# Research Papers — 2026-10 · bundle page 64 of 156

> Full-text markdown of papers #3151–3200 of 7799 published in 2026-10, newest first. AI-readable mirrors of the full 7799-paper 2026-10 archive (50 papers per file).

Canonical page: https://aipod.dev/papers/archive/2026-10/page/64
[← Previous bundle](https://aipod.dev/papers/archive/2026-10/page/63.md)
[Next bundle →](https://aipod.dev/papers/archive/2026-10/page/65.md)

### From Transfer Probes to Expert Grouping: Evidence from LoRA Post-Training

Authors: Xueyang Wu
Published: 2026-10-10
Categories: cs.LG
arXiv: https://arxiv.org/abs/2610.14437v1

A platform that post-trains one base model for many business lines often has more lines than experts it can maintain, so lines must share experts, yet lines seldom come with a taxonomy saying which should share. We ask whether transfer probes (train on one source, measure paired changes on all sources) can discover a grouping when the standard one is unknown. To test probes against a reference, we use seven public sources that do have a label tree (five mathematics subfields and two kinds of programming, standing in for business lines) and treat the tree's grouping as a label reference that the evaluator holds and the method does not. With four LoRA experts under fixed per-source exposure and per-expert budgets, we compare a purely probe-derived grouping and an uninformed uniform default (whose tied routing serves every line from one expert) against this reference. Trained once for one interval from a shared warm-up state and routed by a rule fixed by each allocation, neither reached the prespecified practical gain of 2 percentage points, and both had lower overall utility than the reference ($-2.43$ and $-2.54$ pp; 95\% intervals below zero). In a post hoc comparison, the probe-derived grouping was indistinguishable from the uniform default ($+0.11$ pp, $(-1.25,+1.54)$). In fixed-budget LoRA post-training, a grouping built from initial-state single-source probes thus did not deliver the advantage predicted from the probes once carried to a later training state, a finite budget, and specified routing.

---

### Design Choices for Multimodal Transformer Learning with Missing and Longitudinal Healthcare Data

Authors: Rita Cordeiro Mendes, Maria Manuel Pereira, Carlos Santiago, Catarina Barata
Published: 2026-10-10
Categories: cs.LG, cs.CV
arXiv: https://arxiv.org/abs/2610.14436v1

Multimodal learning offers the potential to improve predictions by integrating complementary information from heterogeneous data sources. However, multimodal datasets are frequently incomplete, raising important questions about how models should handle missing information. Although recent approaches increasingly rely on Transformer-based architectures for multimodal fusion, incomplete multimodal data introduces challenges such as missing modalities and intra-modal missingness, which motivate key architectural design choices, such as tokenization granularity. In addition, many datasets are longitudinal, with observations collected across multiple time points. The sequential nature of the Transformer architectures provides a natural framework for modeling such data, but introduces additional considerations, such as how temporal information should be represented. In this study, we systematically investigate these design choices using three different cancer datasets. We compare modality-level and feature-level tokenization and evaluate attention masking and imputation as strategies for handling missing information, both naturally occurring in the datasets and synthetically induced. We further investigate approaches for incorporating temporal information in our training process and how to model it. Experiments are conducted with unimodal and multimodal configurations, as well as different levels of missingness for binary survival classification and multiple time encoding strategies for discrete-time survival prediction. Our results show that while multimodal outperforms unimodal results, missing data heavily impacts predictions and thus effective design choices such as tokenization granularity, and intra-missing handling is essential for reliable prediction.

---

### MemForm: Benchmarking Memory Formation Poisoning in LLM Agents

Authors: Haowen Pan, Pengyang Shao, Yunshan Ma, Yanzheng Jin, Naixin Zhai, Xiang Wang, Xun Yang
Published: 2026-10-10
Categories: cs.CR, cs.CL
arXiv: https://arxiv.org/abs/2610.14431v1

Long-term memory enables LLM agents to learn from past interactions but exposes memory formation to adversarial influence. Existing studies primarily organize memory poisoning by attack channels, corruption patterns, or lifecycle stages, leaving formation interfaces insufficiently characterized. We formulate Memory Formation Poisoning as adversarial manipulation of the interfaces through which persistent agent memory is produced and committed, and introduce MemForm, a controlled diagnostic benchmark covering five interfaces: Content, Evidence, Scope, Selection, and Path Poisoning. We introduce Selection and Path Poisoning as controlled intervention categories that isolate retention and consolidation order effects, respectively. Each category pairs clean and poisoned conditions with interface-specific controls. Evaluations across six LLMs and two response formats reveal broad susceptibility at all five interfaces under the evaluated memory protocols and implementations. Crucially, authentic experiences with successful outcomes do not guarantee reliable behavior after memory formation: changes to contextual support, retention, or consolidation order can induce undesirable downstream behavior. In particular, consolidation order can alter behavior even with identical retained experiences, depending on the memory formation procedure. Additional analyses show that interface applicability depends on memory architecture and that Path Poisoning susceptibility depends on the updater. The evaluated defenses provide incomplete mitigation that varies by interface. These findings motivate protecting both what is committed to persistent memory and how memory is formed from experiences. Code and data are available at https://github.com/opanhw/memform-code.

---

### Frustratingly Easy Agentic RAG via Portable Skills: A Case Study in Customer Support

Authors: Istok Bako, Bonan Min, Tiffany Shao, Claire Cheng, Yashar Mehdad
Published: 2026-10-10
Categories: cs.AI
arXiv: https://arxiv.org/abs/2610.14428v1

Enterprise workflow automation typically requires retrieving a complete and accurate set of relevant information from large operational document corpora (SOPs, policy documents) given a specific scenario or request. In customer support (CS), for example, human specialists must determine the correct sequence of resolution steps for each case, a task that demands precise grounding in policy documentation. Meeting this bar in production typically requires complex document chunking, indexing, and retrieval pipelines, often layered with learned re-rankers, query rewriters, and bespoke agentic RAG infrastructure. We report a deployment case study of a simpler alternative: a portable skill consisting of a natural-language instruction, lightweight tools (a retrieval API, format-checking scripts), and a corpus enriched with an LLM-generated navigation layer, which is a structured "Wikipedia-style" index over the underlying documents. This agentic RAG approach is straightforward to implement (1-2 engineer days), portable across capable agentic harnesses (e.g., Claude Code), and performs strongly on the challenging task of CS resolution workflow generation, achieving 99.2% procedural equivalence to human-approved ground truth and 83.3% evidence sufficiency. It is also roughly an order of magnitude slower per case than our deployed production assistant (~90 s vs. ~8-9 s), so we claim no latency advantage. Our claim is scoped to that setting: a mature, well-curated corpus, a latency-tolerant workflow, and human review of every case. There, the skill did the work of a complex RAG stack, and maintenance became a matter of editing text and data rather than retraining models.

---

### ByteMark: A Unified Byte-Level Watermark for Detection Across Language Models

Authors: Xiaopu Wang, Runze Li
Published: 2026-10-10
Categories: cs.CR, cs.LG, stat.ML
arXiv: https://arxiv.org/abs/2610.14427v1

As watermarking of LLM-generated text enters deployment, users increasingly need to check text whose source is unknown. Existing detectors read the watermark through a specific tokenizer or embedding pipeline, and even if all providers adopted the same watermarking method, differences in these representations would leave their watermarks incompatible, and the text alone does not reveal which representation produced it. We propose ByteMark, a unified watermarking scheme that defines its signal on the UTF-8 bytes of published text. Models with different tokenizers embed this signal under one public specification and key, and a single detector verifies the text from the text and key alone. When the tokenizer or embedding model of a baseline detector is replaced, its detection degrades sharply, whereas ByteMark, which uses neither, retains its detection rate; ByteMark also remains detectable when participating models continue or rewrite the text.

---

### Scaling Laws for Multimodal Data Mixtures

Authors: Aditi Khandelwal, Ayush K. Tarun, Yixuan Xu, Imanol Schlag, Golnoosh Farnadi, Siva Reddy, Luke Zettlemoyer, Antoine Bosselut
Published: 2026-10-10
Categories: cs.CL, cs.LG
arXiv: https://arxiv.org/abs/2610.14423v1

Frontier AI systems are increasingly natively multimodal, jointly pretrained on multiple modalities, such as text, vision, and audio. However, the scientific understanding of the multimodal data mixtures used to pretrain these systems --- how much of each modality to mix, and at what potential interference cost to the others --- is largely absent or gatekept. In this work, we derive data mixing scaling laws for three modalities: text, vision, and audio. Multimodal Large Language Models (MLLMs) increasingly adopt Mixture-of-Experts (MoE) architectures, as MoEs scale efficiently and partially limit cross-modal interference. We therefore use MoEs to conduct 268 experiments ranging from 457M to 8.3B parameters, on up to 150 billion tokens. Our work leads to multiple actionable insights into multimodal data mixing for training MLLMs. First, compute optima are modality-specific: audio modality is model-heavy, favoring scaling parameters over tokens, while text and vision require balanced scaling of model size and data. Second, repeating vision or audio data beyond 2x yields negligible benefits. Finally, cross-modal interactions are highly asymmetric: vision strongly interferes with audio, whereas audio's interference with learning text and vision is negligible. Overall, we provide a principled foundation for understanding the multimodal data mixtures needed to train frontier MLLMs.

