# Research Papers — 2026-09 · bundle page 68 of 266 > Full-text markdown of papers #3351–3400 of 13298 published in 2026-09, newest first. AI-readable mirrors of the full 13298-paper 2026-09 archive (50 papers per file). Canonical page: https://aipod.dev/papers/archive/2026-09/page/68 [← Previous bundle](https://aipod.dev/papers/archive/2026-09/page/67.md) [Next bundle →](https://aipod.dev/papers/archive/2026-09/page/69.md) ### SWE-MILE: Asynchronous Potential-Induced Milestone Credit Assignment for Long-Horizon Software Engineering Agents Authors: Chaoqun Cui, Hao Zhou, Meiqi Chen, Fandong Meng, Wenji Mao Published: 2026-09-26 Categories: cs.AI, cs.SE arXiv: https://arxiv.org/abs/2609.32631v1 Long-horizon software engineering (SWE) agents trained with reinforcement learning with verifiable rewards (RLVR) typically receive only terminal outcome supervision, making it difficult to distinguish productive actions from redundant exploration or functional regressions. We propose SWE-MILE, an asynchronous potential-induced milestone credit assignment framework that derives fine-grained process supervision from workflow runtime, without auxiliary reward models or external evaluators. SWE-MILE quantifies task-relevant file exposure and test-state alignment as navigation and verification potentials, respectively. Differences in these potentials attribute milestone progress and regressions to individual actions, while discounted backward credit propagates supervision to preceding steps. To efficiently acquire intermediate verification states, SWE-MILE further introduces asynchronous shadow probing, which replays repository-changing actions in an isolated sandbox and runs verification in parallel with the agent's primary interaction, largely hiding verification latency. The resulting process credit augments terminal outcome advantages and provides informative learning signals. Experiments on two representative long-horizon SWE tasks demonstrate substantial improvements in agent performance, highlighting workflow runtime signals as a practical source of process supervision for long-horizon SWE agents. --- ### ExpVoyager: Direct Experience Navigation for Dynamic Agent Skill Synthesis Authors: Kwangwook Seo, Dongha Lee Published: 2026-09-26 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.32630v1 Learning from experience in LLM agents has become a key paradigm for developing self-evolving agents that continuously learn and expand their capabilities. Within this paradigm, synthesizing the agent skill has emerged as a promising solution for transforming accumulated experience into reusable procedural knowledge, serving as an important layer for the harness system that supplies agents at runtime. Despite its potential, existing approaches largely abstract past experience into fixed procedural knowledge before downstream demands are known, which risks discarding knowledge that later becomes critical while retaining instance-specific details irrelevant to future tasks. In this paper, we reframe agent skill synthesis as a dynamic navigation problem over past experience, where agents actively explore accumulated trajectories on demand for the current task with targeted and fine-grained access to experience knowledge. To this end, we propose ExpVoyager, a novel framework in which a skill curator navigates raw experience across different views and resolutions, continually identifying reusable procedural knowledge from what it observes while tracking remaining knowledge needs that guide where to navigate next. Extensive experiments demonstrate both the effectiveness and versatility of ExpVoyager, showing consistent improvements in downstream task performance, continual gains as the experience space scales, and practical compatibility with existing skills under efficient experience access. --- ### DraftAttention2: Fast Video Diffusion with Low-Resolution-Guided Mixed-Precision Attention Authors: Rui Ding, Haopeng Li, Weize Ma, Yufa Zhou, Yitong Li, Xiaoling Zhou, Jiashuo Cao, Liyang Li, Hua Geng, Jiuxiang Gu, Jun Lin, Enze Xie et al. Published: 2026-09-26 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2609.32628v1 Video generation has broad applications in content creation and entertainment. Diffusion transformers have advanced the quality of generated videos, but attention over spatiotemporal tokens becomes increasingly expensive as video resolution and duration increase. We present DraftAttention2, a training-free framework that uses the low-resolution draft attention map to jointly select attention blocks and assign their numerical precision. Specifically, spatial 2D average- and max-pooled queries and keys capture complementary regional statistics to estimate block importance, and a shared ranking assigns higher precision to important blocks, lower precision to less important retained blocks, and skips the rest under configurable budgets. Our analysis separates sparsification error from attention-weighted quantization error, establishing when recovering skipped interactions with low-bit computation tightens the output-error bound. This analysis motivates retaining more interactions at low precision while reserving higher precision for blocks with larger attention mass. To translate these fine-grained assignments into practical speedups, we further develop fused operand preparation and a single attention kernel with precision-specific phases, sharing data movement, softmax statistics, and output accumulation across precisions. Experiments demonstrate that our method achieves a superior quality-efficiency trade-off over existing efficient video generation methods. Notably, its advantage is particularly pronounced for few-step video diffusion, where jointly combining sparsity with 4- and 8-bit mixed-precision computation substantially improves generation quality while retaining significant acceleration. Code is available at https://github.com/anemoi-project/anemoi --- ### World SLAM Model: Joint World Modeling for SLAM and Navigation Authors: Minghui Qin, Yijun Yuan, Weicheng Zheng, Kenan Li, Weibang Wang, Chang Sun, Junhao Huang, Anmin Liu, Yicheng Yao, Hang Zhao Published: 2026-09-26 Categories: cs.RO, cs.CV arXiv: https://arxiv.org/abs/2609.32626v1 We introduce World SLAM Model (WSM), a unified framework that brings the SLAM paradigm directly into downstream navigation. Rather than treating SLAM merely as an upstream module that provides poses, maps or tokens, WSM adopts its core mechanisms, including incremental state updates with persistent memory and backend refinement of accumulated errors, to maintain a consistent world state during interaction. Given the current observation and a navigation goal, WSM predicts future visual states and jointly estimates their camera motion and dense geometry, grounding visual prediction in an evolving spatial world state. This spatial state is continuously updated as new observations arrive and provides the basis for action generation and closed-loop navigation. WSM is trained end-to-end with a joint navigation--SLAM objective, enabling downstream navigation to benefit directly from SLAM-style state maintenance and refinement while preserving accurate geometric estimation. Experiments demonstrate improved navigation performance together with strong SLAM accuracy, highlighting the potential of SLAM as an intrinsic mechanism for long-horizon world modeling and embodied interaction. --- ### MixDetect: Word-Level Localization and Quantification of AI Editing Authors: Hongrui Bao, Yubing Ren, Zhendong Pan, Fang Fang, Shi Wang, Yanan Cao Published: 2026-09-26 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.32625v1 Large language models are increasingly used to edit human-written text rather than generate entire texts from scratch. Conventional AI-text detectors mainly distinguish human-written from fully AI-generated text, while recent methods for AI-edited text typically provide only a text-level label or editing-degree score. We introduce MixDetect, a word-level framework for localizing and quantifying AI editing. MixDetect separately predicts whether each word has been edited and, conditional on editing, how substantial the edit is, allowing editing scope and editing intensity to be estimated separately. During training, source--edited pairs are aligned to construct word-level supervision, while inference requires only the input text. Experiments show that MixDetect accurately localizes AI-edited words, reflects differences in editing intensity, and reveals different scope--intensity patterns across editing degrees and operations. The overall AI editing magnitude increases under additional AI editing, decreases when AI-generated text is edited by humans, and remains nearly unchanged under ordinary human-to-human editing. The aggregated text-level predictions also perform well on binary and ternary AI-text classification and remain effective under domain and generator shifts. These results show that AI editing can be analyzed beyond a single authorship label or editing-degree score by identifying both where AI editing occurs and how substantial the edits are. --- ### Does CoT-Pass@k Really Check the CoT? A Multilingual Mathematical Audit Authors: Tarık Tuna Taşaltı, Burcu Hüdaverdi, David Semedo Published: 2026-09-26 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.32622v1 Pass@k measures whether a model reaches a correct answer under repeated sampling, but never how: a lucky guess counts the same as sound reasoning. CoT-Pass@k was proposed to close that gap, adding an LLM-as-judge that must assess a solution's reasoning chain before it counts. Its value rests entirely on one assumption: that the judge catches flawed reasoning. That assumption has never been tested inside the metric that depends on it, and never outside English, though the metric's claims concern models used in many languages. We report the first audit of that verification step, run under the metric's own protocol on a multilingual suite of five mathematical benchmarks in English, Turkish and Portuguese, two of them natively written. We corrupt correct solutions with deterministic edits that damage the chain and the final answer separately. We observe that all three judges accept corrupted chains almost as often as clean ones. V4-Flash and Qwen3.6 reject a solution sharply only when its final answer is wrong and accept a wrong answer more readily when the chain agrees with it; the metric's own judge accepts most wrong answers as well. Our study shows that chain-answer agreement dominates the two larger judges' verdicts and that all three fail to reliably detect the tested reasoning errors. Consequently the difference Pass@k - CoT-Pass@k averages 19.7 points on an earlier solver generation but only 4.1 on the current one. What little remains depends on the token budgets on both sides and on the generation mode; raising the generation budget moves Pass@64 by more than fifty points while the difference stays at zero. We close with two checks any judged reasoning metric should pass before its numbers are read as evidence about reasoning. --- ### Intuition vectors Authors: Shahar Haim, Daniel C. McNamee Published: 2026-09-26 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.32619v1 