arXiv:2606.10856v2 Announce Type: replace Abstract: While end-to-end (E2E) autonomous driving has become the dominant research direction, production vehicles continue to rely on modular multi-NN pipelines for a non-trivial transitional period. The subject of this paper is the design of an architecture that, during this phase, supports a modular pipeline and an E2E path side by side and embeds a path for staged migration. Transplanted to a production SoC, egalitarian late fusion is compute-inefficient and offers no natural unit for staged E2E substitution. As an alternative, we propose three design principles: (i) Primary-Path, which selects a primary perception-to-planner chain and prioritizes its enclosure within a single SoC over the non-critical paths, (ii) Exposure-Time-Aligned, which propagates the primary sensor's exposure time $\tau_{\mathrm{exp}}$ as a tag along the chain and event-drives the fusion node on matched $\tau_{\mathrm{exp}}$ rather than a fixed cycle, and (iii) Co-Path Coexistence, which, building on (i) and (ii), lets an E2E output path co-run with the modular pipeline within the same $\tau_{\mathrm{exp}}$ cycle. On a Dual-SoC production AD-ECU, the implementation achieves a camera-shutter to planner-output latency of 296 ms on average within a 350 ms design budget, with the modular pipeline and the E2E path coexisting within the shared $\tau_{\mathrm{exp}}$ cycle.
Science Journals
arXiv:2606.11815v3 Announce Type: replace Abstract: Scalable distributed systems form the backbone of modern computing infrastructure. However, as scale grows, system complexity may lead to scalability faults. Scalability faults are challenging to uncover and diagnose, as they are often latent and only manifest at large-scale deployment. In this paper, we present the first comprehensive study on scalability faults and propose an approach for their detection. First, we systematically investigate 444 scalability issue reports from 10 large-scale distributed systems to understand the common anti-patterns and root causes of scalability faults. We found that the majority of these faults are caused by the synergy between dimensional code fragments and anti-patterns associated with them. Second, based on our findings, we design and implement ScaleLens, a novel approach to detect scalability faults. ScaleLens combines dynamic and static analyses to pinpoint dimensional code fragments and match them with anti-patterns. Our evaluation shows that ScaleLens detects 4.2x more dimensional code fragments associated with known scalability faults compared to the baseline. On the latest stable versions of Cassandra, HDFS, and Ignite, ScaleLens detects 334 dimensional code fragments with confirmed problematic behavior.
arXiv:2604.20702v2 Announce Type: replace Abstract: Direct satellite uplink is severely constrained by limited link budgets, which hinder the exploitation of wideband resources, and ultimately limit the throughout. This paper presents a pilot-less coded modulation scheme based on sparse superposition coding (SSC) to enable efficient wideband usage in coverage-limited scenarios. This scheme leverages the structured Zadoff-Chu quasi-orthogonal (ZC-QO) dictionary to support scalable transmission. To address decoding complexity, the SSC transmitted signal embeds root index information via indicator sequences, allowing the receiver to restrict the decoding search space. In addition, a multi-codeword transmission framework with repetition and stop-feedback is developed, enabling reliable communication and better resource utilization. Simulation results show that the proposed scheme achieves throughput gains compared to a more conventional narrow-band multi-dimensional constellation-based approach.
arXiv:2606.12913v2 Announce Type: replace Abstract: The rapid growth of modern training datasets has significantly increased computational cost, motivating dataset pruning~(DP) methods which retain only a subset of informative samples to reduce training cost. Existing pruning criteria typically rely on either intrinsic signals that assess samples independently or extrinsic signals that promote diversity via pairwise relations. While effective in their own specific regimes, each captures only one aspect of sample utility and lacks robustness across different pruning ratios or data distribution. In this work, we present a unified graph-based DP framework. By modeling the dataset as a weighted graph, where node weights encode intrinsic value and edge weights encode extrinsic value, DP can be cast as a Maximum Weight Clique Problem (MWCP). Although MWCP is NP-hard, its structure admits a principled greedy solution based on sample-wise marginal gains. Under a few mild conditions, we further prove that this unified objective enjoys a formal approximation guarantee, which applies to a broad family of importance metrics and provides practical design guidelines. Extensive experiments show that our method outperforms existing DP methods while substantially reducing training cost, reducing training time by over 40\% without sacrificing accuracy on ImageNet-1k with ResNet-50.