---

### Prior-Informed Generalizable Low-Rank Representation Learning via Regularized Adversarial Reweighting

Authors: Keyao Zhan, Junwei Lu, Tianxi Cai
Published: 2026-10-10
Categories: stat.ME, math.ST, stat.ML
arXiv: https://arxiv.org/abs/2610.14414v1

Modern learning problems increasingly leverage data from heterogeneous sources to improve robustness and generalizability under distribution shift, often with high-dimensional parameters admitting low-rank structure. Existing group distributionally robust optimization (group-DRO) approaches optimize against adverse source mixtures, but extending this principle to low-rank matrix estimation with a broad class of losses leads to a challenging nonconvex-concave problem for which methods and theory remain limited. We propose Prior-Informed generalizable Low-rank representation Learning via Adversarial Reweighting (PILLAR), which smooths the group-DRO objective by regularizing adversarial source weights toward a user-specified reference distribution while learning a shared low-rank representation. By tempering extreme adversarial reweighting, PILLAR improves estimation stability and reduces the potential conservativeness of group-DRO. Using Burer-Monteiro factorization, we optimize the nonconvex-concave objective by bifactored gradient descent with mirror ascent. We establish linear convergence to a statistical neighborhood of the population saddle point and derive an error bound that separates low-rank estimation from adversarial reweighting. We further show that PILLAR improves on vanilla group DRO when the latter's excess conservativeness dominates source heterogeneity and prior misspecification. Simulations and an electronic health record study of suicide-attempt risk demonstrate improved worst-source and transported predictive accuracy.

---

### Coarse-Graining Recursive Expert Computation: Continuation-Relative Interfaces and Failure Modes

Authors: Shihua Yu, Yanzhi Wang, Tianchi Chen
Published: 2026-10-10
Categories: cs.LG, cs.AI, cs.CC, cs.CE
arXiv: https://arxiv.org/abs/2610.14413v1

Computational coarse-graining aims to replace a sequence of updates with a cheaper module while preserving task performance. We present a continuation-relative evaluation framework that assesses replacements through the computation that follows them, using downstream task quality and total execution cost as joint criteria. The framework distinguishes boundary representations, summaries of skipped computation, executed operations, and downstream loss. It makes interfaces and permitted deployment information explicit, combines fine-path references with directly trained controls using the same shortened architecture, and accounts for selection overhead. An exact loss decomposition separates replacement effects at a fixed downstream action from effects of changing that action. A finite construction shows how retained information can reduce the stipulated cost of zero-error prediction under fixed information and controller restrictions. Experiments in two frozen pretrained language models show that representation fidelity and task loss can favor different fitting objectives. In a prespecified comparison on a 1.7-billion-parameter model, next-token-loss training lowers in-distribution negative log-likelihood by 0.1901 nats per scored token compared with reconstructing the skipped blocks' output, using the same optimizer and update count. This ranking reverses in all nine paired comparisons under distribution shift. In a recursive expert prototype, the jointly adapted candidate lowers task error relative to frozen-bank and router-only baselines, but trails directly trained controls, and the tested input-only selectors fail the joint quality-cost criteria. Together, the protocol and studies provide explicit criteria and diagnostics for distinguishing useful computational replacement from local approximation gains within specified interfaces and evaluation settings.

---

### PotentialEP: Backprop-Free Training of CIFAR-Scale Generative Models via Equilibrium Propagation

Authors: Alex Gower
Published: 2026-10-10
Categories: cs.LG, cs.ET, cs.NE
arXiv: https://arxiv.org/abs/2610.14409v2

Generating an image with a diffusion or flow-matching model typically involves integrating a velocity field for hundreds of sequential steps. That computation suits a GPU poorly and an analog neuromorphic substrate well, where one fast, low-power physical relaxation could replace the whole sequence. Two things are needed: a generative algorithm whose computation is a relaxation, and a way to train it without backpropagation, whose global gradient transport is hard to realise in analog. Energy Matching supplies the first, sampling by gradient-descent relaxation on a time-independent potential - but obtaining the gradient that descent follows is itself a backward pass, once per integration step. We prove that a predictive coding network with a single linear injection term added to its energy relaxes to a state whose visible layer carries that gradient inherently, with an error that vanishes for small injection strengths. An Energy Matching model trained by backpropagation can therefore be loaded into this network unchanged and sampled by relaxation alone, matching the original model's FID to within 0.04 at CIFAR-10 and ImageNet-32 scale. The network can also be trained without backpropagation: Equilibrium Propagation reproduces backpropagation's parameter updates to a cosine similarity of 0.9998, and training end to end with those updates reaches an FID of 3.64 on unconditional CIFAR-10 against 3.54 for backpropagation. To our knowledge this is the first CIFAR-10 generative model trained with Equilibrium Propagation, and the first at competitive quality by any backpropagation-free rule. This quality does not require group normalisation or attention: an architecture of only convolutions and pointwise nonlinearities, both native to an analog crossbar, reaches an FID of 3.51 when trained by backpropagation.

---

### Neural Block-wise Auto-exposure for Single-Shot High Dynamic Range Imaging

Authors: Zhengguo Li, Chaobing Zheng, Wei Wang
Published: 2026-10-09
Categories: cs.CV
arXiv: https://arxiv.org/abs/2610.18733v1

Block-wise varying exposures are effective for cap- turing appearance of a real-world high dynamic range (HDR) scene via a single-shot. However, brightness order reversal (BOR) artifacts, blocking artifacts and loss of details in highlight/shadow regions are possible issues for block-wise varying exposures. In this paper, a novel neural block-wise auto-exposure algorithm is proposed to address the issues. The proposed algorithm maps a temporary image and its exposure time to the exposure times of all non-overlapping 16 X 16 blocks. The temporary image can be either an sRGB or a raw image. Although the sRGB image is widely adopted in existing auto-exposure algorithms, exposure times of blocks in the highlight and shadow regions might not be accurately learned due to the limited radiance measurement capability of the sRGB image. This paper thus also explores the feasibility of temporary image as a raw image. Due to the better radiance measurement capability of the raw image, the exposure times of blocks are learned more accurately, and more details in the highlight/shadow regions of the real-world HDR scene are captured by the resultant HDR imaging system. Experiment results validate the proposed neural block-wise auto-exposure algorithm.

---

### TheoPhysEval: Assessing AI agents at autonomous theoretical physics problem solving

Authors: A. Ramesh Chandra, Aaghaz Mahajan, Ainesh Sanyal, Mianqi Wang, Shubhadeep Chakraborty, Sudip Chakraborty, Shailesh Chandrasekharan, Rama Govindarajan, Akshit Goyal, Jonathan J. Heckman, Yasir Iqbal, Divya Jaganathan et al.
Published: 2026-10-09
Categories: hep-th, cond-mat.stat-mech, cond-mat.str-el, cs.AI, gr-qc
arXiv: https://arxiv.org/abs/2610.16733v1

We present a suite of 50 unpublished research-level theoretical physics problems with verifiable answers. Without AI assistance, a typical problem in our collection could be solved by a world expert in about a day or less but would take considerably longer for experts in adjacent specialties. We test the ability of off-the-shelf AI agents to tackle these problems autonomously, without human hints. When given access to tools available in a realistic setting, including the ability to search the literature and write scripts for numerical and symbolic work, the best-performing agent, GPT-6-Astra harnessed in Codex, obtains the final answers with a near-perfect accuracy of 98% at an average cost under USD 3 per problem. On average, agents show only small changes in accuracy when denied web search and access to background research papers, suggesting significant memorization of the literature. These results suggest that state-of-the-art agents can accurately perform concrete calculations required across theoretical physics research. We comment on the broader implications of these findings and discuss possible explanations for the current lack of agent-led breakthroughs in theoretical physics despite their outstanding problem-solving abilities.

---

### Memory Without Curation: Learning Multi-Stream Latent Memory for Robot Policies

Authors: Yilin Wu, Rutav Shah, Shuaiyi Huang, Sumedh A. Sontakke, Homanga Bharadhwaj, Andrea Bajcsy, Max Simchowitz, Guanya Shi
Published: 2026-10-09
Categories: cs.RO, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2610.14406v1

It is widely appreciated that robotic policies require long-term memory to perform complex tasks. However, directly conditioning policies on all past observations leaves policy learning susceptible to learning spurious shortcuts, and introduces large latency at policy deployment. As such, today's memory-augmented policies curate what to remember, either by retaining only recent frames, selected keyframes, or textual summaries, or by learning latent representations of the visual stream alone. In this work, we propose an alternative approach IMPRINT, a policy-agnostic latent memory framework that augments vision-language-action (VLA) models with multi-stream memory without explicitly constraining what the policy should remember. IMPRINT compresses different streams into a latent memory that is updated incrementally at inference, enabling low latency. To avoid shortcut learning, IMPRINT adopts a two-stage training procedure, first supervising memory to reconstruct multiple streams (observations, actions, proprioception) of past information, as well as estimate current task progress, followed by end-to-end finetuning, jointly with the policy, on action prediction. Across both simulation and real-world tasks demanding memory, IMPRINT improves average success over the base VLA by at least 26%, outperforming popular alternatives, and maintaining lowest inference latency among memory-augmented VLAs. Through careful ablations, we demonstrate the utility of incorporating diverse information streams into robotic memory, validating the advantages of foregoing history curation. Project website: https://yilin-wu98.github.io/imprint_website/.