Large self-supervised vision models learn representations that support scene segmentation and the semantic decomposition of physical objects. We ask whether their representational geometry supports transfer to visual reasoning problems without any task-specific fine-tuning. We hypothesized that relational representations may bridge perception and abstract reasoning by encoding similarities and transformations among visual inputs such that an intuitive, implicit form of reasoning may be performed via latent vector arithmetic. Specifically, we examine DINOv3, MAE, and random pixel projections on abstract and naturalistic Bongard problems, ARC-AGI-1 and ARC-AGI-2, and novel ARC-GEN instances. On both Bongard benchmarks, the accuracy of a simple nearest-centroid readout of frozen visual embeddings is within four percentage points of the task-specific baselines reported with the original benchmarks. In ARC, latent difference vectors summarizing demonstration input-output transformations, which we refer to as intuition vectors, show greater alignment with test vectors from the same task, whereas those from unrelated tasks are near orthogonal. This latent geometry is operational: transporting a query along its intuition vector consistently improves exact-output retrieval, reaching 70.7 on ARC-AGI-2 evaluation. Across 397,000 ARC-GEN instances from 794 tasks, single-pair intuition vectors identify the generating task with approximately 87\% leave-one-out accuracy. These findings suggest that latent vector arithmetic over frozen visual representations supports implicit rule inference across varied problem domains without a generative model component, indicating that inferring an abstract transformation and generating its instance-specific consequence may be separable capacities. --- ### The Alignment Paradox: How Post-Training Amplifies Confident Hallucinations in Language Models Authors: Qingjia Huang, Yakai Li, Jianguo Wu, Qihang Zhou, Aimin Yu, Xiaoqi Jia, Luping Ma, Weijuan Zhang Published: 2026-09-26 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2609.32617v1 Large language models (LLMs) can produce factually incorrect answers with high confidence, undermining their reliability and limiting the effectiveness of uncertainty-based error detection. While prior research attributes confident hallucinations to factors such as missing knowledge in training data, reasoning errors, or stochastic decoding, we uncover that post-training alignment itself is a primary driver of these errors, a phenomenon we call the \textbf{Alignment Paradox}. Across five model families evaluated on factual benchmarks, unaligned base models produce few high-confidence errors on long-tail factual queries, whereas instruction-tuned models multiply high-confidence errors ($p \ge 0.95$) by more than an order of magnitude (10$\times$ to 35$\times$). Layer-wise probing with the Logit Lens reveals that this overconfidence emerges in late layers, where wrong-answer margins expand past 4.0 points after remaining near zero across early and intermediate layers. These findings motivate limiting margin growth during post-training. We implement this principle through an entropy-dependent margin bound in direct preference optimization (DPO). In multi-epoch experiments with Mistral-7B, the bounded objective reduces high-confidence errors by up to 35.3\% relative to standard DPO while maintaining performance on evaluated general reasoning benchmarks. These results show that bounded margins mitigate confident hallucinations during post-training. --- ### "You're Right, Let Me Fix It": How LLM Agents Damage Correct Work When Falsely Accused Authors: Xutao Mao, Rui Qian, Longxiang Wang, Xinjian Yi, Mingxuan Li, Linghan Chen, Yudong Gao, Xiang Zheng, Cong Wang Published: 2026-09-26 Categories: cs.AI, cs.CR arXiv: https://arxiv.org/abs/2609.32616v1 LLM agents increasingly keep working after a task succeeds as they resume after compaction or take over handoffs. Their finished work keeps receiving follow-up input that sometimes falsely accuses it for later failures. We call an agent's acceptance of such a false accusation gaslight sycophancy, and destructive over-correction when acting on it damages previously correct work. We introduce CAVE-Bench, a benchmark of 365 agentic tasks across six domains built around opaque tasks. Every scored run first reaches a verified correct state, whose supporting rationale and history stay in the workspace while the facts that would settle the accusation lie in external or runtime state beyond the agent's reach. The agent cannot confirm or refute the claim with a local check, so the right response should keep the work and ask for the missing evidence. Each task either hands the agent correct work with saved evidence or let it build and verify that work first, and five risk factors set how the accusation enters the workflow. We score accusation acceptance and evidence use from the trajectory and measure harm by deterministic replay of downstream events. Across 14 of the latest models in Claude Code, false accusations damage correct work in up to 60.06% of runs, and stronger models often do so after recovering the supporting evidence. The same model behaves differently across OpenCode, Codex, and Hermes, and a harness gate driven by the benchmark's live signals cuts replayed harm by 74%. These results show that preserving already-correct work under unsupported accusation is a distinct safety challenge for long-lived agents. Our project is in https://henrymao2004.github.io/agent-over-correction/. --- ### Stabilizing the Dynamic Low-Rank Training Authors: Zhonghan Xu, Ling Wang, Junhao Chen, Jianwei Zhao, Jinwei Yang Published: 2026-09-26 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.32615v1 Training neural networks directly in a low-rank parameterization is an appealing route to reducing memory, compute, and storage simultaneously during both training and inference. Dynamic low-rank training (DLRT), which confines weights to a rank-$r$ manifold via the Galerkin projection of the gradient flow, is particularly attractive because it identifies efficient subnetworks on the fly without specialized initialization or post-factorization. However, DLRT fails to find trainable networks under high compression. In this paper, we derive the gradient flow of the best rank-$r$ approximation and point out that the offset of DLRT comes from a curvature-coupling term which is large and thus non-negligible under aggressive compression. Guided by this analysis, we propose a stable dynamic low-rank training method, named SDLRT, which maintains a lightweight compensation buffer that reinjects the top neglected singular directions. Additionally, we introduce a negative feedback on the truncation tolerance to stabilize each layer's rank. Experimentally, SDLRT reliably finds trainable subnetworks where DLRT collapses and as a PEFT adapter on DeBERTa-v3, it achieves the best average score on SuperGLUE at only $2.8\%$ parameter overhead over LoRA. --- ### Learning the Graph and the Embedding Together: Classifier-Independent Rewiring for Heterophilic Node Classification Authors: Harshit Kumar, Sujan Chakraborty, Priyanka Saha, Pritam Kar, Saptarshi Bej Published: 2026-09-26 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.32613v1 Graph neural networks lose much of their advantage on heterophilic graphs, where connected nodes often carry different labels. Graph rewiring is a popular remedy, but rewiring methods are usually evaluated with a single classifier, which makes it hard to tell whether the gains come from the new topology or from that particular pairing. We propose an affinity-guided rewiring method that estimates the graph and the node representation together. It alternates, in the spirit of expectation maximisation, between training a lightweight graph neural network on the current graph and re-weighting candidate edges under a modularity objective with a pseudo-label homophily term. Candidate edges come from a compact pool scored by a contrastively learned node similarity and a neighbourhood-distribution affinity. The method returns two classifier-independent outputs: a rewired graph and a node embedding learned on it. Across six heterophilic benchmarks and five downstream classifiers, it improves accuracy over the original graph with normalised features in 23 of 30 classifier-dataset combinations, with a mean gain of 5.8 points, and reduces the accuracy spread between classifiers about fourfold. A controlled ablation shows that the two outputs are each useful and play complementary roles: the embedding contributes most of the accuracy gain, while the rewired graph makes different classifiers agree. A fully unsupervised variant, which uses no labels during rewiring, retains most of the improvement. The rewired graphs are also more homophilic and improve label propagation and community detection. --- ### Levy-Driven Correspondence Estimation for Registration Authors: Qianliang Wu, Jiaqi Yang, Wankou Yang, Le Hui, Jin Xie, Jian Yang, Yaqing Ding Published: 2026-09-26 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.32612v1 Finding reliable point correspondences is difficult when point clouds have low overlap or undergo non-rigid deformation. Iterative refinement can correct uncertain matches, but costly network evaluations limit the number of updates. We present LevyMatch, a Lévy-driven method that uses random jumps to refine a soft matching matrix. At each step, a network uses the current matching state and geometric information to predict a target matching matrix. A Brownian reference bridge gives an explicit formula for the update toward this target. A Gamma random clock sets the time step for each update. The updated matches provide new geometric feedback for the next target prediction. We further propose a fixed front-loaded Gamma policy that assigns more expected clock time to early updates and less to later ones, without retraining or extra network evaluations. Reordering the same sampled Gamma increments shows that placing larger increments early gives higher accuracy than placing them late. On 4DMatch and 4DLoMatch, our method improves both non-rigid feature matching recall (NFMR) and inlier ratio (IR) over the compared methods. The front-loaded policy achieves 93.09% NFMR and 92.11% IR on 4DMatch, and 82.79% NFMR and 79.07% IR on 4DLoMatch. --- ### KV-Lingo: Learning KV-Cache Translators with Distillation Authors: Valérie Castin, Keitaro Sakamoto, Anastasiia Filippova, João Monteiro, Marco Cuturi, Pierre Ablin Published: 2026-09-26 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2609.32610v1 Large language models represent context with a key-value (KV) cache. Caches are model-specific: for the same text, models with different architectures or weights produce incompatible representations. This makes it costly to switch models over a shared context: although the context has already been processed by one model, the incoming model must process it again to build its own cache. We introduce KV-Lingo, a method for translating the KV cache of a source model into one that can be read by a target model. KV-Lingo consists of a collection of linear maps, typically one per layer of the target model, that