# Research Papers — 2026-07 · bundle page 147 of 222

> Full-text markdown of papers #7301–7350 of 11075 published in 2026-07, newest first. AI-readable mirrors of the full 11075-paper 2026-07 archive (50 papers per file).

Canonical page: https://aipod.dev/papers/archive/2026-07/page/147
[← Previous bundle](https://aipod.dev/papers/archive/2026-07/page/146.md)
[Next bundle →](https://aipod.dev/papers/archive/2026-07/page/148.md)

### Equal Accuracy, Unequal Evidence: Search APIs as Decision Surfaces for Tool-Using Agents

Authors: Sriram Selvam, Anneswa Ghosh
Published: 2026-07-11
Categories: cs.CL
arXiv: https://arxiv.org/abs/2607.10198v1

Search APIs are the fundamental retrieval layer for many agents and are often their most frequently used tool. Traditional search APIs provide URLs, titles, and snippets that preview website contents. Because full-page retrieval is token-intensive, agent retrieval architectures increasingly use progressive disclosure: the agent first sees snippets and then chooses whether to fetch full pages. In such systems, search API performance is often evaluated primarily by answer accuracy. We argue that a commercial search API is better understood as a decision surface: the ranked snippets, URLs, and metadata that determine whether an agent answers immediately, searches again, or spends tokens opening pages. We test this claim with one frozen GPT-5.4 agent, two tools (search_web and fetch_page), and 100 questions from SEALQA-HARD, varying only the search provider (Brave, Tavily, Firecrawl). A Kimi-K2.6 oracle labels every content element visible to the agent (URL, title, snippet, and fetched page, when fetched), producing 6,869 valid per-URL judgments. We use an audited correct-answer label, semantic match, which preserves exact matches while accepting harmless formatting and naming variants. Under this measure, the providers remain close (25, 25, 26 / 100), but their evidence economies differ sharply: Brave offers gold-answer-rich snippets, Tavily concentrates gold-supporting URLs at rank 1, and Firecrawl is associated with broader exploration under this fixed agent policy. We also introduce a surface contradiction-to-gold URL ratio, which varies from 0.92 to 2.59. Provider choice is therefore a retrieval-budget and policy decision, not merely a recall decision.

---

### GRATE: Temporal Extensions for Inductive KG Foundation Models via Gated Rotary Attention

Authors: Jiaxin Pan, Osama Mohammed, Daniel Hernández, Steffen Staab
Published: 2026-07-11
Categories: cs.AI
arXiv: https://arxiv.org/abs/2607.10197v1

Knowledge graph foundation models such as Ultra and Trix achieve strong inductive transfer by learning relation-graph representations that generalise to unseen entities and relations. Extending this transferability to temporal knowledge graphs (TKGs) remains challenging: existing temporal models tie their parameters to dataset-specific entities, relations, or timestamps and are not designed to transfer to TKGs with disjoint vocabularies. We propose GRATE (Gated Rotary Attention for Temporal Encoding), an entity-side message function that adds no learnable parameters and encodes time through relative time differences by rotating each edge message according to its time gap to the query and applying a query-conditioned gate to select temporally relevant signals. GRATE integrates into NBFNet-style KG foundation models while preserving structural transferability. Existing TKG benchmarks evaluate within shared train/test vocabularies and cannot directly test cross-dataset temporal transfer; we therefore construct GDELTIndT and WIKIIndT, inductive transfer benchmark suites with disjoint entities, relations, and timestamps spanning both interpolation and extrapolation. Across these benchmarks and held-out forecasting datasets, a single jointly pretrained GRATE checkpoint improves over the static base model in most settings.