---

### SPROUT: Scaling Agentic Reinforcement Learning for Long-Horizon Software Engineering

Authors: Mengyang Liu, Weiming Wang, Haixin Nan, Yu Wu, Peize Li, Shihong Huang, Xuning He, Xue Jiang, Yihong Dong
Published: 2026-10-09
Categories: cs.SE, cs.CL, cs.LG
arXiv: https://arxiv.org/abs/2610.14405v1

Reinforcement learning (RL) has driven substantial progress in coding agents by relying on verified successful trajectories as learning signals. However, such trajectories are scarce in long-horizon software engineering tasks, where a minor error can invalidate the entire rollout. Repeatedly restarting from the initial state is commonly used yet inefficient, and resetting only the context with working directory misses state changes such as installed dependencies. Therefore, branching from intermediate steps to preserve earlier progress requires restoring both the agent context and execution environments. In this paper, we present SPROUT (SnaPshot-based RollOUT), an agentic RL rollout framework that scales long-horizon coding tasks by resuming and branching from intermediate states. Specifically, SPROUT was built on top of a lightweight infrastructure that checkpoints both the filesystem and the agent context after each assistant turn. A verifier-guided backtracking policy then selects where to resume a failed attempt and how to proceed. Extensive experiments on long-horizon coding tasks show that SPROUT achieves up to 23.6% improvement over pass@k across various LLMs, while using $32$--$52%$ of their tokens. For RL training of Qwen3.8-27B, SPROUT stably outperforms standard GRPO by $13.9$ points on three long-horizon coding benchmarks while cutting the time per training step by $39.5%$. SPROUT thus makes RL on long-horizon coding tasks both more effective and cheaper to scale.

---

### VisualCoding: Learning Executable Skills for Reliable Visual Reasoning

Authors: Liu Jiaqi, Jin Ruinan, Li Xiaoxiao
Published: 2026-10-09
Categories: cs.CV
arXiv: https://arxiv.org/abs/2610.14403v1

Learning reusable skills can improve visual reasoning agents, yet applying a skill still requires the agent to implement or invoke a procedure and produce the final answer. Errors and hallucinations at these steps can prevent acquired experience from translating into reliable single-run performance. We propose VisualCoding, which combines execution-guided acquisition of reusable visual programs with direct adoption of their answers on applicable inputs. Execution-Guided Program Acquisition composes visual primitives and additional code into family-level solvers, then iteratively tests and revises them on labeled training instances. Scope-Aware Direct Execution runs a retained solver and returns its valid in-scope answer. We evaluate VisualCoding on four models across four benchmarks spanning 69 task families. VisualCoding exceeds the strongest evaluated baseline in mean accuracy by 4.45 percentage points with Gemini-3.1-Pro and by 14.78 points when reusing the frozen Pro-acquired artifacts with Flash-Lite. Ablations support execution-guided acquisition and show that direct execution outperforms textual or callable reuse of the same programs. Code is available at https://github.com/jiaq-liu/VisualCoding.

---

### SAT-RL: Training Language Models for Logical Reasoning with Symbolic Solvers

Authors: Jialiang Sun, Kuldeep Meel
Published: 2026-10-09
Categories: cs.AI, cs.SC
arXiv: https://arxiv.org/abs/2610.14402v1

Large language models (LLMs) have achieved strong performance in mathematical and logical reasoning, yet they still suffer from the curse of complexity and struggle as problem difficulty increases with the number of constraint interactions and the size of the search space. In contrast, satisfiability solvers such as Z3 provide scalable and verifiable reasoning through constraint solving under formal theories, but lack the flexibility to interpret natural-language problem descriptions. In this work, we propose SAT-RL, a reinforcement-learning (RL) framework that trains LLMs to interact with symbolic solvers through multi-turn tool execution. SAT-RL leverages synthetic data generation with verifiable solutions, supervised warm-up on solver-augmented traces, and solver-grounded reinforcement learning with tool-aware rewards. Experiments show that SAT-RL substantially improves symbolic reasoning with tools on both in-domain and out-of-domain logic tasks. Our code is available on https://github.com/sunjia72/SAT-RL.

---

### Sparse-TD-JEPA: Improving Zero-Shot Reinforcement Learning With Sparse Latent Task Representations

Authors: Duaa Nakshbandi, Zhiyu An, Wan Du
Published: 2026-10-09
Categories: cs.LG
arXiv: https://arxiv.org/abs/2610.14400v1

Successor-feature-based zero-shot reinforcement learning methods pre-train a task encoder and successor-feature predictor on offline, reward-free data, then recover a policy at test time by solving a closed-form linear regression over a small labeled dataset. We identify an underexplored limitation of this approach: when the task encoder produces dense representations, the inferred reward signal becomes diffuse across the state space, reducing the contrast between goal-relevant states and the background and providing a weaker signal for the policy to follow. We address this limitation by introducing an l1 sparsity penalty that concentrates reward-predictive structure into a smaller effective subspace of the task encoder's latent space. The penalty is applied only during training and requires no architectural change or modification to the closed-form inference procedure. We instantiate this approach as Sparse-TD-JEPA, a single-line of code addition based on Temporal-Difference Joint-Embedding Predictive Architecture (TD-JEPA) that significantly improved its performance in zero-shot reinforcement learning. We evaluate Sparse-TD-JEPA on 65 tasks across the DeepMind Control (DMC) Suite and OGBench, covering locomotion, navigation, trajectory stitching, puzzle solving, and robotic manipulation. Sparse-TD-JEPA improves over TD-JEPA in all 13 domain-level comparisons. On DMC, it increases the average return from 661 +- 6 to 714 +- 3. On OGBench, it increases the average success rate from 38.0 +- 0.8 to 48.4 +- 2.3, with improvements of 13.4-18.0 percentage points on the evaluated antmaze navigation and stitching domains. Overall, Sparse-TD-JEPA achieves the highest mean performance among established zero-shot baselines in nine of the 13 domains. Analysis further shows that sparsity produces a sharper and more task-informative inferred reward landscape.

---

### RealityBench: Simulated Environments to Evaluate a Vision Language Model's Ability to Document Human Physical Work

Authors: Andrew Perrault, Anmol Takiar, Dominik Winecki, Arnab Nandi
Published: 2026-10-09
Categories: cs.CV, cs.AI, cs.CL
arXiv: https://arxiv.org/abs/2610.14397v2

Augmenting worker capabilities in the physical world through guidance has been a decades-long goal for AI agents. One key supporting capability is to understand what the worker is doing, seeing through the human's camera or smart glasses, and assist them in documentation. Despite advances in vision language models, their capabilities in these settings remain uncertain and are challenging to benchmark because most physical work data is proprietary, and involves too much variability to study in a principled manner. Unlike structured settings such as kitchens and warehouses that have predictable lighting, geometry, and objects, field-facing tasks such as rooftop solar installation and vehicle repair are unconstrained and high variance. Here, we model real-world physical work in Blender across eleven environments, incorporating variants in both image acceptability and documentation. The acceptability task asks whether a particular view would be suitable proof-of-work and documentation asks whether an agent can understand the task-relevant state of the environment from that view. Across 26 vision-language models, we find that top models largely solve documentation when the answer is text in frame, but are weak at reading measurements and at judging small parts and spatial relations, such as whether a tape measure leans. Acceptability is harder to pin down because human raters disagree with each other. Measured against a scorer that agrees with the raters' majority about as often as a single rater does, the best models cannot be distinguished from a single non-author rater, at a cost within the range of a human pass. In the three environments where we can trace acceptability errors, most wrong rejections come from misreading the scene rather than the requirement. On the three documentation questions for which we have real images, model rankings on rendered frames transfer to real images.

---

### Clinical value claims in multimodal AI outpace supporting evidence

Authors: Jiayu Feng
Published: 2026-10-09
Categories: cs.LG, cs.CY
arXiv: https://arxiv.org/abs/2610.14395v1

Multimodal prediction models add a data source, such as imaging or an assay, to clinical information, and health systems must decide whether that added test is worth acquiring. We audited 177 studies comparing multimodal with unimodal models: 81 from a systematic review and 96 from a probability sample of 2025-2026 PubMed records. Gains in the area under the receiver operating characteristic curve (AUROC) were smaller when a study fitted a separate model for every data source (adjusted difference, -0.030; 95% CI, -0.056 to -0.005). Within 78 such studies, leaving out the strongest source's model would have raised the average gain from 0.046 to 0.108. This complete benchmarking fell from 69.1% of review studies to an estimated 46.0% of recent ones. An estimated 22% assessed decision utility and none evaluated a model prospectively, yet most claimed potential clinical usefulness or improved care (84.0% and an estimated 87.7%).

---

### Understanding and Mitigating Training Instability in Large-Scale Recommender Systems: A Cascade Perspective

Authors: Yuening Li, Zeyu Wang, Zhen Zhang, Jiayi Liu, Zhimeng Jiang, Yuji Roh, Weikai Xu, Mingyan Gao, Onkar Dalal, Chuan He, Liang Liu
Published: 2026-10-09
Categories: cs.IR, cs.LG
arXiv: https://arxiv.org/abs/2610.14394v1

Training large-scale recommender systems on dynamic industrial platforms, such as Short-Form Video (SFV) services with billions of daily interactions, faces critical stability challenges. Rapid distribution shifts in user data and increasing model complexity frequently trigger a divergence cascade, typically initiating with gradient explosion, propagating to weight explosion, and culminating in loss explosion. This G-W-L cascade severely impedes model convergence and performance. This paper provides a systematic analysis of this G-W-L cascade within an industrial SFV recommender context. We then propose and evaluate a structured mitigation framework where interventions are explicitly mapped to counteract specific stages of the cascade: gradient (G) control, weight/activation (W) stabilization, and loss (L) shaping. Our contribution provides a structured framework for understanding and tackling training divergence, offering practical insights for deploying robust, continuously updated recommenders in large-scale, non-stationary settings. We empirically validate the effectiveness of these strategies on a major industrial SFV recommendation system.