arXiv:2603.02070v3 Announce Type: replace Abstract: When automating plan generation for a real-world sequential decision problem, the goal is often not to replace the human planner, but to facilitate an iterative reasoning and elicitation process, where the human's role is to guide the AI planner according to their preferences and expertise. In this context, explanations that respond to users' questions are crucial to improve their understanding of potential solutions and increase their trust in the system. To enable natural interaction with such a system, we present a multi-agent Large Language Model (LLM) architecture that is agnostic to the explanation framework and enables user- and context-dependent interactive explanations. We also describe an instantiation of this framework for goal-conflict explanations, which we use to conduct a user study comparing the LLM-powered interaction with a baseline template-based explanation interface.
arXiv:2607.04839v1 Announce Type: cross Abstract: With the increasing adoption of deep learning for applications such as image compression, improvements in the rate-distortion trade-off have been achieved at the cost of increasingly larger and more opaque ''black-box'' models. Autoencoders are among the most widely used architectures for this task; however, without a clear understanding of their internal behavior, these models tend to grow in complexity to achieve more performance gains. In this paper, we investigate whether universal behaviors can be detected from the internal operations of bias-free autoencoders through Jacobian analysis. If such behaviors exist, they may be extracted to design low-complexity image compression models inspired by high-complexity deep learning architectures.
arXiv:2607.04998v1 Announce Type: cross Abstract: Many real-world coupled dynamical systems have the interaction structure and strength that evolve or adapt over time. Here, we investigate how one can control the state of a system by tuning its temporal interaction network. We present a framework based on nonlinear optimal control, where one has control over the coupling matrix of a dynamical system. We show how to obtain the gradient of the Lagrangian function of the system using the adjoint method. We then focus on a linear time-variant system for which we illustrate the framework. Finally, we explore how the states at the nodes can be steered to target trajectories, by controlling the coupling matrix, imposing various constraint on its structure. The workflow presented here can be leveraged to steer the dynamics of systems with artificial or engineered interaction that is tunable.
arXiv:2406.04425v2 Announce Type: replace Abstract: A fundamental problem in machine learning is understanding the effect of early stopping on the parameters obtained and the generalization capabilities of the model. Even for linear models, the effect is not fully understood for arbitrary learning rates and data. In this paper, we analyze the dynamics of discrete full batch gradient descent for linear regression. With minimal distributional assumptions, we characterize the trajectory of the parameters and the expected excess risk. Using this characterization, we show that when training with any learning rate schedule and finite time horizon, the early stopped solution is equivalent to the minimum norm solution for a generalized ridge regression problem. We also prove that early stopping is beneficial for generic data with arbitrary spectrum and for a wide variety of learning rate schedules. We provide an estimate for the optimal stopping time and empirically demonstrate the accuracy of our estimate.
arXiv:2603.03283v2 Announce Type: replace Abstract: We dream of a future where point clouds from all domains can come together to shape a single model that benefits them all. Toward this goal, we present Utonia, a first step toward training a single self-supervised point transformer encoder across diverse domains, spanning remote sensing, outdoor LiDAR, indoor RGB-D sequences, object-centric CAD models, and point clouds lifted from RGB-only videos. Despite their distinct sensing geometries, densities, and priors, Utonia learns a consistent representation space that transfers across domains. This unification improves perception capability while revealing intriguing emergent behaviors that arise only when domains are trained jointly. Beyond perception, we observe that Utonia representations can also benefit embodied and multimodal reasoning: conditioning vision-language-action policies on Utonia features improves robotic manipulation, and integrating them into vision-language models yields gains on spatial reasoning. We hope Utonia can serve as a step toward foundation models for sparse 3D data, and support downstream applications in AR/VR, robotics, and autonomous driving.