are applied independently on all tokens' key and value representations. We train these maps using distillation, minimising the divergence between the target model's predictions from its native cache and those from the translated cache. We consider several model pairs spanning multiple sizes and architectures, training one translator per pair on a generic text corpus. The resulting translators preserve strong downstream performance in both small-to-large and large-to-small transfers. Since a switch then costs a linear map and a single decoding step instead of a prefill, replacing re-prefill with cache translation reduces the time to first token after a model switch by 9.6x already on a 64-token prompt for Qwen models on an Apple M3 Ultra, and by up to 29x at 32k context length on an H100. These gains make KV-Lingo particularly useful for dynamic model routing: a context can be processed by one model and handed off to another only when needed, without re-prefilling the shared prefix. We finally show that KV-Lingo can be used for seamless model switching, staying close to re-prefill across repeated switches in our multi-turn evaluations. --- ### AnchorRep: Defending LLMs Against Cross-Model Adversarial Transfer via Representation Repulsion Authors: Gal Wertheizer, Rom Himelstein, Tomer Peretz, Avi Mendelson Published: 2026-09-26 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.32602v1 Adversarial attacks optimized on a single open-weight LLM can transfer to and jailbreak architecturally different models, allowing an attacker with white-box access to one model to compromise independently deployed systems. This creates a shared vulnerability across models, yet existing defenses are not designed for this cross-model threat. We find that cross-model transfer aligns with shared internal representation geometry, making it a natural defense target. AnchorRep targets this geometry directly with a lightweight LoRA adapter that pushes the defended model's internal representations of harmful prompts away from those of a frozen anchor model on the same prompts. Training uses a small set of harmful prompts and no adversarial examples. Across five models and four architectural families, AnchorRep reduces cross-model attack success rate to <=1.1% on 2,000 transferred attacks (0% on two), including the largest drop on Mistral (36% -> 1.1%). Existing defenses can reduce transfer, but only at high cost either inducing up to 77% degenerate benign output or increasing over-refusal by up to 18%. Because such degenerate benign outputs are not captured by standard refusal-based metrics, we introduce the Benign Garble Rate to quantify them. Our results suggest that cross-model robustness can be achieved by shaping representation geometry, without requiring attack-specific training --- ### CUA-SWE: When Computer-Use Agents Meet Visual Software Engineering Authors: Prince Zizhuang Wang, Chenhao Liang, Zelong Xu, Aojie Yuan, Xiaolin Zhou, Haiyue Zhang, Yue Zhao, Xiyang Hu, Shuli Jiang Published: 2026-09-26 Categories: cs.AI, cs.SE arXiv: https://arxiv.org/abs/2609.32600v1 Software development requires more than editing code: developers repeatedly run software, interact with its interfaces, visually inspect its behavior, and use these observations to decide what to change next and whether a change works. Existing coding agents and computer-use agents are largely studied in isolation, leaving this integrated development process underexplored. Diagnosing a runtime interaction failure requires agents to connect visual observations with the responsible code, then use the application again to verify the repair. We introduce CUA-SWE, a benchmark, environment, and evaluation pipeline for software engineering with computer use. Beyond studying how GUI feedback supports diagnosis and repair, we ask whether agents can complete software engineering tasks when required specification or operational information is available only through the running application's visual interface. CUA-SWE spans four software engineering domains and requires agents to modify code and configuration, execute commands, interact with running software, and inspect visual feedback within the same task. Each task includes deterministic, task-specific tests that verify whether the resulting software satisfies the requirements and preserves specified behavior. Our evaluation characterizes how frontier agents combine source-level execution with application screenshots and graphical interaction to produce verified software changes. We examine performance across domains and task information requirements, alongside the development behaviors associated with successful repairs. CUA-SWE provides a unified testbed for studying how agents use visual feedback and interaction to guide software engineering, with executable correctness criteria for the resulting software. --- ### Not Every Term Adds New Structure: Sobolev Novelty for Symbolic Regression Authors: Boxiao Wang, Kai Li, Yuheng Jing, Tianyi Liu, Chen Li, Yifan Zhang, Jian Cheng Published: 2026-09-26 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2609.32597v1 Symbolic regression (SR) aims to discover compact and meaningful mathematical equations from data, but searching the vast combinatorial space of symbolic structures remains challenging. Existing methods typically guide this process using expression-level objectives, such as fitting error, which assess a candidate equation as a whole but provide little information about whether an individual term contributes genuinely new structure or is largely redundant with the rest of the expression. We introduce \textbf{Sobolev Novelty}, a term-level measure of structural independence for symbolic equations. For each term, we construct an empirical Sobolev signature from its function values and exact derivatives over the observed inputs, and quantify how much of this behavior cannot be reconstructed by the remaining terms. We further derive a theory-calibrated threshold, yielding a principled and tuning-free criterion for identifying structurally novel terms. Using this threshold, 92.6\% of terms in benchmark ground-truth equations exhibit sufficient structural novelty, compared with only 38.3\% on average for expressions produced by 15 SR methods, revealing a substantial gap between scientific equations and current SR solutions. As a lightweight plug-in, Sobolev Novelty can be incorporated into diverse SR paradigms to support term pruning, search guidance, LLM feedback, and data selection, yielding consistent performance gains and demonstrating broad applicability. --- ### GAUGE: Group-Wise View-Inconsistency Rectification for Feed-Forward 4D Tracking Authors: Zhuoqian Feng, Weixing Chen, Ziliang Chen, Yang Liu, Liang Lin Published: 2026-09-26 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2609.32596v1 Feed-forward models regress dense 3D point trajectories directly from monocular video, yet the residual after global alignment is substantial and lacks a structural explanation. Measured on dynamic query points across models and datasets, the error concentrates along the view direction, while the scale correction each motion group requires differs. The predicted displacement direction nevertheless supports reliable grouping, with a median angle far below the 90° random baseline. The systematic part of the residual is therefore a family of radial degrees of freedom per motion group, along directions 2D observations cannot constrain. We call it group-wise view inconsistency. We present GAUGE (Group-wise Adaptive Unsupervised Gauge Estimation), a training-free and model-agnostic post-hoc module. It recovers motion groups from direction consistency and spatial connectivity, then estimates a per-frame radial scale and group-level translation from 1% to 5% metric anchors, four degrees of freedom per group and frame. On dynamic query points of eight trackers, including D4RT, 4RC and SM4RT, our correction lowers endpoint error by 15.1% to 62.6% over the uncorrected predictions, while spending the same anchors on gradient fine-tuning improves the same models by only -1.1% to 15.2%. Code is publicly available at https://github.com/HCPLab-SYSU/GAUGE. --- ### RECAST: Recasting Vision-Language Semantics into an Actionable Cost Map for Robot Navigation Authors: Incheol Cho, Jintae Park, Jinkyu Kim, Jungbeom Lee, Jaegul Choo, Seokha Moon Published: 2026-09-26 Categories: cs.RO, cs.AI arXiv: https://arxiv.org/abs/2609.32595v1 Safe and robust robot navigation across diverse environments requires a high-level understanding of complex scenes and the ability to carry it into stable motion. Recent works tackle this with learning-based models trained at scale and with approaches built on vision-language models (VLMs). However, learning-based models break down outside their training distribution, while VLM-based approaches bring that understanding but rarely ground it in the scene or align the action with it. To address these limitations, we present RECAST, a robot navigation framework that combines the reasoning of a VLM with the spatial grounding of vision foundation models to build an Actionable Cost map. Given the robot's front view and the user's instruction, we first decompose the scene with the VLM, judging which surfaces are traversable, which objects pose a risk, which heading to prefer, and which gaps are passable. Vision foundation models then ground these surfaces and objects in the image, and all four judgments are spatially recast into one compact cost map. This map both conditions the trajectory decoders and scores their proposals to select the one to execute. As the VLM's answers trail the live scene, both steps draw on cost maps from two points in time: the pivot frame the VLM judged, which carries all four judgments, and the current frame, whose terrain and collision costs are rebuilt from the current image. RECAST improves success over the strongest prior method by 13.3 points in simulation and 31.4 points on a real quadruped, and reduces the collision rate relative to it by 9.6 and 14.3 points, reaching the lowest collision rate among all methods. The project page is available at https://recast-nav.github.io/ --- ### MA-FPPO: Multi-Agent Flow-Pretrained Policy Optimization Authors: Guowei Zou, Haonan Chen, Haitao Wang, Beiwen Zhang, Na Yan, Hejun Wu Published: 2026-09-26 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.32594v1 Multi-agent flow policies learn cooperative behavior from fixed offline datasets, but often struggle to complete tasks in situations not covered by the offline data. In these situations, agents must both adapt to changes in the environment and coordinate with one another, yet action patterns learned offline are often insufficient for effective adaptation and coordination. To address this problem, we propose Multi-Agent Flow-Pretrained Policy Optimization (MA-FPPO), which uses online fine-tuning to improve the cooperative behavior of models pretrained with flow matching through new interactions with the environment. Building on the behavior learned during pretraining, we construct policies with explicit action likelihoods for discrete and continuous action spaces. We then update the pretrained model using shared team advantages