arXiv:2603.28955v1 Announce Type: new Abstract: This paper presents the World-Action Model (WAM), an action-regularized world model that jointly reasons over future visual observations and the actions that drive state transitions. Unlike conventional world models trained solely via image prediction, WAM incorporates an inverse dynamics objective into DreamerV2 that predicts actions from latent state transitions, encouraging the learned representations to capture action-relevant structure critical for downstream control. We evaluate WAM on enhancing policy learning across eight manipulation tasks from the CALVIN benchmark. We first pretrain a diffusion policy via behavioral cloning on world model latents, then refine it with model-based PPO inside the frozen world model. Without modifying the policy architecture or training procedure, WAM improves average behavioral cloning success from 59.4% to 71.2% over DreamerV2 and DiWA baselines. After PPO fine-tuning, WAM achieves 92.8% average success versus 79.8% for the baseline, with two tasks reaching 100%, using 8.7x fewer training steps.

arXiv:2603.29085v1 Announce Type: new Abstract: Large language models (LLMs) remain brittle on multi-hop question answering (MHQA), where answering requires combining evidence across documents through retrieval and reasoning. Iterative retrieval systems can fail by locking onto an early low-recall trajectory and amplifying downstream errors, while planning-only approaches may produce static query sets that cannot adapt when intermediate evidence changes. We propose \textbf{Planned Active Retrieval and Reasoning RAG (PAR$^2$-RAG)}, a two-stage framework that separates \emph{coverage} from \emph{commitment}. PAR$^2$-RAG first performs breadth-first anchoring to build a high-recall evidence frontier, then applies depth-first refinement with evidence sufficiency control in an iterative loop. Across four MHQA benchmarks, PAR$^2$-RAG consistently outperforms existing state-of-the-art baselines, compared with IRCoT, PAR$^2$-RAG achieves up to \textbf{23.5\%} higher accuracy, with retrieval gains of up to \textbf{10.5\%} in NDCG.

arXiv:2603.29112v1 Announce Type: new Abstract: We introduce GISTBench, a benchmark for evaluating Large Language Models' (LLMs) ability to understand users from their interaction histories in recommendation systems. Unlike traditional RecSys benchmarks that focus on item prediction accuracy, our benchmark evaluates how well LLMs can extract and verify user interests from engagement data. We propose two novel metric families: Interest Groundedness (IG), decomposed into precision and recall components to separately penalize hallucinated interest categories and reward coverage, and Interest Specificity (IS), which assesses the distinctiveness of verified LLM-predicted user profiles. We release a synthetic dataset constructed on real user interactions on a global short-form video platform. Our dataset contains both implicit and explicit engagement signals and rich textual descriptions. We validate our dataset fidelity against user surveys, and evaluate eight open-weight LLMs spanning 7B to 120B parameters. Our findings reveal performance bottlenecks in current LLMs, particularly their limited ability to accurately count and attribute engagement signals across heterogeneous interaction types.

arXiv:2603.29176v1 Announce Type: new Abstract: Parkinson's disease (PD) affects over ten million people worldwide. Although temporal interference (TI) and deep brain stimulation (DBS) are promising therapies, inter-individual variability limits empirical treatment selection, increasing non-negligible surgical risk and cost. Previous explorations either resort to limited statistical biomarkers that are insufficient to characterize variability, or employ AI-driven methods which is prone to overfitting and opacity. We bridge this gap with a pretraining-finetuning framework to predict outcomes directly from resting-state fMRI. Critically, a generative virtual brain foundation model, pretrained on a collective dataset (2707 subjects, 5621 sessions) to capture universal disorder patterns, was finetuned on PD cohorts receiving TI (n=51) or DBS (n=55) to yield individualized virtual brains with high fidelity to empirical functional connectivity (r=0.935). By constructing counterfactual estimations between pathological and healthy neural states within these personalized models, we predicted clinical responses (TI: AUPR=0.853; DBS: AUPR=0.915), substantially outperforming baselines. External and prospective validations (n=14, n=11) highlight the feasibility of clinical translation. Moreover, our framework provides state-dependent regional patterns linked to response, offering hypothesis-generating mechanistic insights.