---

### Generative Augmentation of Raman Spectra for Glioma Classification

Authors: Andrei Iuşan, Iulian Vasile, Daria Voiculescu, Ion Petre, Andrei Păun, Bogdan Oancea, Mihaela Păun
Published: 2026-07-11
Categories: cs.LG
arXiv: https://arxiv.org/abs/2607.10196v1

Access to sufficiently large biomedical datasets remains a major obstacle for machine learning in Raman spectroscopy-based diagnostics. In particular, for glioma analysis, datasets are typically small and heterogeneous, affected by acquisition-specific variability. This work investigates the utility of deep generative augmentation in such a small-cohort setting. We analyze glioma biopsy spectra acquired from 58 tumor samples and consider both binary IDH-status classification and 6-class methylation subtype classification problems. To address the limited size and imbalance of the dataset, we develop a conditional variational autoencoder ($β$-CVAE) capable of generating class-conditioned synthetic Raman spectra. The generated data are evaluated in Train-on-Synthetic, Test-on-Real (TS/TR) and Train-on-Synthetic+Real, Test-on-Real (TSR/TR) settings under a strict patient-isolated cross-validation protocol. Models trained exclusively on synthetic data underperform models trained on real spectra, indicating a substantial domain gap between synthetic and real distributions. However, augmenting the real training data with synthetic spectra consistently improves classification performance across multiple models. These findings indicate that, even with a limited number of independent patient samples, generative models can capture sufficient structure to provide useful regularization for downstream classifiers. We also investigate a reconstruction-based inference strategy, termed Classification by Reconstruction (CbR), in which class prediction is based on reconstruction error under different class conditions. Overall, the results support the use of deep generative augmentation as a practical strategy for improving machine learning robustness in Raman spectroscopy applications characterized by limited biomedical datasets.

---

### Instruction Set and Language for Hypergraphs

Authors: Mario Pascual-Gonzalez, Ezequiel Lopez-Rubio
Published: 2026-07-11
Categories: cs.CL, cs.AI, cs.PL
arXiv: https://arxiv.org/abs/2607.10194v2

We present IsalHG, a method for representing the structure of any finite, connected hypergraph of bounded hyperedge arity as a string over a compact instruction alphabet $Σ_{\mathrm{HG}}$. The encoding is executed by a small virtual machine comprising a sparse hypergraph, a circular doubly-linked list (CDLL) of node references, and $k$ traversal pointers, where $k$ bounds the hyperedge arity. Instructions either move a pointer through the CDLL or insert a hyperedge, optionally together with new nodes, into the hypergraph. Every string over $Σ_{\mathrm{HG}}$ decodes to a valid hypergraph; the alphabet is closed. A greedy \emph{HypergraphToString} (h2s) algorithm encodes any connected hypergraph into a string; a backtracking variant seeded at nodes of lexicographically maximal structural tuple produces a \emph{canonical string} $w^{*}$, which we conjecture to be a complete isomorphism invariant. Canonical-string equality then decides hypergraph isomorphism natively, without the standard reduction to the Levi incidence graph followed by a graph-isomorphism engine. We verify the round-trip property $s2h(h2s(H)) \cong H$ on 150 connected random uniform hypergraphs and on named combinatorial designs, and we benchmark the canonical algorithm against the three practically available exact baselines -- nauty, Traces, and bliss operating on the 2-coloured Levi graph -- across a $(n, c)$ grid with ten seeds per cell. All four methods agree on every one of 600 isomorphism verdicts, consistent with the completeness conjecture. On wall-clock time the Levi baselines dominate every tested cell by three to five orders of magnitude (geometric-mean ratio $311\times$ to $117{,}672\times$), which we report as measured. We contribute the representation framework, a conjecture of canonical completeness, and the first native-versus-Levi benchmark for hypergraph isomorphism.