arXiv:2605.09018v3 Announce Type: replace Abstract: We introduce Evolutionary Ensemble (EvE), a decentralized framework that organizes existing, highly capable coding agents into a live, co-evolving system for algorithmic discovery. Rather than reinventing the wheel within the "LLMs as optimizers" paradigm, EvE fixes the base agent substrate and focuses entirely on evolving the cumulative guidance and skills that dictate agent behaviors. By maintaining two co-evolving populations, namely functional code solvers and agent guidance states, the system evaluates agents through a synchronous race, updating their empirical Elo ratings based on the marginal gains they contribute to the current solver state. When applied to a research bottleneck in In-Context Operator Networks (ICON), EvE autonomously discovered a robust rescale-then-interpolate mechanism that enables reliable example-count generalization. Crucially, controlled ablations reveal the absolute necessity of stage-dependent agent adaptation to navigate the shifting search landscapes of complex codebases. Compared to variants driven by a fixed initial agent or even a frozen "best-evolved" agent, EvE uniquely avoids phase mismatch, demonstrating that organizing agents into a self-revising ensemble is the fundamental driver for breaking through static performance ceilings.
arXiv:2606.13156v2 Announce Type: replace Abstract: Letting a vision-language model (VLM) think longer at test time has driven much recent progress. A natural way to bring this to spatial grounding is visual self-correction: the model predicts a bounding box, sees it rendered on the image, and refines it over several steps. We build a faithful instance of this idea, Iterative Visual Thinking (IVT), with a two-phase recipe: a supervised warm-up in which the base model's own predictions serve as realistic errors that a teacher VLM turns into corrective reasoning traces (yielding training data without human annotation), followed by GRPO with a simple IoU reward. Measured the way such systems are usually reported, it works: the trained model surpasses the single-shot base by +2.4pp Acc@0.5. We show this gain is a measurement mirage. The reported number silently keeps, per sample, the trajectory step closest to the ground-truth box: an oracle that needs the very answer it predicts. Re-scored under deployable, label-free stopping rules the improvement vanishes, and the best policy is not to iterate at all: stopping at step 0 matches the base and beats every shippable rule. The cause is a verification failure, since the model can generate a better box somewhere in its trajectory but cannot identify it. Self-verification confidence correlates only weakly with correctness (r about 0.22), and a counterfactual overlay shows the loop reacts to the presence of a rendered box rather than its correctness. We distill the lesson into an honest-trajectory evaluation protocol: accuracy under fixed label-free policies plus an explicit oracle-shippable gap.
arXiv:2604.24197v2 Announce Type: replace Abstract: Frontier image generation has moved from artistic synthesis toward synthetic visual evidence. Systems such as GPT Image 2, Nano Banana Pro, Nano Banana 2, Nano Banana 2 Lite, Grok Imagine Image Quality, Qwen Image 2.0 Pro, and Seedream 5.0 Lite combine photorealistic rendering, readable typography, reference consistency, editing control, and in several cases reasoning or search-grounded image construction. These capabilities create large benefits for design, education, accessibility, and communication, yet they also weaken one of society's most common trust shortcuts: the belief that a plausible picture is a reliable record. This paper provides a source-grounded technical and policy analysis of synthetic visual risk. We first summarize the public capabilities of recent image models, then analyze public incidents involving fake crisis images, celebrity and public-figure imagery, medical scans, forged-looking documents, synthetic screenshots, phishing assets, and market-moving rumors. We introduce a capability-weighted risk framework that links model affordances to real-world harm in finance, medicine, news, law, emergency response, identity verification, and civic discourse. Our findings show that risk is driven less by photorealism alone than by the convergence of realism, legible text, identity persistence, fast iteration, and distribution context. We argue for layered control: model-side restrictions, cryptographic provenance, visible labeling, platform friction, sector-grade verification, and incident response. The paper closes with practical recommendations for model providers, platforms, newsrooms, financial institutions, healthcare systems, legal organizations, regulators, and ordinary users.