arXiv:2603.29318v1 Announce Type: new Abstract: Smartphone GUI agents execute tasks by operating directly on app interfaces, offering a path to broad capability without deep system integration. However, real-world smartphone use is highly personalized: users adopt diverse workflows and preferences, challenging agents to deliver customized assistance rather than generic solutions. Existing GUI agent benchmarks cannot adequately capture this personalization dimension due to sparse user-specific data and the lack of fine-grained evaluation metrics. To address this gap, we present PSPA-Bench, the benchmark dedicated to evaluating personalization in smartphone GUI agents. PSPA-Bench comprises over 12,855 personalized instructions aligned with real-world user behaviors across 10 representative daily-use scenarios and 22 mobile apps, and introduces a structure-aware process evaluation method that measures agents' personalized capabilities at a fine-grained level. Through PSPA-Bench, we benchmark 11 state-of-the-art GUI agents. Results reveal that current methods perform poorly under personalized settings, with even the strongest agent achieving limited success. Our analysis further highlights three directions for advancing personalized GUI agents: (1) reasoning-oriented models consistently outperform general LLMs, (2) perception remains a simple yet critical capability, and (3) reflection and long-term memory mechanisms are key to improving adaptation. Together, these findings establish PSPA-Bench as a foundation for systematic study and future progress in personalized GUI agents.

arXiv:2603.29500v1 Announce Type: new Abstract: Large language models (LLMs) have recently demonstrated impressive performance on complex, multi-step reasoning tasks, especially when post-trained with outcome-rewarded reinforcement learning Guo et al. 2025. However, it has been observed that outcome rewards often overlook flawed intermediate steps, leading to unreliable reasoning steps even when final answers are correct. To address this unreliable reasoning, we propose PRoSFI (Process Reward over Structured Formal Intermediates), a novel reward method that enhances reasoning reliability without compromising accuracy. Instead of generating formal proofs directly, which is rarely accomplishable for a modest-sized (7B) model, the model outputs structured intermediate steps aligned with its natural language reasoning. Each step is then verified by a formal prover. Only fully validated reasoning chains receive high rewards. The integration of formal verification guides the model towards generating step-by-step machine-checkable proofs, thereby yielding more credible final answers. PRoSFI offers a simple and effective approach to training trustworthy reasoning models.

arXiv:2603.29640v1 Announce Type: new Abstract: Can AI accelerate the development of AI itself? While recent agentic systems have shown strong performance on well-scoped tasks with rapid feedback, it remains unclear whether they can tackle the costly, long-horizon, and weakly supervised research loops that drive real AI progress. We present ASI-Evolve, an agentic framework for AI-for-AI research that closes this loop through a learn-design-experiment-analyze cycle. ASI-Evolve augments standard evolutionary agents with two key components: a cognition base that injects accumulated human priors into each round of exploration, and a dedicated analyzer that distills complex experimental outcomes into reusable insights for future iterations. To our knowledge, ASI-Evolve is the first unified framework to demonstrate AI-driven discovery across three central components of AI development: data, architectures, and learning algorithms. In neural architecture design, it discovered 105 SOTA linear attention architectures, with the best discovered model surpassing DeltaNet by +0.97 points, nearly 3x the gain of recent human-designed improvements. In pretraining data curation, the evolved pipeline improves average benchmark performance by +3.96 points, with gains exceeding 18 points on MMLU. In reinforcement learning algorithm design, discovered algorithms outperform GRPO by up to +12.5 points on AMC32, +11.67 points on AIME24, and +5.04 points on OlympiadBench. We further provide initial evidence that this AI-for-AI paradigm can transfer beyond the AI stack through experiments in mathematics and biomedicine. Together, these results suggest that ASI-Evolve represents a promising step toward enabling AI to accelerate AI across the foundational stages of development, offering early evidence for the feasibility of closed-loop AI research.