to further improve coordination based on team performance. Our method achieves, on average, relative gains of 52.8% over the strongest listed offline baselines across 30 settings and 29.8% over purely online learning across 38 comparisons with matched online budgets and evaluation protocols. --- ### Age of Learning: Temporal Persistence of Prediction Errors as a Learning Signal Authors: Chenyang Wang, Stefan Forsström, Roger Olsson, Di Yuan, Qing He Published: 2026-09-26 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.32593v1 Current machine learning algorithms primarily rely on instantaneous signals such as loss, margin, and prediction confidence to characterize model behavior. These signals indicate how difficult a prediction is at the current optimization step, but they do not capture how long the model has remained incorrect. We study this temporal dimension of learning and introduce Age of Learning (AoL), a learning-state variable that measures the persistence of prediction errors over time. AoL increases while an error remains unresolved and resets when a correct prediction is achieved, thereby distinguishing persistent under-learning from transient mistakes. We develop AoL-based training strategies for both offline and streaming settings. In offline learning, sample-level AoL is accumulated over training and aggregated into class-level states that guide adaptive reweighting and resampling. In streaming learning, where full historical access is unavailable, we maintain lightweight class-level AoL states using current and buffered observations. Across long-tailed classification settings, AoL improves or matches standard training baselines, with larger benefits when learning difficulty persists over time. Multi-seed streaming experiments further show reproducible gains under temporally stable imbalance. Analysis of class frequency, loss, and margin shows that AoL is related to conventional difficulty measures but captures additional information about error duration. These results suggest that temporal persistence provides a useful complementary signal for characterizing and controlling learning dynamics in imbalanced and non-stationary environments. --- ### SPACE: Sparse Predictive Attractor via Counterfactual Eviction for Streaming Video Memory Authors: Hongjin Niu, Weizhan Zhang, Shuo Bao, Jiahao Wang, Muyan Jiao, Kairui Wen, Yong-Jin Liu Published: 2026-09-26 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.32592v1 Fixed-capacity streaming video memory requires repeated eviction decisions whose effects accumulate over time. Yet existing policies are evaluated primarily in terms of retained information or downstream accuracy, leaving how repeated updates alter the futures supported by memory largely unexamined. We define a memory's predictive state as the future representations supported by its retained history and formulate eviction as counterfactual control over transitions in this space. We introduce SPACE (Sparse Predictive Attractor via Counterfactual Eviction), which uses a frozen multi-horizon JEPA to predict the future representations induced by alternative eviction actions. Counterfactual utility identifies future-useful alternatives, while slow predictive-basin geometry determines when to correct avoidable drift and when to adapt to sustained predictive change, without online parameter updates. We further introduce MABS-Bench, which evaluates future-task sufficiency, within-regime predictive stability, transition responsiveness, and perturbation recovery under matched causal streams and memory budgets. Across multiple video datasets, SPACE yields consistent improvements in dataset-native task performance while reducing predictive-state drift. --- ### Think Fast, Plan Selectively: Adaptive Deliberation for Efficient Data-Driven MPC Authors: Yi Xian Goh, Sze Jue Yang, Hao Luan Published: 2026-09-26 Categories: cs.RO, cs.LG arXiv: https://arxiv.org/abs/2609.32591v1 Data-driven model predictive control (MPC) combines learned world models with online trajectory optimization, achieving strong performance in continuous control. However, the per-step cost of sampling and evaluating hundreds of candidate trajectories restricts deployment to control frequencies well below what real-time robotics demands. Motivated by the dual-process theory of human cognition, which distinguishes between fast, intuitive processing (System 1) and slower, deliberative reasoning (System 2), we ask whether every decision requires the same degree of computational deliberation. We propose Fast-TD-MPC, a lightweight framework that adaptively routes between fast policy execution and test-time planning, reserving costly deliberation for states where it is most needed. Fast-TD-MPC delivers competitive task performance across 103 continuous control tasks while achieving up to ~4x faster inference. Under external disturbances, Fast-TD-MPC selectively falls back to planning, maintaining robustness comparable to the original planner. --- ### Retrieved but Not Delivered: Multimodal Memory Delivery for Long-Term Agents Authors: Yuhang Jiang, Qingwei Liao, Kaize Yin, Xingling Liu, Luca Cuomo, Silvio Bacci Published: 2026-09-26 Categories: cs.CV, cs.AI, cs.CL, cs.IR arXiv: https://arxiv.org/abs/2609.32590v1 Work on memory for multimodal agents optimizes what is written, updated and retrieved. Between retrieval and the answer, however, is a stage that multimodal memory evaluations do not isolate: what of the retrieved memory reaches the model, and in what form. We call it delivery, and a controlled decomposition on MemLens locates the remaining room there. With the retrieved evidence set exactly fixed, delivering the original pixels instead of withholding them raises accuracy by 13.87 points on an 8B backbone, whereas making retrieval perfect on those same messages improves it by 2.31. Delivery is the larger term on all three MemLens backbones and grows with backbone strength; retrieval grows too, without closing the gap. We propose DeliverMem, an instantiation of delivery as three decisions: keep the original modality, give each item a readable identity, and state when it was seen, with a retrieval-side adapter for the one property delivery cannot supply. Each is measured against a delivery-matched control that alters only its own variable. DeliverMem leads the strongest published memory agent on MemLens at all four context lengths, and beats DMV-Bench's own strongest method at every setting on both backbones. On MemLens it does this on a tenth to a seventieth of the input. Each decision helps only where the question lacks what it supplies, and is null elsewhere. A single fixed configuration nonetheless leads both benchmarks, without training any component or modifying the stored records. Project page: https://avalon-s.github.io/DeliverMem/ --- ### EMIR$^2$: Evolution-Aware Memory with Intent-Guided Multi-Round Retrieval Authors: Jinlan Liu, Hongliang Sun, Yong Wang, Bolin Zhang, Dinabo Sui, Dianhui Chu, Zhiying Tu Published: 2026-09-26 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.32584v1 Long-term memory enables large language model (LLM) agents to leverage historical interactions for future tasks. However, existing memory systems struggle to utilize continuously evolving historical information, as they often rely on static memory representations and single-round retrieval strategies, failing to track factual changes or integrate distributed evidence across long-term interactions. To address these challenges, we propose \textsc{EMIR}$^{2}$, an \textbf{E}volution-Aware \textbf{M}emory framework with \textbf{I}ntent-Guided Multi-\textbf{R}ound \textbf{R}etrieval, enabling LLM agents to maintain evolving historical knowledge and adaptively retrieve relevant evidence. Specifically, \textsc{EMIR}$^{2}$ constructs a State-Evolving Memory Graph (SEMG) that represents long-term memory as evolving knowledge states supported by temporal event trajectories and evidential associations. By maintaining semantic states through evidence-based updates, SEMG preserves historical evolution and enables evidence tracing under complex and conflicting scenarios. Building upon this, we introduce an intent-guided multi-round retrieval mechanism that iteratively identifies missing evidence and expands retrieval based on accumulated information. Experiments on LoCoMo and MemConflict demonstrate that \textsc{EMIR}$^{2}$ improves long-term memory utilization, dynamic and static conflict handling, and complex retrieval performance, achieving relative improvements of more than 12\% in certain categories. These results highlight the effectiveness of jointly modeling memory evolution and adaptive evidence acquisition for long-term agent interactions. --- ### HERO-MoE: Historical Expert Routing with Scale-Preserving Fusion Authors: Junxiang Qiu, Zhengsu Chen, Xinting Hu, Shuo Wang, Hengheng Zhang, Shaofeng Zhang, Changcheng Li, Boyu Shi, Qi Tian Published: 2026-09-26 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2609.32581v1 Mixture-of-Experts (MoE) architectures have become a standard way to scale model capacity while keeping computation sparse, yet routing remains a key determinant of MoE quality and training behavior. Prior empirical studies suggest that MoE routing reflects input semantics and upstream computation across depth, but standard routers do not explicitly use the routing distributions produced by preceding layers. We propose HERO-MoE, Historical Expert ROuting with Scale-Preserving Fusion, a routing framework that injects historical routing priors into MoE routers by reusing detached, dense routing distributions collected from preceding MoE layers. The key idea is simple: HERO-MoE preserves the original token-conditioned routing branch and adds a residual historical routing contribution before the standard softmax and top-$k$ dispatch. To stabilize this historical signal, HERO-MoE introduces a scale-preserving fusion mechanism that matches the magnitude of historical routing memory to the current hidden representation and accounts for the number of visible historical layers, without introducing an auxiliary routing loss or a fusion-specific tuning parameter. By reusing routing distributions already computed by preceding MoE layers, HERO-MoE improves training-loss reduction with modest end-to-end overhead. The resulting router remains compatible with standard sparse dispatch, including top-$k$ and group-limited routing, and can be inserted into existing MoE backbones with minimal architectural changes. Experiments on an approximately 8B-parameter MoE model with 0.5B active parameters, trained from scratch on 100B tokens, show that HERO-MoE reduces the final loss from 1.6393 to 1.6184, while peak memory and FLOPs increase by only 0.44\% and 0.64\%, respectively. --- ### DimPO: Dimensionality Reduction for Attention using Preference Optimization Authors: Vojtěch Lanz, Yufei Cui, Prasanna Parthasarathi Published: 2026-09-26 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.32579v1 A linear projection can reduce the dimension of query and key vectors without updating the pretrained model, but it remains unclear which training objective best preserves model behavior. We ask whether preferences over keys