---

### SciCurateBench: Curating Experimental Data from Heterogeneous Scientific Literature with Verbal-Action Agents

Authors: Minuk Ma, Kaiyun Guo, Jiarui Ding
Published: 2026-10-09
Categories: cs.CL
arXiv: https://arxiv.org/abs/2610.14393v1

Scientific research increasingly depends on curating experimental data from literature. Scientific data curation requires verifying exact experimental conditions (e.g., specific assays, clinical outcomes, and data availability) across long, heterogeneous files, which is a highly time-consuming process. Existing benchmarks evaluate only broad relevance on short queries, failing to capture this complexity. We introduce SciCurateBench, a benchmark of 200 expert-annotated papers from major life-science journals that challenges systems to verify conjunctive biological experimental criteria from raw, unprocessed scientific literature. We show that plain LLMs and popular reasoning, tool-using, and feedback-based agents all struggle with this task. To study what makes the task tractable, we further propose LangPoli, an integrative system in which a policy agent generates natural language action guidelines that actor agents convert into executable tools. LangPoli showed strong performance on both Paper Classification (83.16% F1 vs. 58.22% for its base LLM) and Structured Information Retrieval, and also improves re-ranking on the public benchmarks of LitSearch, BEIR, and TREC. We further explore curriculum transfer, a method that distills generalizable strategies from solved tasks to accelerate the solving of related problems. We hope this work facilitates research in truly practical scientific AI assistance. The anonymized pre-release can be found from https://anonymous.4open. science/r/scicuratebench-D885/.

---

### Measuring the Inference Energy Efficiency of Open-Weight LLMs across Benchmarks, Reasoning Modes, and Hardware

Authors: Zeyu Yang, Wes Armour
Published: 2026-10-09
Categories: cs.AI, cs.PF
arXiv: https://arxiv.org/abs/2610.14389v1

Benchmark scores tell us what an LLM can do, but not how much energy is required to deliver that capability. We directly measure this trade-off for 86 open-weight model checkpoints, evaluated on seven benchmarks and six hardware platforms ranging from a consumer GPU to an eight-GPU server. We find large differences in energy efficiency even among models with similar capability. Most configurations lie away from the score--energy Pareto frontier: they use a median of 2.6--4.8$\times$ more energy than another configuration that scores at least as well, with differences reaching 27$\times$. Reasoning can improve capability, but enabling thinking increases energy by about 3.7$\times$ on average, and the benefit depends strongly on the model family and output budget. Generated-token count is also an imperfect proxy for energy, misordering 18--24\% of model pairs. In contrast, FP8 reduces energy by roughly a quarter with little change in score, while changing hardware largely rescales energy without changing the relative ranking of models. These results show that capability alone gives an incomplete picture of efficiency. We argue that benchmark scores should be reported together with measured energy consumption under a specified serving regime and output budget.

---

### Decomposing A Neural Learner

Authors: Zelong Bi
Published: 2026-10-09
Categories: stat.ML, cs.LG
arXiv: https://arxiv.org/abs/2610.14388v1

Neural learners for classification and regression are often criticized as black boxes. We propose a probabilistic decomposition of neural learners that accounts for heterogeneous predictability across the feature space. A training procedure combining the EM algorithm, stochastic gradient descent, and maximum likelihood estimation is developed to jointly learn the predictable component and an input-dependent reliability score. We show that the optimal decomposed learner recovers the response distribution within the predictable component, while the reliability score characterizes the predictability of individual observations. We further extend the framework to incorporate localized feature transformations. Experiments on synthetic and real-world datasets demonstrate the effectiveness of the proposed framework.

---

### Deep learning from a control vantage point

Authors: Álvaro Rodríguez Abella, João Pedro Silvestre, Bahman Gharesifard, Paulo Tabuada
Published: 2026-10-09
Categories: eess.SY, cs.AI, cs.LG, math.OC
arXiv: https://arxiv.org/abs/2610.14385v1

This article offers an overview of recent developments in deep learning from a control vantage point. It reviews several classes of models including feedforward and residual neural networks, transformers, large-language models, and selective state space models while explaining how control theory plays an important role in understanding their training, properties, or architecture design. The article explains how residual neural networks can be interpreted as the discretization of a continuous-time control system thereby enabling the use of control theoretic techniques for its analysis. To illustrate the power of this bridge between control and deep learning, two technical results are discussed: 1) the ability of a residual neural network to approximate arbitrary continuous functions is a non-standard controllability problem that can be solved using geometric nonlinear control and 2) transformers, the key architectural component of most large language-models, can be seen the discretization of a consensus algorithm and thus large language-models exhibit consensus. Another point of contact between control theory and deep learning is the connection between backpropagation, the algorithm used to train neural networks, and optimal control. The article addresses how the older connection between these problems, established in the 80's using discrete-time models can be reconciled with the more recent approach using continuous-time models. Several open problems and directions for future research are discussed, including selective state-space models, a recent class of models inspired by the models used in control.

---

### ANASSA: From High-Level Framework to Low-Level Architecture for Spatial Agentic AI

Authors: Constantinos Papantoniou, Brian Hilton
Published: 2026-10-09
Categories: cs.AI, cs.MA, cs.SE
arXiv: https://arxiv.org/abs/2610.14382v1

Agentic AI systems operating over geospatial data require more than autonomous reasoning and tool use; they must also support explicit orchestration, spatial validation, governance, provenance, failure handling, and human decision authority. Prior work introduced ANASSA, an agentic AI orchestration framework for spatial intelligence that established high-level components and responsibility boundaries for governed agentic execution (arXiv:2609.14824). However, the framework did not specify how those high-level responsibilities should be systematically decomposed into an implementation-facing low-level architecture. This paper addresses that gap by applying Design Science Research (DSR) to refine the ANASSA framework from high-level architectural (HLA) responsibilities into a low-level architecture (LLA). The resulting design preserves the responsibility boundaries of the foundational framework while translating them into executable architectural structures, including 47 logical roles, versioned task graphs, capability-based agent selection, typed contracts, explicit runtime states, spatial validation mechanisms, governance controls, failure and recovery paths, and provenance records. The architecture is demonstrated through an end-to-end execution trace and analytically evaluated for requirements traceability, internal consistency, responsibility preservation, and technology independence. The resulting high-level-to-low-level design provides an implementation-facing bridge between the ANASSA framework and future system implementations, defining how governed spatial agentic workflows can be made inspectable, reconstructable, and testable without binding the architecture to a specific AI model, geospatial platform, communication protocol, or deployment environment.

---

### ACED-Bench: Evaluating How LLM Agents Acquire and Act on Causal Evidence

Authors: Yiwei Chen, Amila Weerasinghe, Siddharth Biswal, Utkrisht Rajkumar, Olcay Boz, Taylor Berg-Kirkpatrick, Rose Yu
Published: 2026-10-09
Categories: cs.AI
arXiv: https://arxiv.org/abs/2610.14381v1

Large language models are increasingly used as scientific assistants, yet we still lack reliable ways to evaluate whether they can investigate a system rather than merely answer a prompt. A useful evaluation must let agents choose experiments, interpret finite evidence, and make causal decisions, while keeping the environment coherent across repeated observations and interventions and preserving objective ground truth. We introduce ACED-BENCH (Active Causal Evidence and Decision Benchmark), a neuro-symbolic benchmark built around this requirement. The key idea is to separate the readable scientific surface from the formal causal world: LLMs generate natural-language scenarios and task text, while hidden probabilistic graphical models define the causal graph, data-generating process, intervention semantics, validation checks, and oracle answers. This yields scientific worlds that are natural for agents to investigate but exact enough for scalable, programmatic evaluation. The main leaderboard ACED-CORE contains 180 tasks: 60 set-valued causal ancestry/descendancy questions and 120 constrained causal-decision tasks. We also report ACED-QUAL, a 150-question yes/no qualification check using the same hidden-PGM interface. We evaluate seven contemporary LLMs and two earlier-generation models across three active agent designs. ACED-QUAL checks whether agents can use the interface to draw conclusions from evidence, while ACED-CORE tests whether they can decide what evidence to collect and when to commit to an answer. Evidence-ledger analysis localizes failures within the investigation: agents test the wrong intervention, lose relevant evidence, or continue after sufficient evidence has appeared. These results suggest that reliable AI co-scientists need explicit support for experiment design, evidence retention, answer commitment, and stopping.