arXiv:2603.29723v1 Announce Type: new Abstract: Retrosynthetic planning is a fundamental task in organic chemistry, yet remains challenging due to its combinatorial complexity. To address this, conventional approaches typically rely on hybrid frameworks that combine single-step predictions with external search heuristics, inevitably fracturing the logical coherence between local molecular transformations and global planning objectives. To bridge this gap and embed sophisticated strategic foresight directly into the model's chemical reasoning, we introduce ReTriP, an end-to-end generative framework that reformulates retrosynthesis as a direct Chain-of-Thought reasoning task. We establish a path-coherent molecular representation and employ a progressive training curriculum that transitions from reasoning distillation to reinforcement learning with verifiable rewards, effectively aligning stepwise generation with practical route utility. Empirical evaluation on RetroBench demonstrates that ReTriP achieves state-of-the-art performance, exhibiting superior robustness in long-horizon planning compared to hybrid baselines.

arXiv:2603.29902v1 Announce Type: new Abstract: Interleaved text-and-image generation represents a significant frontier for Multimodal Large Language Models (MLLMs), offering a more intuitive way to convey complex information. Current paradigms rely on either image generation or retrieval augmentation, yet they typically treat the two as mutually exclusive paths, failing to unify factuality with creativity. We argue that the next milestone in this field is Agentic Tool Planning, where the model serves as a central controller that autonomously determines when, where, and which tools to invoke to produce interleaved responses for visual-critical queries. To systematically evaluate this paradigm, we introduce ATP-Bench, a novel benchmark comprising 7,702 QA pairs (including 1,592 VQA pairs) across eight categories and 25 visual-critical intents, featuring human-verified queries and ground truths. Furthermore, to evaluate agentic planning independent of end-to-end execution and changing tool backends, we propose a Multi-Agent MLLM-as-a-Judge (MAM) system. MAM evaluates tool-call precision, identifies missed opportunities for tool use, and assesses overall response quality without requiring ground-truth references. Our extensive experiments on 10 state-of-the-art MLLMs reveal that models struggle with coherent interleaved planning and exhibit significant variations in tool-use behavior, highlighting substantial room for improvement and providing actionable guidance for advancing interleaved generation. Dataset and code are available at https://github.com/Qwen-Applications/ATP-Bench.

arXiv:2603.29908v1 Announce Type: new Abstract: Trajectory planning for autonomous driving increasingly leverages large language models (LLMs) for commonsense reasoning, yet LLM outputs are inherently unreliable, posing risks in safety-critical applications. We propose C-TRAIL, a framework built on a Commonsense World that couples LLM-derived commonsense with a trust mechanism to guide trajectory planning. C-TRAIL operates through a closed-loop Recall, Plan, and Update cycle: the Recall module queries an LLM for semantic relations and quantifies their reliability via a dual-trust mechanism; the Plan module injects trust-weighted commonsense into Monte Carlo Tree Search (MCTS) through a Dirichlet trust policy; and the Update module adaptively refines trust scores and policy parameters from environmental feedback. Experiments on four simulated scenarios in Highway-env and two real-world levelXData datasets (highD, rounD) show that C-TRAIL consistently outperforms state-of-the-art baselines, reducing ADE by 40.2%, FDE by 51.7%, and improving SR by 16.9 percentage points on average. The source code is available at https://github.com/ZhihongCui/CTRAIL.

arXiv:2603.29950v1 Announce Type: new Abstract: Effective collaboration requires teams to manage complex cognitive and emotional states through Socially Shared Regulation of Learning (SSRL). Physiological synchrony (i.e., longitudinal alignment in physiological signals) can indicate these states, but is hard to interpret on its own. We investigate the physiological and conversational dynamics of four medical dyads diagnosing a virtual patient case using an intelligent tutoring system. Semantic shifts in dialogue were correlated with transient physiological synchrony peaks. We also coded utterance segments for SSRL and derived cosine similarity using sentence embeddings. The results showed that activating prior knowledge featured significantly lower semantic similarity than simpler task execution. High physiological synchrony was associated with lower semantic similarity, suggesting that such moments involve exploratory and varied language use. Qualitative analysis triangulated these synchrony peaks as ``pivotal moments'': successful teams synchronized during shared discovery, while unsuccessful teams peaked during shared uncertainty. This research advances human-centered AI by demonstrating how biological signals can be fused with dialogues to understand critical moments in problem solving.

