zju.github.io/trace

LLM-based guardrails have emerged as a highly effective defense against prompt injection and jailbreak attacks in autonomous agents. However, we reveal that the very reasoning and task-following capabilities enabling this protection introduce a novel vulnerability: attackers can inject crafted data to trap the guardrail in extended reasoning loops, effectuating a systematic denial-of-service (DoS) attack. To systematically expose this threat, we design a beam-search optimization framework that crafts natural-language payloads to maximize guardrail reasoning length, utilizing an LLM proposer guided by a strategy bank. Based on the observation of guardrail's schema-following nature, we also provide another attack framework driven by mechanism-aware structural mutations with less computational load. The attack efficacy is systematically evaluated in two parts. First, in standalone evaluations, the attack generalizes across diverse guardrail architectures, safety templates, and agent benchmarks. Payloads optimized on a single open-source surrogate successfully transfer to eight leading model backbones (e.g., Claude, GPT, Gemini, DeepSeek, and Qwen), achieving a 13–63$\times$ token amplification. Second, in end-to-end real-world agent deployments (web, desktop, code, and multi-agent systems), the attack reveals up to a 148$\times$ latency amplification. We show that a single poisoned document can saturate shared guardrail infrastructures, effectively starving co-located agents and paralyzing the entire system. By uncovering this availability flaw, our work underscores the urgent need to develop cost-bounded, reasoning-robust guardrails.