---

### Tokenization-Invariant Watermarking: A Byte-Local Method for LLM Text

Authors: Tung-Ling Li
Published: 2026-10-09
Categories: cs.CR, cs.CL
arXiv: https://arxiv.org/abs/2610.14378v1

We introduce Tokenization-Invariant Watermarking (TI-WM), a method that embeds a watermark during token generation but keys and scores it on canonicalized UTF-8 bytes. The detector does not invoke the tokenizer. Canonicalizer-preserving edits leave its score unchanged; other localized canonical-byte edits affect a bounded number of scoring windows. Across three fixed keys and 128 prompts per key and model, a deterministic white-box attack at a 5% word-edit budget leaves mean attacked-AUC 0.530 on Qwen3-8B and 0.664 on Gemma 4 E4B, compared with 0.410 and 0.489 for Unigram with detection-side deletion. These operating points are not jointly matched on quality and clean detectability. Bias sweeps show model-dependent distortion costs, and shared-rule controls support a locality interpretation without establishing a unique advantage of bytes. Visible insertion outside the canonicalizer's deletion set defeats the default design. Alphabetic filtering closes that channel but does not resolve source shift or calibrated detection power. TI-WM thus provides a concrete construction with local scoring guarantees and measured attack-ranking gains, not a general robustness or low-false-positive-rate deployment guarantee.

---

### DPAL: A Test-Time Adaptation Loss for Financial Time Series Forecasting

Authors: Wuzhou Lian, Haochun Wang, Yuefan Deng
Published: 2026-10-09
Categories: cs.LG
arXiv: https://arxiv.org/abs/2610.14376v1

Test-time adaptation (TTA) enables time series forecasters to update after deployment as new observations become available. However, adaptation objectives that focus primarily on prediction errors at individual time steps do not fully address the needs of financial time series forecasting. In addition to pointwise return fitting, return direction and the price trajectories reconstructed from sequences of returns are also important. To better account for these properties, we introduce Direction- and Price-Aware Adaptation Loss (DPAL), which combines pointwise return fitting with directional and price-calibration terms. We apply DPAL within existing TSF-TTA frameworks without modifying the source forecaster or the adapter architecture. Across a diverse collection of financial datasets, DPAL reduces reconstructed-price mean squared error in most evaluated settings.

---

### DexFLEX: Contact-Aware Foundation Controller for Command-Guided Dexterity

Authors: Weikang Wan, Zeyu Lai, Jianyu Hou, Yuchen Zhou, Jiawei Fu, Henrik I Christensen, Hao Su
Published: 2026-10-09
Categories: cs.RO, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2610.14372v1

Dexterous robot hands can receive useful motion intent from teleoperation or learned policies, but successful execution also depends on contact decisions that these commands rarely specify. We introduce DexFLEX, a contact-aware foundation controller that turns upstream fingertip-motion drafts into contact-consistent joint commands. Instead of treating a draft as a trajectory to copy, DexFLEX treats it as evidence about intent: from the current tactile-proprioceptive state, it proposes feasible short-horizon motion chunks, predicts their contact consequences, and selects the candidate that best follows the command while preserving future contact stability. At inference time, a simple Draft-Dream-Select procedure combines pure-prior proposals for recovery with draft-seeded proposals for responsiveness, then decodes the selected motion into executable joint-space targets. Across degraded-command simulation, real-world shared control, and visuomotor policy learning, the same trained controller improves robustness without retraining, increasing real-world teleoperation success from 29.2% to 78.3% and reaching 46.7% policy-learning success, 3.5 times direct joint-action prediction, and 1.56 times prior-only correction. Videos are available at dex-flex.github.io.

---

### When More Memory Fails: Hidden Feasibility Holes in Dynamic Tensor Rematerialization

Authors: Mahesh Reddy Pagadala
Published: 2026-10-09
Categories: cs.LG, cs.DC
arXiv: https://arxiv.org/abs/2610.14369v1

Dynamic Tensor Rematerialization (DTR) trains neural networks under a memory budget by evicting tensors and recomputing them on demand. A natural expectation is that a budget large enough to complete training stays sufficient when it is increased. Using DTR's reference simulator, simrd, to replay the public DTR training traces, we show that this expectation fails. We call the exceptions feasibility holes: memory budgets at which an eviction policy runs out of memory although both a smaller and a larger budget complete, a form of non-monotone feasibility. Under a written sweep protocol with dated amendments, a 0.01-step budget grid shows no such hole for DTR on any of six confirmatory traces, although unresolved timeouts prevent it from establishing monotonicity. Finer sampling finds holes on InceptionV4 (three regions) and U-Net (two regions), in addition to one on the ResNet-32 development trace. Every hole is reproduced by three repeats and by unmodified simrd. For one InceptionV4 pair (0.2343 versus 0.235 of peak memory) we trace the failure to individual allocations. The larger budget avoids one eviction at an operator where the smaller budget was 173,479 bytes short. The extra data it carries forward forces a later eviction of a tensor needed three operators afterwards, and rebuilding that tensor triggers a recomputation cascade that cannot fit. Keeping that one tensor resident under the same budget prevents the failure. The two simple score changes we tested do not remove holes across traces; both add failing budgets on some traces. All results are from simulation, with no allocator model; behaviour on GPU hardware is untested.

---

### History, Adaptation, and Hidden Gain: Two Prospective Tests of Learned Predictive Control in a Synthetic Neural Population

Authors: Mahesh Reddy Pagadala
Published: 2026-10-09
Categories: cs.LG, eess.SY
arXiv: https://arxiv.org/abs/2610.14368v1

We study learned one-step predictive control of a synthetic population of 64 leaky integrate-and-fire neurons whose light sensitivity is scaled by a hidden factor s, using two prospective confirmations on plants simulated only after the analysis was frozen. In the first (50 plants, 11 conditions), removing the 200-ms input-output history substantially worsened a parameter-count-matched neural predictor at low and moderate gain (s = 0.6, 0.8; RMSE rose by 2.47 Hz at s = 0.6, 97.5% CI 2.43-2.51). In secondary comparisons, recursively adapted ARX had the lowest mean error of the deployable controllers in 10 of 11 conditions, tuned PI with anti-windup matched or beat the network for s >= 0.7, and training for 10,000 rather than 900 epochs lowered prediction error but worsened control. Development work suggested the network's learned action sensitivity did not track the hidden gain. A second protocol (50 new plants, 9 sensitivities, 2 noise levels) confirmed three pre-specified predictions at the primary noise level. Re-estimating a single multiplicative input gain recovered 87% (95% CI 86-88%) of the benefit of full adaptive ARX. Supplying that online estimate to the network as an extra input reduced high-gain tracking error by 1.84 Hz (1.81, 1.88). Longer training worsened the plain network by 1.09 Hz but improved the gain-input network by 1.07 Hz, a differential of 2.16 Hz (2.13, 2.20). Under high process noise (sigma = 2.0) the effects shrank to about 0.04 Hz or less and gain-only adaptation recovered only 24% of the adaptive-ARX benefit. In this synthetic plant, at the primary noise level, one-parameter online gain estimation accounts for most of the gap between a history-based network and full adaptive ARX, and for its prediction-control mismatch.

---

### ConformalGuard: False-Positive-Controlled Action Gating for LM Classifiers

Authors: Bishnu Bista, Dhanraj Haibatrao Jadhav, Vibha Choudhary, Gaurav Saxena, Arun Pandiyan Perumal, Arjun Subedi, Jignesh Ammineni
Published: 2026-10-09
Categories: cs.LG
arXiv: https://arxiv.org/abs/2610.14365v1

Language model classifiers are increasingly deployed for high-stakes decisions such as phishing screening, content moderation, and fraud triage, yet validation-set thresholds provide no finite-sample control of deployment-time false positives. We present ConformalGuard, a model-agnostic calibration layer that wraps a fixed scoring function with finite-sample control over the rate at which legitimate inputs trigger an intervention tier. Our contribution is empirical and operational, not a new conformal theorem: zero-shot LM API scores can be coarse enough (10--11 unique values here) to make temperature scaling and isotonic regression operationally identical to raw thresholding for low-FPR action gating. We compare these methods with conformal calibration for fine-tuned BERT, Claude Sonnet 4, and GPT-4o on public email corpora. For Claude, conformal calibration reduces FPR from 0.040 to 0.008 while recall falls from 0.938 to 0.875. For deterministic GPT-4o scores, the 1% operating point is too conservative to deploy (recall 0.564); three-run stochastic averaging raises recall to 0.732 at FPR 0.005. At 2% prevalence with BERT, conformal calibration removes 25 false-positive reviews per 10,000 emails (95% CI (10, 43)). The guarantee is legitimate-event-rate control under exchangeability; discrimination, workload effects, and robustness to shift remain empirical.

---

### The Sample Complexity of Multicalibration with an ERM Oracle

Authors: Princewill Okoroafor
Published: 2026-10-09
Categories: cs.LG, cs.DS, cs.GT
arXiv: https://arxiv.org/abs/2610.14363v1

A predictor is multicalibrated with respect to a class of subpopulations (formally, of functions on the inputs, called auditors) if, for every subpopulation in the class and every value the predictor outputs, the average outcome among the members of that subpopulation who receive that value is close to the value itself. For finite classes, about $\varepsilon^{-3}$ samples are necessary and sufficient to learn a predictor with multicalibration error $\varepsilon$, but the algorithms that attain this rate enumerate the class, which is impossible for the large classes of interest. We ask how many samples suffice when the learner accesses the class only through an empirical risk minimization (ERM) oracle. We give an algorithm that uses $\tilde{O}(d/\varepsilon^{3})$ samples and polynomially many ERM calls, where $d$ is the VC dimension of the class. Its dependence on $\varepsilon$ is optimal for all learners. When $d$ is at least about $1/\varepsilon$, every learner that can only access the class through polynomially many ERM calls needs $\tildeΩ(d/\varepsilon^{3})$ samples. Learners with unlimited computation need only $\tilde{O}(\varepsilon^{-3}+d\varepsilon^{-2})$ samples, so the extra factor $1/\varepsilon$ on the dimension term is the price of oracle efficiency. For swap multicalibration, which allows a different function from the class for each predicted value, $\tildeΘ(d/\varepsilon^{3})$ samples are necessary and sufficient. For estimating the error of a given predictor, the ordinary notion is the harder one: the swap error can be estimated with an ERM oracle, whereas the ordinary error is NP-hard to approximate even for classes with linear-time ERM. Consequently, there are classes for which it is computationally easier to learn a multicalibrated predictor than to compute the multicalibration error of a given predictor.