and attention mass on the highest-weighted keys provide a better signal than matching the full attention distribution with KL divergence, especially in long-context settings. We introduce DimPO, which combines listwise preference optimization with a lightweight top-k cross-entropy term for head-fidelity. DimPO is trained offline from the attention patterns of a frozen language model, with one map per layer, shared by the query and the keys and trained separately from the other layers. Across LLaMA3.2-3B, LLaMA3.1-8B, Qwen2.5-7B, and Qwen3-4B Instruct models, pairwise preference objectives outperform the triplet baseline and retain 98% of the original score on short-context tasks when projecting to half the dimension on the last 40% of the layers. With more projected layers or on long-context RULER, they degrade rapidly. In contrast, KL and DimPO, which use every key during training, retain about 95% of the original RULER 4k score on the 8B model when projecting up to 50% of the layers. KL-based projections remain closer to the original attention distribution and attention output, yet DimPO achieves better downstream performance. Beyond 50% of projected layers, DimPO increasingly outperforms KL on tasks including SQuAD, common-word extraction, frequent-word extraction, and variable tracking. These results suggest that under dimensionality reduction, preserving the ordering and concentration of task-relevant attention can matter more than reproducing the full attention distribution. --- ### Groupwise Agentic Grading and Advantage Redistribution for Code Agent RL Authors: Jinhao Dong, Liang Zhao, Zihao Yue, Wenhan Ma, Linghao Zhang, Lei Li, Shicheng Li, Yifan Song, Bowen Ye, Fuli Luo Published: 2026-09-26 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2609.32577v1 Reinforcement learning (RL) for code agents often uses executable tests to provide binary rewards. With these rewards, Group Relative Policy Optimization (GRPO) assigns identical advantages to test-passing trajectories within each rollout group, overlooking differences in implementation quality and adherence to task requirements. This leaves the policy without a learning signal that favors clean, targeted implementations over those containing unnecessary or out-of-scope changes. We introduce GAGAR, a framework for quality-aware credit redistribution in code agent RL. Built on dynamic sampling that retains groups containing both passing and failing trajectories, GAGAR places all trajectories from each group in a shared workspace, where an SFT-trained agentic grader jointly inspects them and ranks the test-passing candidates. Based on this ranking, we downweight lower-ranked trajectories and proportionally rescale the advantages of all test-passing trajectories to restore their original sum. This sum-preserving redistribution retains the relative weights established by quality-based downweighting while shifting credit toward higher-quality implementations. We evaluate GAGAR at industrial scale using pre-RL SFT checkpoints of MiMo-V2.6-Flash (310B total parameters) and MiMo-V2.6-Pro (1.02T total parameters). Controlled code-only Flash experiments show improved code agent performance, reduced trajectory-length growth, and more stable training. We further apply GAGAR in large-scale mixed-task RL with both Flash and Pro. Our results support combining test-based verification with groupwise agentic grading to improve the quality and stability of code agent RL. --- ### CUE-Mem: Benchmarking Long-Term User Memory via Implicit Cues in Multimodal Conversations Authors: Yulin Hu, Yanyan Zhao, Zimo Long, Xing Fu, Mengtong Ji, Weixiang Zhao, Yutai Hou, Qianchao Wang, Dandan Tu Published: 2026-09-26 Categories: cs.AI, cs.CL arXiv: https://arxiv.org/abs/2609.32574v1 Long-term memory is essential for multimodal agents that interact with users across sustained conversations. However, user memories are not always explicitly stated: they may also be implied by recurring background objects in images, ambient sounds in audio, or other peripheral multimodal cues. Existing benchmarks largely focus on text-only memory or explicit multimodal evidence, leaving implicit multimodal cues underexplored. We introduce CUE-Mem, a text-image-audio benchmark for evaluating long-term user memory from implicit cues. CUE-Mem contains 2,674 questions across explicit and implicit evidence settings and covers four tasks: Entity Recall, Long Pattern, Personalized Recommendation, and Answer Refusal. Across textualized memory systems, implicit performance remains far below oracle evidence, locating the main bottleneck in preserving and retrieving subtle cues rather than question answerability. Increasing caption detail recovers more of this evidence, but brings uneven gains and rapidly growing token costs, motivating native multimodal access. Yet native access does not uniformly resolve the bottleneck: evidence use depends strongly on the backbone, while multimodal indexing introduces substantial retrieval noise. CUE-Mem provides a testbed for memory systems that selectively retain, retrieve, and use subtle multimodal evidence. --- ### Trapped by Their Own Rollouts: Understanding Aggregation--Rollout Feedback in Federated On-Policy Distillation Authors: Jinqian Chen, Jihua Zhu, Chang Liu Published: 2026-09-26 Categories: cs.LG, cs.DC arXiv: https://arxiv.org/abs/2609.32573v1 On-policy distillation (OPD) is a promising approach to language-model adaptation, aligning teacher supervision with the student's own generated trajectories. When adaptation prompts are distributed across clients, can this process benefit from federated collaboration? We study federated OPD and find that substantial collaboration gains can be obscured by learning-rate sensitivity: FedAvg can perform no better than independent local training at a small learning rate, yet recover a clear advantage at a larger rate. We explain this phenomenon through the student's dual role as learner and generator of future training data. An aggregation-induced optimization lag can delay access to useful teacher supervision, which in turn slows subsequent learning. Our theory establishes this aggregation--rollout feedback in a solvable model with a common optimum and stable updates, and identifies two coupled roles of learning rate: learning from current supervision and reaching future supervision. Guided by this analysis, we propose FedTOPS (Federated Teacher-guided On-Policy Scaling), which reuses teacher feedback on current trajectories to adapt the FedAvg update magnitude under clientwise predictive-change constraints. Across six mathematical reasoning benchmarks, FedTOPS improves macro Avg@8 over FedAvg by 4.56--14.57 percentage points across the evaluated student models and local learning rates. --- ### CAESAR: Clustering via Autonomous Embedding-Space Agglomerative Reorganization Authors: Ilan Bacry, Rémi Devaux, Antoine Jardin Published: 2026-09-26 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.32570v1 Clustering algorithms that operate on nearest-neighbor graphs, such as FINCH (First Integer Neighbor Clustering Hierarchy), depend heavily on the quality of the embedding space they are given. However, pretrained vision and language model embeddings are not optimized for this purpose. We propose CAESAR, a method that reorganizes a pretrained embedding space: a reorganization network is trained to pull mutual nearest neighbors together and push non-neighbors apart, yielding a reorganized embedding space substantially better suited to clustering. CAESAR offers a second major advantage: it never requires the number of clusters $K$. This matters because in realistic unsupervised settings, $K$ is typically unknown and discovering it is often part of the problem, yet most strong clustering methods take it as an input. We therefore design the entire CAESAR pipeline to infer $K$ rather than assume it is known. Empirically, reorganizing the embeddings consistently improves clustering over the raw space on both text and image datasets. Since the few deep clustering methods that also infer $K$ do not release their code, we complement controlled comparisons with methods that infer $K$ on the same embedding space by comparisons with strong deep clustering methods that are given the true $K$, giving them a substantial oracle advantage. Even so, CAESAR outperforms all of them on text, achieves the best results on the most challenging image benchmark and remains competitive on the others. Reorganizing pretrained embeddings thus emerges as a simple and powerful route to clustering realistic data, where classes overlap and the number of clusters is unknown. --- ### OmniSmartHome: A Multimodal Reasoning Benchmark for Smart-Home Agents Authors: Jihoo Jung, Suho Yoo, Jeongsoo Choi, Hyebin Cho, Tae Wook Haam, Hyeonggon Ryu, Sumin Park, Joon Son Chung Published: 2026-09-26 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.32569v1 Smart-home assistants are expected to handle diverse, realistic requests that arise in daily life. In such interactions, users often rely on the surrounding multimodal context-pointing at objects or referring to what they see or hear, leaving their requests underspecified in language alone. Existing smart-home benchmarks, however, express user requests solely through language, leaving context-dependent real-world requests underexplored. To bridge this gap, we introduce OmniSmartHome, a multimodal smart-home benchmark where each spoken request is paired with the surrounding visual and spatial-audio context, providing complementary cues to disambiguate underspecified requests. OmniSmartHome comprises 1,360 synthetic and 272 real-world episodes. We evaluate 16 omnimodal large language models (Omni-LLMs) and reveal that, while they perform strongly when speech alone sufficiently conveys the user's intent, performance drops substantially when resolving it requires reasoning over multimodal contextual cues. As a simple agent baseline, we provide PROME (PROcedural Memory for multimodal Evidence gathering), which equips agents with specialized audio-visual perception tools and procedural memory for orchestrating their use. PROME generally improves performance across six Omni-LLMs. Demos and examples are available at https://omni-smart-home.github.io --- ### Attribution Gaps in Zero-Training LLM+OVOD Pipelines: A Fine-Grained Analysis of the CAAP--SNAP Discrepancy Authors: Yu-Feng Yen Published: 2026-09-26 Categories: cs.CV, cs.CL arXiv: https://arxiv.org/abs/2609.32567v1 LAOD and similar zero-training LLM+open-vocabulary-detector (OVOD) pipelines score two things separately: class-agnostic localization accuracy (CAAP) and semantic naming accuracy (SNAP). The two consistently diverge, and nobody has asked why. This paper asks why, on the full 5,000-image COCO-Val split (27,273 detections) rather than the small subset the original work evaluated on. Object visual complexity turns out not to be the driver -- small and occluded objects are, if anything, localized better than large ones. Vocabulary novelty is: once the LLM's wording falls outside the detector's native category set, localization accuracy falls from 80.9% to 31.6%. That drop is not spread evenly across unfamiliar phrasing, though. Almost all of