---

### Instruction Set and Language for Hypergraphs

Authors: Mario Pascual-Gonzalez, Ezequiel Lopez-Rubio
Published: 2026-07-11
Categories: cs.CL, cs.AI, cs.PL
arXiv: https://arxiv.org/abs/2607.10194v1

We present IsalHG, a method for representing the structure of any finite, connected hypergraph of bounded hyperedge arity as a string over a compact instruction alphabet $Σ_{\mathrm{HG}}$. The encoding is executed by a small virtual machine comprising a sparse hypergraph, a circular doubly-linked list (CDLL) of node references, and $k$ traversal pointers, where $k$ bounds the hyperedge arity. Instructions either move a pointer through the CDLL or insert a hyperedge, optionally together with new nodes, into the hypergraph. Every string over $Σ_{\mathrm{HG}}$ decodes to a valid hypergraph; the alphabet is closed. A greedy \emph{HypergraphToString} (h2s) algorithm encodes any connected hypergraph into a string; a backtracking variant seeded at nodes of lexicographically maximal structural tuple produces a \emph{canonical string} $w^{*}$, which we conjecture to be a complete isomorphism invariant. Canonical-string equality then decides hypergraph isomorphism natively, without the standard reduction to the Levi incidence graph followed by a graph-isomorphism engine. We verify the round-trip property $s2h(h2s(H)) \cong H$ on 150 connected random uniform hypergraphs and on named combinatorial designs, and we benchmark the canonical algorithm against the three practically available exact baselines -- nauty, Traces, and bliss operating on the 2-coloured Levi graph -- across a $(n, c)$ grid with ten seeds per cell. All four methods agree on every one of 600 isomorphism verdicts, consistent with the completeness conjecture. On wall-clock time the Levi baselines dominate every tested cell by three to five orders of magnitude (geometric-mean ratio $311\times$ to $117{,}672\times$), which we report as measured. We contribute the representation framework, a conjecture of canonical completeness, and the first native-versus-Levi benchmark for hypergraph isomorphism.

---

### Breaking the Quality--Intelligibility Trade-off in Streaming Target Speaker Extraction via Deep-Feature-Anchored Preference Optimization

Authors: Shuhai Peng, Jinjiang Liu, Hui Lu, Liyang Chen, Guiping Zhong, Jiakui Li, Shiyin Kang, Zhiyong Wu
Published: 2026-07-11
Categories: cs.SD, cs.AI
arXiv: https://arxiv.org/abs/2607.10191v1

Generative streaming models for Target Speaker Extraction (TSE) commonly exhibit a quality--intelligibility trade-off, wherein naive optimization for perceptual audio quality tends to degrade speech intelligibility, and conversely. We reveal that this trade-off arises not from the constraints of streaming architectures, but from an inappropriate choice of optimization anchor. Directly optimizing against audio quality metrics induces catastrophic reward hacking, where content critical to pronunciation and intelligibility is systematically erased to maximize a proxy score. To break this bottleneck, we propose two complementary improvements: an enlarged Conformer convolution kernel for richer local spectro-temporal modeling, and WavLM-anchored Direct Preference Optimization (DPO) fine-tuning strategy. DPO preference pairs are ranked by WavLM cosine similarity, a deep acoustic feature encoding both phonetic structure and speaker identity, providing an optimization anchor that resists hacking. Under a 560 ms streaming chunk size, the proposed method achieves a 10.9% relative intelligibility improvement (word error rate: 0.138 to 0.123), with marginal simultaneous gains in audio quality and speaker similarity.

---

### PhysMRV: Physical Memory Retrieval and Verification for Physics Plausibility Reasoning

Authors: Wenyuan Wang, Lianyu Hu, Hao Wang, Yang Liu
Published: 2026-07-11
Categories: cs.LG, cs.AI, cs.CV
arXiv: https://arxiv.org/abs/2607.10190v1