---

### When Pleasing the User Compromises Care: AI Sycophancy in Caregiver-Mediated Healthcare

Authors: Priyadarsi Mishra, Arjun Beniwal, Ziyu Yi, Yu Ding, Sheng Liu
Published: 2026-10-09
Categories: cs.AI
arXiv: https://arxiv.org/abs/2610.14360v1

Large language model (LLM) alignment often assumes that the user interacting with a model is also the primary beneficiary of its behavior. Yet in many consequential settings, one person directs the system while another bears the consequences. We study this problem in caregiver-mediated healthcare, where parents, pet owners, and caregivers make decisions for children, animals, and older adults. We introduce an evidence-grounded benchmark spanning veterinary, pediatric, and older-adult care, and test whether AI advisors maintain appropriate care when users express clinically inferior preferences. Holding patient facts fixed, we vary stated care preferences and communication styles and evaluate both initial recommendations and multi-turn follow-up. Across frontier and open-weight models, inferior preferences substantially increase switching from appropriate to inappropriate care beyond neutral conversational drift and increase agreement with the requested action, while communication style alone has a smaller effect. These failures accumulate across turns, including on cases models initially answer correctly, and often carry moderate or severe potential harm. In matched eldercare cases, the effect of an inferior preference is similar whether it is expressed by the patient or by a caregiver. Together with models' limited tendency to elicit patient preferences in caregiver conversations, this suggests that current models are highly sensitive to expressed user preferences but less sensitive to whose preferences are represented and who bears the consequences. Our results expose a broader limitation of user-centric alignment in settings with multiple stakeholders.

---

### Actionable Meta-Requirements for AI Robustness Auditing

Authors: Joshua Owotogbe, Roya Nasiri, Indika Kumara, Damian Andrew Tamburri, Willem-Jan van den Heuvel
Published: 2026-10-09
Categories: cs.AI, cs.SE
arXiv: https://arxiv.org/abs/2610.14356v1

AI-enabled systems must remain dependable when the conditions under which they operate change (e.g., due to input perturbations, distribution shifts, or changes in external dependencies). This makes robustness auditing of such systems important for judging whether the available evidence is sufficient to support a defined robustness conclusion. However, designing an AI robustness auditing framework is challenging. A key unresolved issue is determining the requirements such a framework should support; these requirements have not yet been consolidated. We address this gap through meta-requirements: higher-order requirements specifying what a class of auditing frameworks should support. In design science research, literature and domain experts are common sources of meta-requirements; accordingly, we derive the initial set from literature and validate and refine it with practitioners. Our systematic literature review retained 74 of 2,890 records and derived 43 meta-requirements across 16 categories. We evaluated them with 128 practitioners, obtaining 5,504 ratings and 586 substantive comments. Overall, 81.20% of ratings indicated agreement or strong agreement. In particular, 41 of the 43 literature-derived meta-requirements were classified as strongly supported. Practitioner feedback refined the requirements and identified an additional requirement concerning external components and supply chain dependencies. The final set comprises 44 meta-requirements across 16 categories, of which 43 received quantitative validation. These meta-requirements can help organizations implement AI robustness auditing frameworks, assess existing frameworks for completeness, and refine them.

---

### Class-Conditional Conformal Prediction for Signal Temporal Logic Inference with Many Classes

Authors: Yixuan Wang, Danyang Li, Matthew Cleaveland, Roberto Tron, Mingyu Cai
Published: 2026-10-09
Categories: cs.LG
arXiv: https://arxiv.org/abs/2610.14352v1

Signal Temporal Logic (STL) inference provides interpretable formulas for time series classification. For multiclass classification, however, accurate predictions do not necessarily ensure that the inferred formulas are satisfied consistently with the true class. We introduce the positive satisfaction rate and codeword satisfaction rate to quantify satisfaction of required formulas and consistency with the complete class codeword, respectively. We develop a two-stage procedure that first learns discriminative formulas using a margin-based objective and then refines predicate thresholds toward a small positive robustness target while fixing the formula structure and temporal intervals. Satisfaction and ranking penalties discourage violations of required formulas and loss of class separation. The decoding loss is used as a nonconformity score and calibrated separately for each class, providing finite-sample class-conditional coverage under intra-class exchangeability. Experiments on naval, synthetic, and robot manipulation trajectories demonstrate discrepancies between classification accuracy and formula satisfaction. The results further reveal a data-dependent tradeoff between robustness tightness and satisfaction, while class-conditional calibration mitigates undercoverage that can be hidden by pooled calibration.

---

### Learning Transferable World-Action Models from Task-Paired Human Videos

Authors: Jeongeun Park, Taekyung Kim, Dongyoon Han, Sangdoo Yun
Published: 2026-10-09
Categories: cs.CV, cs.RO
arXiv: https://arxiv.org/abs/2610.14351v1

World-action models inherit broad visual and physical priors from video pre-training. However, standard within-video next-frame prediction allows them to rely on appearance shared between the context and future, rather than learning interactions that transfer across scenes and embodiments. We introduce WATCH, a cross-video pre-training that learns from task-paired human demonstrations collected across diverse environments. Conditioned on the video and actions of one demonstration, the model jointly predicts another demonstration's actions and future frames from its current observation. Predicting one demonstration from another encourages the model to learn task-relevant interactions that transfer across scenes and demonstrators. Our pre-training shows that these capabilities can be learned from readily scalable human-human pairs alone, without requiring matched human-robot demonstrations. Analyses before robot post-training show improved motion alignment across embodiments and better action and video prediction on unseen robots, with representations closer to those learned from robot pairs. After robot post-training, WATCH improves MimicDroid in-context success from 40.9% to 68.3% with Cosmos Policy and from 59.8% to 77.7% with Cosmos3, with the largest gains on the unseen levels. Gains persist under context-free post-training across robot embodiments. WATCH outperforms within-video pre-training and co-training baselines, supporting cross-video prediction as a scalable approach to learning transferable world-action representations.

---

### Natural Code Transformation Confuses Agents in Vulnerability Analysis

Authors: Jie Zhu, Ronik Bhaskar, Heather Zheng, Ben Y. Zhao, Kexin Pei
Published: 2026-10-09
Categories: cs.CR, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2610.14350v2

Having demonstrated a strong potential for automating vulnerability discovery, AI agents can also be used to attack software systems at scale. Today's software systems need mechanisms to level the playing field by slowing down potential attacks, giving them time to patch vulnerabilities and strengthen software. Along this line, our study of agent trajectories in vulnerability analysis shows a surprising finding: agents produce dramatically different results across real-world code commits that carry the same functionality and the same vulnerability. This phenomenon appears consistently in multiple code domains and vulnerability categories. It motivates us to study natural and functionality-preserving code transformations derived from real-world software development, and their potential as a defense against attacker agents. We build a corpus of natural transformations by mining recurring development patterns from real-world code commits, and apply them to evaluate impact on agentic vulnerability analysis. Our results show that even a single form of transformation can effectively "confuse" frontier agents, e.g., Codex with GPT-6 Sol, reducing successes in vulnerability discovery. We also study different agent failures, whether transformations can be composed to provide stronger protection, and adaptive attackers who seek to reverse them. Overall, our study of natural code transformation significantly expands the research space of using code transformations to defend against malicious agents.

---

### Multimodal Co-Generation With Continuous Bitstream Diffusion

Authors: Georgios Batzolis, Georgios Kontos, Bruno Trentini, Pranam Chatterjee, Micha Livne, Mark Girolami, Luca Ambrogioni
Published: 2026-10-09
Categories: cs.CL
arXiv: https://arxiv.org/abs/2610.14349v1

We show that continuous bitstream diffusion (CoBit), originally developed for language modeling, provides a unified framework for joint and conditional multimodal generation using the same training and sampling protocol. Across image-text, speech-text and protein sequence-structure modeling, models are trained separately from scratch, but each uses a single input projection, transformer backbone, bitwise prediction head and binary score-matching objective shared across its modalities. Within each setting, a single checkpoint supports joint generation and the conditional directions included in training by changing which parts of the bitstream are observed. We first validate the approach on MNIST-Sum, a controlled image-text task pairing raw binary images with equations encoded as text, achieving near-perfect conditional accuracy and high joint consistency without learned tokenizers. We then apply the same protocol at larger scale to speech-text and protein sequence-structure modeling. Trained without text pretraining, a single speech-text model performs speech synthesis, recognition, continuation, and unconditional co-generation. Jointly generated speech and text achieve 3.5% word error rate between the generated text and automatic transcriptions of the generated speech. In protein modeling, a single CoBit checkpoint supports forward folding, inverse folding, and joint sequence-structure generation. For jointly generated proteins spanning 100-500 residues, computational refolding of CoBit's sequences yields structures that agree with the generated backbones at a mean self-consistency TM-score of 0.911. Together, these results show that CoBit provides a common generative framework for substantially different modalities and forms of cross-modal dependence, without introducing a domain-specific objective or sampling procedure.