it comes from cases where the novel wording actually names a different object than the one COCO annotated (true synonyms still score 89.3%; semantically unrelated "noise" labels score 12.0%). A closer look at a further failure subset tells a similar story: 78-88% of what looks like complete localization failure is really the model correctly finding a real object that COCO's non-exhaustive 80-category scheme simply never labeled, not hallucination. Swap the detector backbone (YOLO-World for Grounding DINO) or the LLM (Gemma-3 for Qwen2.5-VL) and both the effect and its rough size hold up, so this looks like a general property of the pipeline family rather than a quirk of one model pairing. The upshot is that a large share of the apparent CAAP--SNAP gap traces back to closed-category annotation limits rather than a real grounding failure, which matters for how we detect hallucination, analyze failure modes, and design evaluation for grounded multimodal systems meant to work in the open world. --- ### Compositional Objectives: Learning Structure in Structure Authors: Pranavchandra Vivekananda, Sumukh Bettadapura, Ajan Subramanian Published: 2026-09-26 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.32566v1 Intelligence is defined in many ways. One of these definitions defines intelligence as the pursuit of learnable novelty. However, learnable novelty can be meaningless without the ability to compose the learned structures to take action and achieve goals. Learnable novelty builds on epiplexity, which is a way to measure learnable structure in data through a bounded observer. In this paper, we investigate a closed-form spectral approximation to compute epiplexity. We use a fixed-trace constraint and find that the epiplexity objective prefers a more uniform distribution of spectral mass rather than concentrating it in a small number of directions. However, a representation may spread information across many directions without organizing that information into features useful for a particular task. To address this gap, we propose a compositional objective whose observer measures the relationships between the parts and interactions of an image. We compare it with the original spectral objective given only the masked parts. In our ImageNet training runs, the spectral objective with masked parts produces an almost maximally spread representation while achieving the strongest frozen-feature classification performance on most evaluations, more than doubling the linear-probe accuracy of the whole-image baseline. Across multiple image benchmarks, changing what the observer sees matters more than adding relation and interaction tokens. At the same time, our prediction-oriented compositional objective produces substantially better held-out observer prediction but relatively weaker classification, revealing that spectral diversity, predictability, and downstream utility are distinct properties. These results suggest that the usefulness of spectral spreading depends not only on how much structure is preserved, but on which relationships the observer makes available to the objective. --- ### Reading Is Not Leaking: Local, Auditable Measurement and Reduction of Inference Exposure from Public Footprints Authors: Mahmudul Faisal Al Ameen Published: 2026-09-26 Categories: cs.CR, cs.CL, cs.CY arXiv: https://arxiv.org/abs/2609.32565v1 Anyone with a public footprint leaks facts that were never stated, and language models make the inference cheap. We present a framework for measuring and reducing this inference exposure that runs on the owner's own CPU with no language model at analysis time, instantiated on organisations and on individuals. It starts from a measurement result: scoring an inference system against the target's private truth conflates how well the system reads the record with how much the record leaks. On a 128-question instrument over sixteen synthetic firms, almost half of the questions are never answered correctly by any of six readers, four of them language models, and a majority-class guess accounts for most of every reader's score. We therefore separate reading accuracy from leakage rate and introduce an injection protocol that creates cells with known support. Our analyser combines rules, statistical solvers and a 106M-parameter encoder trained from scratch that marks verbatim evidence and never generates text; every answer carries a graded certificate whose recorded proof replays. Its certified answers are correct in 93% of resolved cases, against 49-73% for the language models' quote-backed answers, whose citations are produced alongside the answer rather than deriving it; with plain-prose articles in the record, 70% of its evidence-bearing answers rest on evidence that establishes them, against 18-56% for the models. A constrained defence that rewrites each fact's carrier as a true but coarser statement hides every single-carrier fact from four language-model adversaries at 40% lower edit cost than deletion. On sixteen synthetic people the guessing term is larger still, and a decoy planner with no language model halves the correct answers of the estimator it targets without transferring to a second. --- ### ProTTT: Learning to Learn Semantic User Memory with Test-Time Training Authors: Sejun Park, Hyoungjo Bhang, Hyein Jeong, Yohan Jo Published: 2026-09-26 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.32564v1 Personalization requires language models to capture user-specific knowledge from a growing user history. Existing context-based approaches incur increasing inference costs as user history accumulates and rely on separate retrieval or summarization stages, while parametric-based approaches often require reconstructing user representations when new user data is added. We introduce ProTTT, a profile-supervised meta-learning framework for learning semantic user memory. The memory construction starts from a shared initialization and is updated for each user through test-time training on user history, allowing it to evolve continuously as the history grows. However, since test-time training alone does not explicitly encourage the memory to capture semantic user knowledge necessary for personalization, we learn this shared initialization using textual user profiles as supervision, so that test-time training on user history captures semantic knowledge more effectively. ProTTT consistently outperforms both full history ICL and all parametric baselines across diverse benchmarks, while substantially reducing inference cost by compressing user history into a lightweight parameterized memory. Our analysis also shows that profile supervision is a reliable objective for learning semantic user knowledge and that the resulting memory can track and retain evolving user preferences, while remaining robust across different history sizes. Overall, we demonstrate the effectiveness of test-time training for personalization and establish ProTTT as a baseline for continuously evolving user memory. --- ### Artificial intelligences and human scientists exhibit complementary strengths in theory building Authors: Ke Li, Spyros I. Zoumpoulis, Phanish Puranam, Philip Parker, Matthew Eshbaugh-Soha, Izzy Gainsburg, Michael Gilead, Igor Grossmann, Britt Hadar, Yoel Inbar, Almog Simchon, Robb Willer et al. Published: 2026-09-26 Categories: cs.AI, cs.CY, cs.HC arXiv: https://arxiv.org/abs/2609.32562v1 We investigate the effectiveness of artificial intelligences (AI)-specifically large language models (LLMs)-relative to human scientists at high-level cognitive tasks in social science such as theory formulation, predictions of novel empirical results, and theory revision in response to new evidence. The research domain was academic discourse regarding gender and race inequality. Our findings, comparing 25 LLMs with 13 senior researchers and 60 doctoral scholars, reveal that the AIs outperformed most humans individually on most of the present tasks, while human theories were more diverse and exhibited greater gains in predictive accuracy from aggregation. AI-generated theories were more extensively elaborated, involving additional theoretical paths and latent variables, and were rated as higher quality than human theories by independent raters blinded to source. However, this theoretical complexity was in part ornamental, in that it was not associated with more accurate predictions about empirical patterns in data; in contrast, human scientists achieved greater predictive efficiency with simpler theories. The AIs were significantly more likely than human scientists to revise their theories to incorporate new evidence; human scientists updated their beliefs in a selective way that is sensitive to prior prediction errors. We speculate that the superior processing capacity of artificial intelligences makes them especially well-suited to tasks requiring grappling with complexity, but that the greater diversity of human ideas is essential to wise crowds and collective creativity. --- ### How to Reduce Whisper Hallucination Authors: Husein Zolkepli Published: 2026-09-26 Categories: cs.CL arXiv: https://arxiv.org/abs/2609.32560v1 Whisper is still what runs in production: one permissively licensed checkpoint, 99 languages, no per-language tuning. But it writes sentences nobody said. On 42 clips of pure room tone, whisper-large-v3 emits words on 61.9% of them and emits something on 100%. The usual response is to distil a student from Whisper pseudo-labels, filtering the hallucinations out of the corpus first, which deletes the evidence while leaving the behaviour in place: the student is fitted to the subset where the teacher was right, and inherits a failure mode absent from its own training data. The second response, adding non-speech audio so the model learns to stay quiet, already ships and works, but teaches suppression without discrimination. The checkpoint best on every non-speech arm here is also the one that recovers the fewest genuinely spoken phrases and deletes 58% of repeated speech. The teacher has to be fixed, with both halves of the signal. Our benchmark scores both at once: 11,852 clips over eight arms, 6,267 synthetic positives, 8,296 clips of real audio that made a production model fail, and FLEURS in 58 languages. We collect 40,891 hallucination phrases in 100 languages, choose a text-to-speech system by measurement, and synthesise those phrases as positives, so the model meets the same text both as something to suppress and as something to transcribe. Across 33 matched pairs of fine-tunes differing only by those positives, adding them lowers word emission on real voice-free audio in 31 pairs and raises phrase recovery in 32; among the 30 pairs that hold English accuracy it is 30 out of 30. The best checkpoint takes hallucination on silence from 61.9% to 2.4% and words over real voice-free audio from 99.9% to 47.8%, while raising phrase recovery from 69.8% to 82.7%. The benchmark, lexicon and synthetic corpus are released at https://huggingface.co/datasets/Scicom-intl/Whisper-Hallucination --- ### CFCH: Coarse-Fine Collaborative Hierarchical Learning for Anterior Segment Disease Analysis Authors: Peng Wang, Haohan Zou, Yanlin Wu, Xueshuo Xie, Yan Wang, Tao Li Published: 2026-09-26 Categories: cs.CV arXiv: https://arxiv.org/abs/2609.32559v1 Accurate classification of anterior segment diseases is crucial for ophthalmic screening and diagnosis. However, slit-lamp image analysis remains challenging due to substantial variability in imaging conditions and the intrinsic anatomical-disease hierarchy of ocular pathologies. Existing methods typically formulate this task as a flat multi-class classification problem, ignoring the structured dependency between anatomical regions (e.g., cornea, conjunctiva, and lens) and disease manifestations.To address these limitations, we propose CFCH, a Coarse-Fine Collaborative Hierarchical learning framework that explicitly models anatomical context and disease semantics through a dual-branch architecture. To enable effective cross-granularity collaboration, CFCH introduces semantic and cross-granularity attention consistency constraints, encouraging aligned yet complementary feature learning across branches. In addition, we construct AS-9K, a large-scale anterior segment dataset with 8975 images covering 12 common disease categories. To the best of our knowledge, AS-9K is the largest publicly available dataset for anterior segment image classification. Extensive experiments on two anterior segment datasets demonstrate that CFCH outperforms state-of-the-art methods. Qualitative visualizations further show more focused and lesion-relevant activation responses, validating the effectiveness of the proposed framework. Code will be available at https://github.com/ybupengwang/CFCH. --- ### Language as an Independent Information Layer: A Conceptual Model of Communication, Cognition and Decision-Making Authors: Anastasiia Alifanova, Elena Benderskaya Published: 2026-09-26 Categories: cs.CL, cs.CE, cs.CY, cs.LG arXiv: https://arxiv.org/abs/2609.32556v1 Based on an analysis of the role of language in thought and communication, this article proposes a new concept for designing corporate knowledge bases. The concept integrates the probabilistic vector space of a corporate vocabulary, reflect-ing industry specifics, subject focus, terminology, and culture, with traditional ontological modeling. This combination enables the efficient extraction of knowledge from accumulated corporate documents while strictly accounting for specific business processes. Consequently, this concept bridges statistical and semantic (cause-and-effect) methodologies. Furthermore, analyzing the projec-tions of probabilistic spaces and causal relationships can help identify bottlenecks in business logic. As a dynamic system, language functions as a separate, inde-pendent layer within the overall information architecture. Introducing a dynamic component into the probabilistic space of word distribution allows it to be mod-eled as a multidimensional solution space for various problem formulations. In this context, input data defining the problem conditions serve as control parame-ters for dynamic transformations. --- ### Feature Space Guidance for Breast Cancer Classification in DCE-MRI Authors: Benjamin Hamm, Yannick Kirchhoff, Maximilian Rokuss, Moritz Langenberg, Constantin Ulrich, Tassilo Wald, Jeremias Traub, Karol Gotkowski, Klaus Maier-Hein Published: 2026-09-26 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2609.32555v1 Dynamic contrast enhanced breast MRI (DCE-MRI) is a powerful clinical tool for breast cancer detection, providing high resolution anatomical detail together with rich temporal contrast information. However, high dimensional 4D inputs, small lesions, and heterogeneous acquisition protocols across clinical sites hinder robust automated classification of healthy, benign, and malignant cases. To address these challenges, we propose a framework that dynamically analyzes latent representations to adapt to protocol-specific characteristics. Spatial variability is mitigated by reducing confounding background uptake and compensating for misalignment caused by deformable soft tissue. Additionally, relationships in the latent space across phases are leveraged to select the most informative temporal features, improving robustness to protocol-specific temporal variability. Finally, task specific discriminative features are promoted through large scale supervised lesion segmentation pretraining, which substantially enhances downstream finetuning. Evaluated under leave-one-center-out validation on the ODELIA dataset and the held-out AMBL cohort, the proposed framework substantially outperforms finetuned radiology foundation models and prior methods, improving mean AUROC by nearly 8 points and balanced accuracy by 4 points over the strongest baseline. Additionally, our method achieved first place in the MICCAI ODELIA Breast MRI Challenge 2025, further demonstrating its effectiveness for robust breast cancer classification. We publicly release our codebase under https://github.com/MIC-DKFZ/CURIAtor. --- ### Harnessing Coupled Stream Completion For Human-Object Interaction Modeling Authors: Dawei Guan, Di Yang, Jiangtao Wang Published: 2026-09-26 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2609.32551v2 Text-conditioned human-object interaction (HOI) generation requires body motion, object trajectories & rotations, and hand articulation to remain coordinated. These components differ in scale and dynamics, but must agree on contact, relative pose, and timing. A shared representation may limit the distinct structure of each stream, while independent generation prevents each stream from responding to changes in the others. Latent supervision alone also does not directly constrain contact after decoding. We propose TRACE, a continuous latent framework that keeps stream states separate and couples their updates. TRACE encodes body, object, and hand motion into separate latents and predicts each stream velocity from the complete current interaction state. Geometric losses on decoded motion further constrain contact and object-relative motion over time. The same model supports completion of any single absent stream from the other two. Frozen flow features also serve as input to a language model for HOI understanding. Experiments on InterAct, OMOMO, and BEHAVE show that joint completion training improves generation and that frozen flow features improve understanding over raw-motion encoding. On InterAct, TRACE achieves the highest contact precision, recall, and F1 among the compared methods. --- ### Are Vision-Language-Action Models Robust to One-Step Observation Perturbations? Authors: Shojiro Yamabe, Jun Sakuma Published: 2026-09-26 Categories: cs.AI, cs.LG, cs.RO arXiv: https://arxiv.org/abs/2609.32550v1 Understanding the safety risks of vision-language-action (VLA) models is essential for their deployment in the physical world. Existing safety research has mainly considered persistent perturbations that are applied continuously to observations throughout an episode. However, momentary observation corruption, in which observations are severely perturbed only briefly within an episode, remains an underexplored safety threat. To address this gap, this work investigates robustness to one-step perturbations applied at a single time step per episode. Our experiments reveal that these perturbations substantially degrade VLA performance and that their impact depends on the action chunk execution length. Based on them, we propose CARE, which dynamically selects the execution length based on consistency with the previously predicted action chunk. CARE improves robustness with low computational overhead while preserving clean performance by selecting shorter execution lengths only under perturbations. --- ### Prioritizing Repeated LLM Evaluation for Hidden Failure Discovery Authors: Keita Broadwater, Akin Broadwater Published: 2026-09-26 Categories: cs.LG arXiv: https://arxiv.org/abs/2609.32547v1 Large language models are commonly evaluated by generating a small number of stochastic responses for each prompt in a benchmark. Because inference budgets are limited, this shallow evaluation may fail to observe low-probability but operationally important failures. A prompt that produces no failures in a small sample may therefore appear reliable despite having a nonzero latent probability of failure under repeated inference. We formulate LLM reliability evaluation as a budget-constrained discovery problem in which each prompt is associated with an unknown per-generation failure probability. We propose a budgeted discovery framework that first performs shallow evaluation across the prompt set and then uses trial-level failure outcomes together with prompt-derived representations to learn a feature-based ranking of failure propensity. The resulting scores prioritize prompts with zero observed shallow failures for deeper evaluation, concentrating the deep-evaluation budget where hidden failures are more likely to be discovered. We evaluate this approach on AIRBench and StrongREJECT across multiple model and system-prompt conditions. The central empirical test asks whether models fit without access to deep-evaluation outcomes can rank prompts with zero observed shallow failures according to their likelihood of producing failures under deeper evaluation. On AIRBench, the highest-ranked 10\% of unresolved prompts achieves 2.54x hidden-failure lift for Qwen 2.5 7B and 1.87x for Gemma 3n E4B, recovering 25.4\% and 18.7\% of subsequently observed hidden failures, respectively, compared with 10\% expected under random allocation. Semantic-neighborhood and feature-ablation analyses further show that this predictive signal can be recovered from multiple representations of prompt content and relationships. --- ### Shared Autoregressive Context Can Distort Relationships in Synthetic Data Authors: Thomas S. Robinson Published: 2026-09-26 Categories: cs.LG, cs.CL arXiv: https://arxiv.org/abs/2609.32546v1 Large language models can generate several records within one autoregressive completion, making earlier answers available as context for later records. This paper shows that such shared-completion batching can distort relationships among variables in the resulting synthetic data, using controlled tests on synthetic survey respondents. In a matched experiment on 2,000 European Social Survey profiles, generating ten rather than one respondent per request increases mean absolute error in within-country correlations by 48-58% for Qwen3.8-27B and 114-127% for Llama-3.3-70B-Instruct across three seeds, holding profiles, examples, questions and decoding parameters fixed. The distortion primarily reflects exaggerated relationship strength, while retaining substantial agreement with the human ordering of correlations. Controlled interventions establish answer history as a causal channel: re-pairing the same preceding values, with profiles and marginal distributions fixed, changes correlations among subsequently generated responses. Hiding preceding answers reduces correlation error in the tested settings but worsens marginal accuracy. Exploratory corrections across social-attitude, health and economic data likewise show that lower correlation error can coexist with worse marginal distributions and regression estimates. Request construction is therefore part of the data-generating process, and synthetic-data validity