Video-language models (VLMs) have achieved remarkable performance on video understanding and visual question answering, yet they remain unreliable in reasoning about physical plausibility, where understanding object interactions, causal dynamics, and fundamental physical principles is essential. This limitation is particularly evident on challenging physical reasoning benchmarks, revealing a persistent gap in physical commonsense reasoning. To address this challenge, we propose PhysMRV, a training-free physical memory and verification framework for physical plausibility reasoning. Unlike retrieval-augmented VLMs that retrieve semantically similar videos as additional context, PhysMRV transforms training videos into a Hierarchical Memory Bank of structured physical knowledge comprising three complementary levels: scene descriptions capturing visual context, physical-event graphs modeling object interactions and causal structure, and physics-rule summaries distilling reusable physical principles and cues. During inference, PhysMRV retrieves physically relevant memories and leverages their structured physical evidence to guide a frozen VLM in verifying physical plausibility, requiring neither fine-tuning nor parameter updates. We evaluate PhysMRV on three challenging physical reasoning benchmarks, ImplausiBench, IntPhys2, and GRASP Level 2, across multiple state-of-the-art VLMs. Experimental results demonstrate consistent improvements over direct prompting across diverse VLMs and evaluation benchmarks, showing that structured physical memories provide an effective and scalable means of enhancing physical plausibility reasoning without additional training.

---

### BiLoG-Net: A Bi-Context Location-Guided Network for Breast Mass Segmentation and Malignancy Classification in Mammography

Authors: Abu Fatema Mohammad Abdun Noor, Md Imam Ahasan, Md Samiul Ahasan, Kah Ong Michael Goh, S M Hasan Mahmud, Raihana Zannat
Published: 2026-07-11
Categories: cs.CV, cs.LG
arXiv: https://arxiv.org/abs/2607.10188v2

Breast cancer remains the most commonly diagnosed malignancy among women worldwide, yet accurate detection and characterization of breast masses in mammography remain challenging due to subtle intensity variations, heterogeneous tissue densities, and indistinct lesion boundaries that complicate radiological interpretation. To address these limitations, we propose BiLoG-Net, a deep learning framework that jointly performs breast mass segmentation and malignancy classification through bi-context location-aware feature modeling and segmentation-guided attention mechanisms. Our architecture integrates a novel encoder-decoder paradigm with Fire-based feature extraction, lightweight global and local feature enhancement modules, and adaptive location-aware gating to simultaneously capture long-range contextual dependencies and fine-grained boundary-sensitive details. Unlike conventional multi-stage pipelines, our tightly coupled multi-task design enables mutual reinforcement between pixel-level localization and image-level diagnosis, reducing error propagation while producing spatially grounded malignancy predictions. Evaluated on CBIS-DDSM and INBreast benchmarks, BiLoG-Net achieves state-of-the-art performance with Dice scores of 94.20% and 93.10%, classification accuracies of 95.20% and 93.60%, and AUC values of 97.10% and 96.00%, respectively, substantially outperforming existing CNN and transformer-based baselines. By combining precise boundary delineation with reliable malignancy assessment in a single end-to-end model, this work holds strong potential for clinical computer-aided detection systems, helping radiologists prioritize suspicious cases and improve screening efficiency in busy clinical settings.

---

### Knowledge-Conditioned, Single-Pass LLM Synthesis of Executable Unity Game Scenes: A Compiler Error Census across 26 Goal Playable Concepts

Authors: Hugh Xuechen Liu, Kıvanç Tatar
Published: 2026-07-11
Categories: cs.LG, cs.SE
arXiv: https://arxiv.org/abs/2607.10187v1

Large language models (LLMs) write Unity C\# for game scenes. Yet nearly all demonstrations rest on an iterative repair loop that regenerates code until it compiles, conflating what the model writes with what the loop fixes. We remove the loop and evaluate a single pass, where the first draft is final. This isolates the model's parametric knowledge, the most stringent test of unaided generation. Models instantiate Goal Playable Concepts, playable counterparts of goal patterns, across 10,400 generations (four open-weight models, 7B--30B; two generation modes; four intermediate-representation (IR) conditioning levels; 26 goal patterns; 20 seeds). None compiled into a runnable scene, leaving no survivorship bias. To understand how the generated C\# scripts fail, we categorize the 99 error codes behind 90{,}673 compiler-error occurrences as Grounding (invented or misused Unity types and APIs) or Hygiene (structural defects needing no Unity knowledge). The split differs sharply by goal pattern (e.g., Stealth fails mostly on invented engine references; Capture on plain C\# structure). Larger models, stricter IRs, and different generation modes move the errors but never yield a compiling scene. The bottleneck is missing engine-specific knowledge. The census orders goal patterns by that demand, showing designers where single-pass generation breaks.