---

### Evaluating LLMs as Agentic Workflow Executors in Real-World Customer Support Skills

Authors: Md Tahmid Rahman Laskar, Gundeep Singh, Xue-Yong Fu, Karol Chang, Kevin Sanders, Shi Zong, Tania Habib, Julien Bouvier Tremblay, Shayna Gardiner, Harsh Saini, Matthias Lee, Elena Khasanova et al.
Published: 2026-10-09
Categories: cs.CL
arXiv: https://arxiv.org/abs/2610.14346v2

Enterprise customer-support agents increasingly rely on Large Language Models (LLMs) to execute bounded, tool-using workflows. We study this problem through the lens of Skill LLMs: specialized executors that operate within a selected skill context and must complete a narrow, stateful workflow under policy and tool constraints. We present a benchmark construction pipeline that mines recurrent workflows from real-world transcripts, converts them into structured skill descriptions, maps them to tool requirements, simulates environment state with Model Context Protocol (MCP)-compatible mocks, and generates multi-turn tool-using dialogues. Using this benchmark, we conduct a large-scale zero-shot evaluation of 17 language models spanning closed model families and deployment-oriented open-weight models from 1.7B to 14B parameters, each judged independently by five strong LLM judges, plus a deterministic rule-based evaluator that checks tool coverage, hallucinated tool use, and unsupported claims. We find a consistent ranking in which GPT-5.4 leads by a wide margin and QWEN3.5-9B is the strongest open-weight model (below 20B), outperforming several closed baselines under every judge. We also find that judge choice is not a negligible methodological detail: pairwise judge agreement on the pass/fail verdict averages Kappa=0.524 across 10 judge pairs. These results suggest that comprehensive zero-shot evaluation of skill execution requires judge reliability as carefully as model capability.

---

### On the Depth Transfer of Neural Optimizers via L-Smoothness

Authors: Maximilian Wong, Chenyu Xu, Jiajin Li, Yiping Lu
Published: 2026-10-09
Categories: cs.LG, math.OC
arXiv: https://arxiv.org/abs/2610.14344v1

Scaling deep learning models efficiently requires predictable performance transfer across compute regimes. Establishing reliable hyperparameter transfer allows optimal training dynamics and structural choices to be discovered on low-cost proxy models, avoiding catastrophic failures during massive-scale training. Existing depth-transfer analyses are architecture-specific and heuristic: hyperparameter transfer is only justified through toy blocks, and both rely on infinite-depth dynamics, a specific initialization, and feature-learning criteria. Building on $L$-smoothness framework of Xu et al. (2026), we understand hyperparameter transfer as how do the Lipschitz and $L$-smoothness constants of the loss scale with the depth $K$? We show that inverse-depth residual branch scaling $α=Θ(1/K)$ is sufficient to make both constants depth-independent at fixed width, uniformly over the bounded parameter domain and for arbitrary $p\to q$ geometries. Because the bounds depend on each residual branch only through its Lipschitz and smoothness constants, the argument applies to general residual networks (including branches of block depth $\geq 2$ such as transformer and ResNet blocks) without assumptions on initialization or correlations between layer weights. Since the smoothness constant controls the curvature penalty in the descent inequality, our bounds motivate a learning-rate scale without an additional depth-dependent curvature correction for updates bounded in the prescribed norm. This provides a geometric explanation of inverse-depth residual scaling and its application to the operator-norm geometries associated with AdamW, Muon, and MOGA. Learning-rate sweeps on GPT-2 (depths 2--128) and CIFAR-100 ResNets (8--32 residual blocks) confirm that Muon and MOGA exhibit stable learning-rate transfer under these prescriptions, with a common optimal rate across a 64-fold increase in depth.

---

### Continuous-Time Cloth Dynamics with Physics-Informed Neural ODEs

Authors: Yingdong Ru, Lipeng Zhuang, Zhuo He, Gerardo Aragon Camarasa
Published: 2026-10-09
Categories: cs.LG, cs.RO
arXiv: https://arxiv.org/abs/2610.14342v1

We present ContinuousCloth, a continuous-time, physics-informed neural solver for deformable cloth dynamics that enforces a spring--mass ODE within a neural trajectory model. Unlike classical time-stepping simulators that return trajectories only on a discrete integration grid, ContinuousCloth supports querying at arbitrary timestamps and computing displacement, velocity, and acceleration via automatic differentiation. We evaluate ContinuousCloth along two dimensions. In a physics-only setting, ContinuousCloth achieves trajectory accuracy comparable to high-precision solvers across six diverse boundary-condition scenarios while reducing runtime by up to 17.4 times. When fused with sparse and noisy observations, the spring--mass ODE acts as a strong inductive bias for downstream tasks: interpolation and parameter estimation. That is, ContinuousCloth reduces interpolation error by 7.3 times over a data-only neural baseline in simulation and 2.4 times on real-world cloth sequences, and achieves 2.6 times lower parameter error than a differentiable-solver baseline under high-noise conditions, demonstrating consistent robustness under realistic measurement imperfections.

---

### STREGO: A Scalable Implementation of a Trust-Region Framework for Noisy Bayesian Optimization

Authors: Youssef Diouane, Issmail El Hallaoui, Meisseme Kadri, Amina Lamghari
Published: 2026-10-09
Categories: math.OC, cs.LG
arXiv: https://arxiv.org/abs/2610.14338v1

TREGO, a trust-region Bayesian optimization (BO) method for expensive blackbox problems, has convergence guarantees, but only for exact function values. In high dimension, its performance tends to deteriorate. This paper asks whether a trust-region BO method can be extended to handle noisy evaluations while being competitive in high dimension. We first generalize TREGO into a trust-region framework for noisy BO and extend its convergence analysis to noisy function estimates that are sufficiently accurate with sufficiently high probability. We then propose STREGO, a scalable implementation of this framework. Its global phase selects a batch of candidates by solving a bi-objective problem on the posterior mean of a Gaussian process and the integrated variance reduction. Unlike the predictive variance, this second objective does not favor boundary points merely because they are far from the data. Its local phase minimizes the posterior mean of a second Gaussian process, trained only on points close to the incumbent. STREGO is compared with TREGO, TURBO, BAXUS and NSMA-BO on synthetic and real-world problems of dimension $4$ to $124$, with and without noise. It gives the best average results in every synthetic setting tested and the best or tied-best solution on four of the six real-world problems. An implementation of STREGO is available at https://github.com/Mayousk/STREGO.

---

### Which Metadata Matters for LLM Tool Selection in MCP? Evidence from Controlled Factor-Level Interventions

Authors: Xiaofan Li, Fengchun Qiao, Xing Gao
Published: 2026-10-09
Categories: cs.AI, cs.CL, cs.SE
arXiv: https://arxiv.org/abs/2610.14334v1

In Model Context Protocol (MCP) systems, LLMs select tools from metadata rather than from their implementations. Names and descriptions summarize what a tool does, while the parameter schema specifies the fields a tool accepts and how they are organized. Our analysis of metadata for 367,174 MCP tools reveals substantial schema variation among functionally similar tools. Many of these tools share names and similar descriptions but expose different parameter schemas. This raises a key question: can parameter-schema differences influence tool selection before any call is made? We study this question through controlled metadata interventions across five LLMs and five tasks. Each comparison changes exactly one metadata factor (server name, tool name, description, or parameter schema) while keeping the others fixed. We balance tool positions and use conditional logistic regression to estimate position-adjusted effects. We find that: (1) parameter-schema changes produce the largest and most consistent selection shifts among the tested interventions. Models tend to favor separate top-level parameters over grouped or composite forms. (2) Descriptions that explain parameters also tend to increase selection, while name effects are weaker and less consistent. (3) Targeted comparisons show that schema effects depend on the specific parameter names and on how the request represents the same information. When only one schema satisfies an explicit request requirement, all five models select the compatible tool despite position preferences. These results show that parameter schemas influence not only how tools are called, but also whether they are selected. For MCP developers and evaluators, schema design and tool ordering can therefore affect which implementation is invoked even when functionality is unchanged.

---

### Decentralized Bandit Learning of Markov Perfect Equilibria

Authors: Felix Höfer, H. Mete Soner
Published: 2026-10-09
Categories: cs.LG, math.OC
arXiv: https://arxiv.org/abs/2610.14333v1

We study decentralized bandit learning in discounted infinite-horizon, general-sum Markov games with continuous actions. Agents observe the public state, but only their own realized rewards, and use simultaneous perturbation stochastic approximation to update their actions. Under strong monotonicity of the negative running-reward gradients and sufficiently small time steps, we prove that the fully decentralized procedure converges almost surely to the unique stationary Markov perfect equilibrium. We obtain an $O(n^{-1/2})$ convergence rate and present a natural extension to episodic finite-horizon learning.