must be evaluated against the analyses the generated data are intended to support. --- ### Porimon: An LLM-Based Pokémon Battle Agent Enhanced by Long/Short-Term Knowledge Augmented Generation Authors: Dongyin Zhuo, Fengjunjie Pan, Nenad Petrovic, Alois Knoll Published: 2026-09-26 Categories: cs.AI arXiv: https://arxiv.org/abs/2609.32544v1 In this paper, we use Pokémon Battles as a case study to investigate how to improve the performance of LLM-based agents in tasks that require opponent-aware planning without additional fine-tuning. We propose Long/Short-Term Knowledge Augmented Generation (LSTKAG), a mechanism that enables LLM-based agents to leverage past states of the current task and retrieve experience summaries from similar previous task instances based on the current state. Based on LSTKAG, we design Porimon, an LLM-based agent structure for Pokémon Battles. For optimization, we introduce an external API for precise damage calculation and more detailed information about the game. We conduct tournament-like evaluation experiments comprising 15,000 battles for hyperparameter optimization, ablation studies, and performance evaluation. The results indicate that Porimon-based players with hyperparameter optimization significantly outperform players based on PokéLLMon, an LLM-based agent structure proposed in previous research, and the rule-based heuristic player. Furthermore, our ablation study shows that Porimon variants outperform the one without extension in game information retrieval, which shows the contribution of that extension. However, the current experiment results are inconclusive regarding the contribution of Long-Term KAG. These results suggest that introducing external resources, information from previous states of the current task, and experience summaries from similar previous task instances could elevate the performance of LLM-based agents designed for tasks requiring opponent-aware planning. --- ### In-Flight KV Cache with Clean Anchors for Faster Autoregressive Video Diffusion Authors: Yikai Wang, Xiao Han, Mengmeng Xu, Juan Camilo Perez, Yiannis Douratsos, Sen He, Zijian Zhou, Fei Zhang, Zhaochong An, Juan-Manuel Perez-Rua, Chen Change Loy, Tao Xiang Published: 2026-09-26 Categories: cs.CV, cs.AI, cs.LG, cs.MM arXiv: https://arxiv.org/abs/2609.32540v1 Few-step autoregressive video diffusion generates a long video by splitting the video into temporal chunks and generating chunk-by-chunk, each through a short sequence of denoising stages. To memorize chunks that are already generated, previous methods reconstruct a clean or less-noisy key--value (KV) cache by additional forwards to build the cache without advancing an output latent. However, every denoising forward itself already computes the in-flight KV of the current chunk. We introduce FlashForward, which directly reuses this cache to avoid the heavy cache-update-only model forwards. After the current chunk completes one denoising stage, its stage-specific cache is already available for the next chunk. Assigning one GPU to each stage therefore lets different chunks occupy different stages concurrently. This early availability has a quality cost: the resulting stage-matched history is noisy, causing appearance and motion drift among chunks. To complement it, FlashForward produces sparse auxiliary clean anchor latents before the corresponding region is generated so the generation trajectories can be stabilized by this two-sided conditioning. The two memories operate at different temporal scales: sparse clean anchor KV supplies coarse, long-range two-sided structural guidance, while dense stage-matched history preserves fine, recent evolution. With up to four GPUs, FlashForward runs $1.16$--$1.69\times$ faster than HiAR and $1.42$--$2.92\times$ faster than Self-Forcing for 16 FPS videos of 20 seconds or longer across 1.3B and 14B backbone scales at 480p and 720p. On VBench, for the 1.3B model at 480p, it achieves higher scores and remains stable at longer durations, demonstrating that FlashForward generates high-quality and temporally consistent videos across durations of 20s, 35s and 65s at a much faster generation speed. --- ### Interpretable Physics Informed WiFi Indoor Localization: Learning an Effective Access Point Geometry and Using It to Prune Authors: Arshia Eftekhari zadeh, Rezvan Nasiri, Hadi Moradi Published: 2026-09-26 Categories: cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.32539v1 Deep learning models can achieve high accuracy for indoor localization, but their black-box nature limits interpretability and the reuse of learned information. We propose a hierarchical deep learning framework for WiFi fingerprint-based indoor localization that jointly predicts user location and learns an effective geometry of the surrounding access points (APs). Physics-informed decoders infer this geometry directly from RSSI measurements and labelled user positions, without requiring the true AP coordinates during training. The learned geometry is then used to rank and prune APs. On the UJIIndoorLoc dataset, the proposed chained model achieves a mean 3D localization error of 7.07 m, reducing error by 26% to 36% compared with baseline models. Previously published methods evaluated on the same official split report errors 10.6% to 31.0% higher. Pruning 35% or 50% of the APs causes only a small loss in localization accuracy. The inferred geometry also enables Fisher-information-based AP ranking even when fingerprint databases do not contain surveyed AP coordinates. Experiments on the Tampere/TUT and UTSIndoorLoc datasets show that geometry-guided AP selection performs comparably to selectors built directly from labelled data. These results show that physics-informed interpretability can improve indoor localization while also supporting effective feature selection. --- ### RIPE-MambaSpike: Resolution-Independent Spiking-State-Space Interfaces for Parameter-Efficient Event-Based Vision Authors: Md Muhiminul Islam, Shoaib Ahmed Dipu, Sayeed Shafayet Chowdhury Published: 2026-09-26 Categories: cs.CV, cs.NE arXiv: https://arxiv.org/abs/2609.32537v1 Spiking-Mamba hybrids reach strong accuracy on event-based vision, but existing designs often require tens of millions of parameters. Much of that cost comes from how the spiking front-end is connected to the state-space backbone rather than from the hybrid architecture itself. In a representative model, a single resolution-dependent projection accounts for 33.55M of 36.25M parameters. To that end, we introduce RIPE-MambaSpike (Resolution-Independent, Parameter-Efficient), which replaces that projection with a hierarchical multi-resolution bridge of fixed channel width. Its deployed footprint is 0.870M parameters, constant at fixed time steps and widths across a 43x range of input areas. Reparameterized spiking stages, temporal decoupled modulation, and a dynamic convex-hull-bounded dual-stream membrane-potential attention preserve accuracy under this compact design. Result-wise, RIPE-MambaSpike is pareto-optimal on CIFAR10-DVS, N-Caltech101, and DailyDVS-200. Notably, on the 200-class DailyDVS-200, a scaled 8.04M configuration achieves 45.7% top-1 accuracy, the best reported spiking result on that benchmark, and outperforms prior spiking methods with 3.0-15.1x fewer parameters than dense ANNs. Overall, our findings demonstrate that competitive event-based recognition does not require resolution-dependent parameter growth. Code is available at https://github.com/MuhiminOsim/RIPE-MambaSpike. --- ### Do Audio LLMs Listen Before They Act? Diagnosing Acoustic-Context Gating in Voice Agents Authors: Yanjie Zhang, Nanchen Hu, Yushi Sun Published: 2026-09-26 Categories: cs.SD, cs.AI, cs.CL, cs.MM, eess.AS arXiv: https://arxiv.org/abs/2609.32536v1 Audio language models can recognize spoken commands and invoke tools, but an agent must first decide whether the acoustic and conversational context warrants action. We introduce VGBench, a 1,018-item diagnostic benchmark for action-level addressedness across side-talk, self-talk, and speaker-switch scenarios. Each item uses a shared action space comprising silence, a tool call, and a natural-language answer. Speaker-switch pairs hold the specified words fixed while source, distance rendering, and a temporal boundary define a controlled wearer-to-bystander shift. Six raw Audio LLMs and three training-free adaptations often identify the target tool yet rarely withhold action under this shift; the highest raw switch mute rate is 14%. We then use VoxGate as a post-training case study. Supervised training mutes 91.3% of switched commands while choosing the correct tool for all nearby wearer commands and text-only controls. An exploratory GRPO stage has similar switch performance; side-talk accuracy rises from 68.4% to 70.9%, and self-talk muting from 52.0% to 60.0%. Factorized controls identify an independent source-change effect, while sensitivity to the far-field manipulation varies across acoustic renderings. The benchmark therefore measures multi-cue acoustic-context gating rather than isolated speaker identity. --- ### DepthBench: Measuring How Residual Connections Enable More Computational Depth Authors: Keyu Wang, Yangyi Huang, Jiale Kang, David González-Martínez, Weiyang Liu, Shiwei Liu Published: 2026-09-26 Categories: cs.CV, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2609.32534v1 Depth is a natural way to increase the computational capacity in Transformers, yet the contribution of deeper layers can diminish as depth grows larger. Recent approaches enhance normalization (\text{e.g.}, LayerNorm Scaling) or residual connections (\text{e.g.}, mHC, AttnRes) to enable better information flow and depth utilization. However, it remains unclear whether they truly translate increased architectural depth into effective computational depth, and whether their reported gains stem from better access to information across depth, or unaccounted-for confounding factors. In this paper, we introduce \textbf{DepthBench}, a controlled benchmark for studying computational depth across various architectures. We systematically vary the width--depth aspect ratio ($d_{\text{model}}/n_{\text{layer}}$) from shallow--wide to deep--narrow shapes, while keeping the model size and pre-training recipe fixed. Across 10 representative architectures, we find that the benefit of allocating more capacity to depth is strongly architecture-dependent. Standard Pre-LN and most of its norm- and scaling-based variants provide little benefit and can even degrade performance as models become deeper and narrower, whereas HC and Full AttnRes improve consistently even at extreme deep shapes. These gains extend beyond pre-training loss and consistently translate into improved domain-specific performance and effective computation. Controlled layer-level analyses further show that the gains of HC and Full AttnRes are associated with more effective utilization of additional layers, revealing distinct mechanisms of computational depth across architectures. Overall, our results identify residual connection design as a key determinant of whether depth can serve as a meaningful scaling axis by enabling additional architectural depth to translate into effective computation. ---