arXiv:2603.28784v1 Announce Type: cross Abstract: This Data Descriptor presents a fully open, multi-modal dataset for estimating vertical ground reaction force (vGRF) from consumer-grade Apple Watch sensors with laboratory force plate ground truth. Ten healthy adults aged 26--41 years performed five activities: walking, jogging, running, heel drops, and step drops, while wearing two Apple Watches positioned at the left wrist and waist. The dataset contains 492 validated trials with time-aligned inertial measurement unit (IMU) recordings (approximately 100 Hz) and force plate vGRF (Force\_Z, 1000 Hz). The release includes raw and processed time series, trial-level metadata, quality-control flags, and machine-readable data dictionaries. Trial-level matching manifests link recordings across modalities using stable identifiers. Of the 492 validated trials, 395 are triad-complete, containing wrist, waist, and force plate data, enabling cross-sensor analyses and reproducible model evaluation. Dataset quality is characterised through a three-phase cross-sensor plausibility and consistency framework, repeatability analysis of peak vGRF (intraclass correlation coefficient 0.871--0.990), and systematic checks of force ranges and trial completeness. Monte Carlo sensitivity analysis showed that correlation-based validation metrics were robust to single-sample timing perturbations at the IMU sampling resolution. All data are released under CC BY 4.0, with analysis scripts archived alongside the dataset and mirrored on GitHub. This resource supports reproducible research in wearable biomechanics, benchmarking of machine learning models for vGRF estimation, and investigation of sensor placement effects using widely available consumer wearables.

arXiv:2603.28803v1 Announce Type: cross Abstract: Double-Deck Multi-Agent Pickup and Delivery (DD-MAPD) models the multi-robot shelf rearrangement problem in automated warehouses. MAPF-DECOMP is a recent framework that first computes collision-free shelf trajectories with a MAPF solver and then assigns agents to execute them. While efficient, it enforces strict trajectory dependencies, often leading to poor execution quality due to idle agents and unnecessary shelf switching. We introduce CREST, a new execution framework that achieves more continuous shelf carrying by proactively releasing trajectory constraints during execution. Experiments on diverse warehouse layouts show that CREST consistently outperforms MAPF-DECOMP, reducing metrics related to agent travel, makespan, and shelf switching by up to 40.5\%, 33.3\%, and 44.4\%, respectively, with even greater benefits under lift/place overhead. These results underscore the importance of execution-aware constraint release for scalable warehouse rearrangement. Code and data are available at https://github.com/ChristinaTan0704/CREST.

arXiv:2603.28809v1 Announce Type: cross Abstract: Selecting appropriate values for the configurable parameters of Database Management Systems (DBMS) to improve performance is a significant challenge. Recent machine learning (ML)-based tuning systems have shown strong potential, but their practical adoption is often limited by the high tuning cost. This cost arises from two main factors: (1) the system needs to evaluate a large number of configurations to identify a satisfactory one, and (2) for each configuration, the system must execute the entire target workload on the DBMS, which is both time-consuming. Existing studies have primarily addressed the first factor by improving sample efficiency, that is, by reducing the number of configurations evaluated. However, the second factor, improving runtime efficiency by reducing the time required for each evaluation, has received limited attention and remains an underexplored direction. We develop WAter, a runtime-efficient and workload-adaptive tuning system that finds near-optimal configurations at a fraction of the tuning cost compared with state-of-the-art methods. We divide the tuning process into multiple time slices and evaluate only a small subset of queries from the workload in each slice. Different subsets are evaluated across slices, and a runtime profile is used to dynamically identify more representative subsets for evaluation in subsequent slices. At the end of each time slice, the most promising configurations are evaluated on the original workload to measure their actual performance. Evaluations demonstrate that WAter identifies the best-performing configurations with up to 73.5% less tuning time and achieves up to 16.2% higher performance than the best-performing alternative.