arXiv:2607.05006v1 Announce Type: new Abstract: While various works address reflective symmetry understanding in 3D data and images, pixel-level semantic left-right prediction of in-the-wild images remains challenging, due to certain difficulties including the lack of 3D information, occlusion, object pose variation, partiality, etc. In this work, we propose an unsupervised learning framework to tackle this challenge. Leveraging recent advances in vertex-wise semantic left-right understanding of 3D data, our unsupervised learning method jointly utilises 3D shape and image datasets to infer pixel-wise semantic left-right predictions in single-view images. In particular, we show that a medium-scale 3D shape dataset comprising mainly of human- and quadruped animal-like shapes, combined with diverse in-the-wild image data, are sufficient to achieve high-quality semantic left-right prediction in images, even for entirely unseen 3D object categories, such as cars or trains. Overall, our approach achieves superior performance in dense pixel-wise semantic left-right predictions on both rendered and in-the-wild image datasets when compared to existing state-of-the-art methods.
arXiv:2601.22956v2 Announce Type: replace Abstract: Large language model (LLM) research in software engineering has largely focused on tasks such as code generation and bug repair. In practice, teams often draft multiple candidate proposals for fixing an issue and then deliberate on one golden proposal for implementation. This selection requires not only assessing the issue's scope, impact, and urgency, but also a clear understanding of each proposal's strengths and weaknesses. A good selection could make issue resolution more reliable while reducing regression and operational risk, whereas a poor choice can increase risk and even cause unpredictable failures. We first conduct a manual study of real-world issues to characterize the rationales maintainers use when selecting among competing proposals. Motivated by these findings, we introduce SWE-Manager, a joint selection and synthesis approach that selects the best proposal and synthesizes a golden proposal. SWE-Manager is an 8B model trained via reinforcement learning (RL) to compare proposals, justify its choice, and synthesize a golden proposal for implementation. We view proposal selection as a reasoning task, mirroring how technical managers review competing proposals by weighing issue context and each proposal's solution without executing code or running tests. On the SWE-Lancer Manager benchmark, SWE-Manager achieves 53.21 selection accuracy and 57.75 earn rate, earning 152,750 dollars and outperforming strong baselines including GPT-5. To further evaluate the effectiveness of SWE-Manager in real-world issue resolution, we design the P2A framework, which simulates a real-world workflow where multiple proposals are drafted, reviewed, and a golden proposal is selected for implementation ...
arXiv:2602.02908v2 Announce Type: replace Abstract: Diffusion models trained on different, non-overlapping subsets of a dataset often produce strikingly similar outputs when given the same noise seed. We trace this consistency to a simple linear effect: the shared Gaussian statistics across splits already predict much of the generated images. To formalize this, we develop a random matrix theory (RMT) framework that quantifies how finite datasets shape the expectation and variance of the learned denoiser and sampling map in the linear setting. For expectations, sampling variability acts as a renormalization of the noise level through a self-consistent relation $\sigma^2 \mapsto \kappa(\sigma^2)$, explaining why limited data overshrink low-variance directions and pull samples toward the dataset mean. For fluctuations, our variance formulas reveal three key factors behind cross-split disagreement: \textit{anisotropy} across eigenmodes, \textit{inhomogeneity} across inputs, and overall scaling with dataset size. Extending deterministic-equivalence tools to fractional matrix powers further allows us to analyze entire sampling trajectories. The theory sharply predicts the behavior of linear diffusion models, and we validate its predictions on UNet and DiT architectures in their non-memorization regime, identifying where and how samples deviates across training data split. This provides a principled baseline for reproducibility in diffusion training, linking spectral properties of data to the stability of generative outputs.
arXiv:2606.13654v2 Announce Type: replace Abstract: Orbital-optimized (OO) density functional calculations provide a time-independent, variational route to electronic excitations, alternative to presently widely used time-dependent density functional theory (TDDFT) approaches. As the orbitals are optimized in a state specific way, these methods can provide a balanced description of excited states with different character, thereby overcoming several limitations of practical implementations of TDDFT. Driven by recent developments in algorithms for obtaining excited states as saddle points on the electronic energy surface, OO methods have attracted increasing interest, maturing into an active and rapidly expanding area of research. Here, the theoretical foundations of the approach are clarified and an overview of recent methodological developments in excited-state orbital optimization is provided. An overview of methods for treating open-shell singlet excited states and current approaches for computing transition properties and spectra is also provided. Finally, recent applications to molecular Rydberg, charge-transfer, and core excitations are reviewed, with the aim of assessing the present accuracy and range of applicability of OO density functional calculations with common exchange and correlation functionals.