---

### TraceSampler: Making Sampling Errors Predictable and Correctable for Accelerated Generative Sampling

Authors: George Matthew, Guoqiang Zhang, Kenta Niwa
Published: 2026-10-09
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2610.14332v1

Accurate sampling from continuous-time generative models can require many model evaluations. Low-step sampling reduces cost, but coarse updates introduce endpoint errors that can accumulate and degrade generation quality. We introduce TraceSampler, a post-training method that corrects these errors without modifying the pretrained model. Rather than optimising low-step updates directly, TraceSampler models endpoint error as a sample-specific prediction problem. A higher-step reference integration over the same interval provides the supervised target, while a compact representation makes high-dimensional error predictable from the sampling trace. This representation combines shared residual structure, trace-derived directions, and a low-rank remainder, so shallow per-step MLP heads predict sample-dependent coefficients. These coefficients drive a two-stage learned residual correction: the first stage predicts the full endpoint residual, while an intermediate backbone evaluation provides new trace information to predict the remaining residual; this evaluation is reused by the following step. TraceSampler is evaluated with Rectified Flow on CIFAR-10, SiT-XL/2 on ImageNet-256, and SD3.5 Medium for text-to-image generation. At 10 steps, it reduces trajectory MSE relative to matched-step Euler by 90.5% on CIFAR-10 and 85.6% on ImageNet-256, while achieving FID scores of 2.692 and 2.189, close to 120-step Heun references of 2.601 and 2.056 with 24$\times$ fewer backbone evaluations. On SD3.5 Medium, TraceSampler improves image quality and prompt following; at 10 steps, ImageReward rises from 0.752 to 0.906 while trajectory MSE falls by 70.7%. These gains preserve Euler's backbone-evaluation budget, add less than 0.4% inference FLOPs, and incur negligible wall-clock overhead, showing that coarse-step sampling error can be made predictable from the sampling trace and efficiently corrected.

---

### Real-Time Traffic Density Estimation: A Hybrid Computer Vision and Predictive Machine Learning Pipeline

Authors: Prince Kwabena Appiah Boadu
Published: 2026-10-09
Categories: cs.CV
arXiv: https://arxiv.org/abs/2610.14326v1

Traffic density estimation from camera feeds requires connecting visual perception with short-horizon forecasting. We present a two-stage pipeline that combines vehicle detection and tracking with a spatial logic layer, followed by a temporal estimator. The spatial layer filters bounding boxes by their overlap with a precomputed road mask and extracts per-frame traffic features. LSTM, GRU, and Temporal Convolutional Network (TCN) estimators use these features to predict five future steps from a ten-frame input window. We benchmark ten detector variants from six families on a dataset of 5,716 annotated frames using a single NVIDIA GPU. YOLOv5l achieves the highest mAP@50:95 of 0.524 at 56 frames per second, while YOLOv8n reaches 258 frames per second at 0.487 mAP@50:95; these rates measure detector inference without tracking or spatial filtering. Temporal evaluation on 29,243 held-out windows yields aggregate R-squared values of 0.858 for GRU, 0.856 for LSTM, and 0.840 for TCN. These metrics measure forecasts of detector-derived telemetry and are dominated by direction variance, rather than establishing absolute traffic-density accuracy. No road mask was available for the evaluation stream, limiting validation of the spatial filtering component. A live run on a third camera stream examines forecasting behavior within the complete pipeline. The results characterize detector speed-accuracy trade-offs and the limitations of short-window traffic telemetry forecasting.

---

### Manifold Navigation Learning Using Recurrent Networks: Efficient Training Techniques

Authors: Sarah Harkins Dayton, Sathya Rengaswami, Vasileios Maroulas, David L. Boothe, Ioannis D. Schizas
Published: 2026-10-09
Categories: cs.CV
arXiv: https://arxiv.org/abs/2610.14325v1

Position tracking is a well-studied problem in Euclidean navigation, yet comparatively little attention has been given to navigation on general manifolds, where motion is governed by the underlying geometry of the space. In this work, we formalize the problem of manifold navigation as a velocity-driven tracking task and propose a coordinate-chart-based learning framework in which learning models operate entirely in a Euclidean representation. Predicted trajectories are mapped to the manifold via an immersion function prior to error computation, allowing geometric consistency to be evaluated without explicitly enforcing manifold constraints during learning. Within this framework, we focus on Gated Recurrent Units (GRUs) for sequential position prediction on curved manifolds after comparisons with alternative recurrent architectures. We evaluate our GRU models, trained with pertinent custom loss functions, against both a standard MSE loss function and a naive vector-addition baseline. We evaluate our approach on both analytically (and numerically) defined manifolds, as well as a surface reconstructed from a LiDAR point cloud. Our results show that recurrent models can learn to approximate manifold-constrained dynamics from velocity sequences and consistently outperform Euclidean baselines. Our results demonstrate that certain custom loss functions significantly improve long-horizon prediction accuracy compared to conventional point-wise loss functions.

---

### EV-JEPA: Elastic V-JEPA for Any Device and Cross-Device Inference

Authors: Anxhelo Diko, Jieyu Lin, Stefano Ambrogio, Ziyun Li, Mahmoud Assran, Théo Moutakanni, Barbara De Salvo, Chiao Liu
Published: 2026-10-09
Categories: cs.CV, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2610.14324v2

Foundation models like V-JEPA are increasingly deployed across a heterogeneous edge-cloud continuum spanning smart glasses, phones, robots, and servers. Yet, they are typically trained as rigid, fixed-capacity networks bound to one device, supporting neither reconfigurable compute budgets nor collaborative execution. Existing approaches address only one axis at a time: slimmable and NAS-based methods provide width elasticity but assume single-device inference, whereas split-computing methods enable cross-device collaboration only for fixed partitions and static capacities. We introduce Elastic V-JEPA (EV-JEPA), a single model trained once and deployable everywhere along the continuum. Its width is reconfigurable, and it is natively partitionable across heterogeneous devices. At its core is Transition-Aware Triplet Training (T3). At each step we pick a narrow and a wide width, and train three paths: each width alone, plus a hybrid path that runs the early layers narrow and the rest wide. The hybrid switch point is where one device hands off to another, so the same weights learn both single-device and cross-device execution. Across various benchmarks, EV-JEPA falls within 1 accuracy point of independently trained baselines while improving over previous elastic training approaches by up to 11 accuracy points. Narrow-to-wide cross-device hybrid execution recovers to within 1.1 points of the highest-capacity standalone model and supports many-to-one, any-to-any, and variable-split configurations. Finally, we show that the elastic principle extends beyond the backbone to an elastic token compression mechanism that exposes controllable trade-offs among on-device computation, latency, and accuracy.

---

### De-biased Distribution Estimation for Locally Differential Private Survival Data

Authors: Yi Xiong, Li-Pang Chen, Liyue Fan
Published: 2026-10-09
Categories: stat.ME, stat.ML
arXiv: https://arxiv.org/abs/2610.14320v1

Sharing time-to-event data can leak sensitive information about individual participants in clinical studies. To protect the privacy of data, differential privacy (DP) framework offers rigorous protection for data sharing. The local variant of differential privacy is particularly appealing because each record is perturbed before it leaves the data owner or contributor. However, the injected perturbation distorts the true distribution of the source data, yielding low utility and invalid statistical inference with the released data. In this work, we propose a nonparametric estimator for the true survivor function via the deconvolution kernel function with locally differential private (LDP) right-censored survival data in which both the event times and the event indicators are privatized. Our estimator simultaneously de-biases the perturbation applied to the times and the errors induced by flipping the event indicators due to LDP. Theoretically, we establish the uniform consistency and point-wise asymptotic normality of the de-biased distribution estimator, and derive its uniform convergence rate together with the optimal bandwidth attaining that rate. Simulation studies and the analysis of a semi-synthetic dataset generated from the METABRIC breast cancer cohort demonstrate the performance of the proposed estimator across privacy budgets, censoring levels, and sample sizes. By the post-processing property of DP, the proposed de-biased estimator inherits the privacy guarantee while restoring valid inference on the survivor function.

---

### MvFFN: Multi-view Floor plan Feed-forward Network for Unposed Wide-Baseline Panorama Layout Reconstruction

Authors: Yuguang Li, Yichuan Deng, Zixuan Liu, Ivaylo Boyadzhiev, Xiuyuan Wu, Xinyu Zou, Linda Shapiro, Alex Colburn
Published: 2026-10-09
Categories: cs.CV, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2610.14315v2

Reconstructing accurate camera poses and floor-plan layouts from sparse, unposed, wide-baseline RGB panoramas remains a challenging open problem. Direct multi-room floor-plan prediction is especially difficult in this setting: pose, geometry, semantics, and global layout structure are tightly coupled, yet a complete floor-plan is hard to learn as a single global output. Our key insight is to keep the model's predictions strictly local. Multi-view Ordered Wall Instance Segmentation (MOIS) is designed around this concept. Each panorama column predicts only the wall it observes plus the next two adjacent walls in clockwise order. The global wall partition, room layouts, and shared coordinates emerge from deterministic aggregation of overlapping local predictions. Building on MOIS, we propose Multi-view floor-plan Feed-Forward Network (MvFFN), which jointly predicts coarse camera poses, dense global geometry, and per-wall semantic outputs. A downstream Layout Chaining (LC) post processing pipeline then assembles these predictions into multi-room floor-plans with shared global coordinates, consistent wall identities, and connected room structure. The dense MOIS predictions recover cross-view wall identity and local within-room connectivity more accurately than composed modular baselines, and the assembled layouts substantially outperform prior methods on wall and junction level accuracy. We also introduce the ZInD-CrossView dataset, which augments ZInD with globally unique wall instances and cross-view segmentation labels for this task.

---