arXiv:2603.28823v1 Announce Type: cross Abstract: Scaling laws relate model quality to compute budget (FLOPs), but practitioners face wall-clock time constraints, not compute budgets. We study optimal model sizing under fixed time budgets from 5 minutes to 24 hours on consumer GPUs (RTX 4090). Across 70+ runs spanning 50M--1031M parameters, we find: (1)~at each time budget a U-shaped curve emerges where too-small models overfit and too-large models undertrain; (2)~optimal model size follows $N^* \propto t^{0.60}$, growing \emph{faster} than Chinchilla's $N^* \propto C^{0.50}$, with $\alpha = 0.60 \pm 0.07$ robustly exceeding compute-optimal across all sensitivity analyses; (3)~a \emph{dual U-shape mechanism}: short-budget U-curves arise from compute bottlenecks, while long-budget U-curves emerge from data bottlenecks (overfitting), with an intermediate regime where the U-curve temporarily disappears. These findings have immediate implications for researchers training on consumer hardware, where wall-clock time -- not FLOPs -- is the binding constraint. We release all code, logs, and 70+ experimental configurations.

arXiv:2603.28824v1 Announce Type: cross Abstract: Dataset condensation aims to synthesize compact yet informative datasets that retain the training efficacy of full-scale data, offering substantial gains in efficiency. Recent studies reveal that the condensation process can be vulnerable to backdoor attacks, where malicious triggers are injected into the condensation dataset, manipulating model behavior during inference. While prior approaches have made progress in balancing attack success rate and clean test accuracy, they often fall short in preserving stealthiness, especially in concealing the visual artifacts of condensed data or the perturbations introduced during inference. To address this challenge, we introduce Sneakdoor, which enhances stealthiness without compromising attack effectiveness. Sneakdoor exploits the inherent vulnerability of class decision boundaries and incorporates a generative module that constructs input-aware triggers aligned with local feature geometry, thereby minimizing detectability. This joint design enables the attack to remain imperceptible to both human inspection and statistical detection. Extensive experiments across multiple datasets demonstrate that Sneakdoor achieves a compelling balance among attack success rate, clean test accuracy, and stealthiness, substantially improving the invisibility of both the synthetic data and triggered samples while maintaining high attack efficacy. The code is available at https://github.com/XJTU-AI-Lab/SneakDoor.

arXiv:2603.28943v1 Announce Type: cross Abstract: This paper presents a hybrid CPU-GPU framework for solving combinatorial scheduling problems formulated as Integer Linear Programming (ILP). While scheduling underpins many optimization tasks in computing systems, solving these problems optimally at scale remains a long-standing challenge due to their NP-hard nature. We introduce a novel approach that combines differentiable optimization with classical ILP solving. Specifically, we utilize differentiable presolving to rapidly generate high-quality partial solutions, which serve as warm-starts for commercial ILP solvers (CPLEX, Gurobi) and rising open-source solver HiGHS. This method enables significantly improved early pruning compared to state-of-the-art standalone solvers. Empirical results across industry-scale benchmarks demonstrate up to a $10\times$ performance gain over baselines, narrowing the optimality gap to $

arXiv:2603.28998v1 Announce Type: cross Abstract: As Large Language Models (LLMs) and multi-agent AI systems are demonstrating increasing potential in cybersecurity operations, organizations, policymakers, model providers, and researchers in the AI and cybersecurity communities are interested in quantifying the capabilities of such AI systems to achieve more autonomous SOCs (security operation centers) and reduce manual effort. In particular, the AI and cybersecurity communities have recently developed several benchmarks for evaluating the red team capabilities of multi-agent AI systems. However, because the operations in SOCs are dominated by blue team operations, the capabilities of AI systems & agents to achieve more autonomous SOCs cannot be evaluated without a benchmark focused on blue team operations. To our best knowledge, no systematic benchmark for evaluating coordinated multi-task blue team AI has been proposed in the literature. Existing blue team benchmarks focus on a particular task. The goal of this work is to develop a set of design principles for the construction of a benchmark, which is denoted as SOC-bench, to evaluate the blue team capabilities of AI. Following these design principles, we have developed a conceptual design of SOC-bench, which consists of a family of five blue team tasks in the context of large-scale ransomware attack incident response.