arXiv:2606.14216v2 Announce Type: replace Abstract: Accurate and computationally efficient vehicle models are essential for motion planning of autonomous vehicles, where positional accuracy directly affects trajectory feasibility and safety. However, the positional accuracy has not been systematically evaluated against real measurements. Therefore, this paper compares the short-horizon positional accuracy of three single-track vehicle models against vehicle measurements across various driving maneuvers. Model parameters are identified through dedicated experiments with the instrumented test vehicle. Rather than identifying a single best model, this work aims to provide insight into the trade-offs between model complexity, parameterization quality, and positional accuracy for informed model selection in Model Predictive Control applications.
arXiv:2606.14721v2 Announce Type: replace Abstract: Text-to-motion generation requires modeling both global action structure and fine-grained motion dynamics from natural language. Existing approaches typically rely on either continuous diffusion models or vector-quantized discrete representations. Diffusion models generate smooth motions but lack explicit compositional structure for temporal planning, while discrete token-based methods improve controllability but compress motion into finite codebooks, losing fine-grained dynamics. We argue that this limitation stems from a representation mismatch: action semantics such as intent, phase transitions, and temporal layout are inherently discrete and compositional, whereas joint trajectories and motion dynamics are continuous and locally correlated. To address this, we propose DC-Motion, a discrete-continuous factorized framework for human motion generation. DC-Motion decomposes motion into discrete structural tokens capturing global action layout and continuous residual latents modeling fine-grained dynamics. A text-conditioned structure generator predicts discrete tokens via iterative masked modeling, and a diffusion-based residual generator produces continuous motion conditioned on the structure. Experiments on HumanML3D and KIT-ML demonstrate that DC-Motion achieves strong performance in both FID and R-Precision, outperforming representative diffusion-based and discrete-token baselines.
arXiv:2606.15198v2 Announce Type: replace Abstract: City landscapes viewed through home windows influence quality of life, yet perceptions of actual window views at the urban scale remain understudied. This study presents an approach for large-scale mapping of perceptions using 12,334 window view images (WVIs) collected from actual residential properties listed on real estate platforms in Wuhan, China, representing a rarely explored form of urban view imagery that offers advantages over the rendered or simulated window views commonly examined in previous studies. Through a non-immersive virtual reality platform, we collected 27,477 pairwise comparisons across six perceptual dimensions (e.g. preference) from 304 participants based on 499 WVIs. A hybrid neural network model was trained to predict human perceptions of all crowdsourced WVIs and map their spatial distribution. Results reveal significant spatial autocorrelation with distinct hot and cold spots across the whole city. Floor level strongly influences human perceptions: while higher floors offer more preferred and extensive window views, lower-floor windows provide residents with quiet and vivid views. An inference model further shows that window view composition matters considerably: high ratios of sky, trees, and low-rise buildings enhance people's preferences and perceptions of vividness, whereas high ratios of high-rise buildings increase perceptions of monotony and oppression. Importantly, these effects are non-linear: the excessive presence of certain elements can alter their impact on human perception. This work advances urban-scale understanding of residents' visual experiences and offers a transferable, human-centric method to inform urban planning and design aimed at improving the visual quality of window views.
arXiv:2606.15220v2 Announce Type: replace Abstract: We present an SE(3)-invariant transformer for 3D-molecule generation, the Atomic Design Transformer (ADT). ADT places atoms one at a time, autoregressively. The SE(3) invariance is achieved by tokenization: each new atom's position is encoded in a local coordinate frame of a previously placed atom. The network is a plain causal transformer. For evaluation, we introduce xTB-Validated Rate (XVR), which checks whether the molecular topology is preserved after xTB GFN2 geometry relaxation. On QM9, ADT is competitive with state-of-the-art baselines. For GEOM-Drugs, we present a quantitative benchmark of scaffold-conditional 3D generation, evaluating seven scaffolds (benzene, pyridine, pyrimidine, pyrazine, furan, thiophene, cyclohexane) on the 30-atom-truncated ground-state GEOM-Drugs dataset. XVR ranges from 11.1% (pyrazine) to 29.7% (benzene), all from a single trained model. The generated molecules are chemically diverse, and generation is fast by virtue of the plain causal-transformer backbone. Application to the 50-atom-truncated GEOM-Drugs dataset yields lower XVR while the per-bond and per-angle quality is retained. These results position ADT as a practical proposer for in silico molecular design.
arXiv:2606.16337v3 Announce Type: replace Abstract: Predictive modeling for clinical decision support requires not only strong predictive performance but also transparent decision logic. Although deep learning and tree-based ensemble methods can achieve high accuracy, their black-box nature remains a major obstacle to clinical deployment. This challenge is further compounded by common characteristics of medical data, including limited sample sizes, severe class imbalance, and feature evolution arising from changes in diagnostic criteria and clinical documentation. To address these issues, we propose Medical Heuristic Learning (MHL), an instantiation of the learning beyond gradients paradigm for clinical prediction from structured medical data. Instead of relying on neural network weight updates, MHL uses a large language model (LLM) driven workflow that integrates statistical probes, medical knowledge probes, rule synthesis, and code-level iterative refinement to optimize a deterministic and executable rule-based expert system. The resulting model is expressed not as opaque parameters, but as versioned pure Python decision rules that are explicitly interpretable, fully auditable, and clinically grounded. MHL also supports continual learning by starting from previously validated rules and iteratively revising them using updated feature information under data drift or feature evolution. Comprehensive experiments on medical datasets show that MHL achieves performance comparable to state-of-the-art methods while maintaining strong behavior in small-sample and highly imbalanced settings. The results further indicate that this explicit rule-update mechanism can help alleviate catastrophic forgetting under feature evolution. Overall, these findings suggest that non-gradient-based heuristic systems offer a transparent and adaptable alternative for high-stakes clinical decision support.
arXiv:2606.16349v3 Announce Type: replace Abstract: Safety alignment requires language models to refuse harmful requests without losing the ability to answer benign ones. Existing robustness evaluations, however, do not reveal whether a model has learned to recognize harmfulness, to activate a refusal policy, or to couple these two processes. We study this question with a dual safety-geometry protocol that measures harmfulness carriers, refusal carriers, and their coupling across aligned instruction-tuned anchors and matched Mistral-7B-v0.1 SFT/R2D2 training trajectories. The aligned anchors validate the protocol: refusal-side interventions reopen attack success more strongly than harmfulness-only interventions, while harmfulness and refusal carriers remain nearly orthogonal. Along the Mistral trajectory, R2D2 exhibits a high-coupling early phase with strong fixed-source robustness, saturated safe-prompt refusal, and collapsed benign utility. Later checkpoints move to a lower-coupling regime with partial utility recovery and reopened attack success. SFT provides an important contrast: it also reaches low coupling, but remains substantially less robust, showing that low coupling alone is not a safety guarantee. All-anchor diagnostics and sparse GCG/AutoDAN transfer experiments further show that H/R coupling is informative in the R2D2 regime, whereas SFT transfer is better summarized by drift or behavior-state measures. Causal sweeps support fixed-protocol sensitivity relative to matched unit-direction controls, but do not establish independent harmfulness and refusal pathways. These results frame harmfulness--refusal coupling as an operational diagnostic for safety-geometry dynamics under adversarial fine-tuning.
arXiv:2606.16428v3 Announce Type: replace Abstract: Effective personalized AI-assisted learning demands systems that can not only generate accurate learner-specific educational materials, but also dynamically adapt their instruction to diverse learners. However, existing educational agents have primarily focused on lecture content automation and simulations, which often fall short of modelling multimodal and embodied instructional methods tailored for the individual learner. To this end, we propose Lect\=uraAgents - a multi-agent framework that enables personalized learning through end-to-end adaptive embodied teaching. At its core, Lect\=uraAgents mirrors a professor-student relationship, in which a ProfessorAgent leads a collaborative team of specialized subordinate agents through research, planning, review, and embodied delivery of lecture contents that adapt to a learner's needs. The framework offers three main contributions: (1) a hierarchical multi-agent architecture for end-to-end personalized learning; (2) an adaptive embodied teaching mechanism, wherein the ProfessorAgent executes visible and pedagogically motivated teaching actions (e.g., handwrite, highlight, underline, etc.) over contents in a teaching environment; and (3) a Teaching Action-Speech Alignment (TASA) algorithm that employs salience-based heuristics and temporal semantic segmentation to generate coherent teaching action sequences aligned with learner profiles. We evaluate Lect\=uraAgents on diverse courses at high school, undergraduate, and graduate levels using sample-specific rubric-based analysis; with generated lecture materials and teaching actions assessed and validated by expert educators. Experimental results show consistent gains in lecture content quality, embodied teaching quality, assessment, and personalization over existing approaches, positioning Lect\=uraAgents as a pedagogically well-grounded framework for personalized learning at scale.
arXiv:2606.16494v3 Announce Type: replace Abstract: Knowledge-based visual question answering (KB-VQA) lets vision-language systems answer questions that exceed their parametric knowledge by conditioning a reader on passages retrieved from a Wikipedia-scale knowledge base. In pure-text long-context LLMs, retrieved-context use follows the U-shaped "lost-in-the-middle" effect of Liu et al. (2024): information at the start and end of context is used, the middle is lost. Whether this transfers to deployed multimodal KB-VQA is open. To close this gap, we design the first controlled probe of reader-side position dependence in multimodal KB-VQA: a gold-position protocol in which only the gold passage's prompt slot varies within question. We run it on three open-source 7B/8B VLM readers and two KB-VQA benchmarks at k up to 20. The shape flips from U to primacy: gold-at-first beats gold-at-last by 16 to 26 points on every reader-by-benchmark cell, an effect we call "Lost at the End". Three targeted ablations narrow the cause: a text-only control shows the multimodal setting amplifies an already-present text-mode primacy 2.2 to 4.5 times, and image-position and distractor-shuffle ablations together pin the locus to prompt slot 0 of the instruction-tuned reader. On a frozen reader, three retrieval-side fixes (MMR, oracle reranking, rank-based reordering) all leave the gap intact (no separable improvement). Our findings indicate that recall@k is the wrong metric for deployed KB-VQA and that closing the gap requires reader-side intervention; we release our protocol as a controlled instrument for evaluating such interventions.
arXiv:2606.16933v2 Announce Type: replace Abstract: Reinforcement learning (RL) systems often degrade when operating conditions differ from those previously encountered, reflecting distributional shifts in the underlying data-generating process. Such shifts may occur between training and evaluation, as in In-Distribution (ID) and Out-of-Distribution (OOD) generalization, or within non-stationary settings where environment dynamics evolve over time. However, the formal relationship between these views remains unclear, and existing work mainly focuses on mitigation rather than the causal origin of shift within the agent-environment interaction. This work develops a unified causal-origin taxonomy that characterizes sources of distributional shift in RL and relates ID/OOD generalization to non-stationary settings. We transfer the classical dataset-shift principle from supervised learning to RL by reformulating distributional shift in terms of the generative interaction process. Using a Partially Observable Markov Decision Process (POMDP), we decompose the interaction into structural components, including the state distribution, observation process, policy, reward, and transition dynamics, together with the shifted-time boundary. The proposed taxonomy distinguishes internal (agent-driven) and external (environment-driven) distributional shifts. The shifted-time boundary perspective further characterizes explicit, implicit, and hybrid shifts. This formulation unifies ID/OOD generalization and non-stationarity as structured changes in the underlying process. We also introduce an evaluation framework for measuring shift impact and adaptation through performance degradation and recovery metrics. By grounding distributional shift in the causal-origin structure of RL, this work supports systematic analysis of robustness under distributional shift.