---

### Automated Tensor Scheduling for Hybrid CPU-GPU LLM Inference on Consumer Devices

Authors: Yangyijian Liu, Hongyi Ye, Mingyang Li, Wu-jun Li
Published: 2026-07-11
Categories: cs.DC, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2607.10183v2

Running large language models on consumer devices such as laptops and desktops is challenging because model weights often exceed GPU memory capacity, making offloading inference necessary to extend effective model capacity with CPU memory. Existing offloading systems, however, typically rely on coarse layer-level or expert-level scheduling, which overlooks substantial heterogeneity among tensors within the same layer and adapts poorly to changing hardware load conditions on such devices. This paper presents ATSInfer, a hybrid CPU-GPU inference system for consumer devices that performs offloading at tensor granularity. ATSInfer combines static tensor placement with load-aware dynamic transfer, and introduces asynchronous CPU-GPU coordination to efficiently schedule hardware storage, data movement, and computation across heterogeneous backends. We implement ATSInfer and evaluate it on representative consumer platforms using both dense and MoE models. Compared with existing systems, ATSInfer improves prefill throughput by up to 1.94$\times$ and decode throughput by up to 3.29$\times$, while also increasing GPU utilization and making more effective use of PCIe bandwidth. These results show that ATSInfer can substantially improve the user experience of local LLM deployment on personal consumer devices.

---

### ActiveFly-Bench: Aligning Embodied Question Answering with Vision-Language-Action for Aerial Embodied Perception

Authors: Weichen Zhang, Shiquan Yu, Yinan Zhu, Peizhi Tang, Shilong Ji, Zhiyuan Deng, Tianyi Lyu, Haoyang Wang, Xin Zeng, Chen Gao, Yong Li, Xinlei Chen
Published: 2026-07-11
Categories: cs.RO, cs.AI
arXiv: https://arxiv.org/abs/2607.10180v2

We introduce ActiveFly-Bench, the first benchmark to bridge cyberspace reasoning and physical-world interaction for UAV embodied perception. The benchmark decomposes active perception into three hierarchical tasks: Aerial Embodied Question Answering (Air-EQA), Observation Behavior Planning (OBP), and Fine-grained Language-guided UAV Control (FLUC), explicitly connecting high-level task understanding, behavior planning, and low-level control. The datasets are collected from both real-world and simulated outdoor environments for training and evaluation. We further develop ActiveFly, a closed-loop UAV agent that integrates visual-language reasoning with fine-grained control, and deploy it on a physical UAV platform. Experiments with representative VLMs and VLA models show that current UAV agents still struggle with behavior planning, viewpoint adjustment, and robust task completion in active perception. These results establish ActiveFly-Bench as a new testbed for embodied aerial intelligence.

---

### ActiveFly-Bench: Aligning Embodied Question Answering with Vision-Language-Action for Aerial Embodied Perception

Authors: Weichen Zhang, Shiquan Yu, Yinan Zhu, Peizhi Tang, Shilong Ji, Zhiyuan Deng, Tianyi Lyu, Haoyang Wang, Xin Zeng, Chen Gao, Yong Li, Xinlei Chen
Published: 2026-07-11
Categories: cs.RO, cs.AI
arXiv: https://arxiv.org/abs/2607.10180v1

We introduce ActiveFly-Bench, the first benchmark to bridge cyberspace reasoning and physical-world interaction for UAV embodied perception. The benchmark decomposes active perception into three hierarchical tasks: Aerial Embodied Question Answering (Air-EQA), Observation Behavior Planning (OBP), and Fine-grained Language-guided UAV Control (FLUC), explicitly connecting high-level task understanding, behavior planning, and low-level control. The datasets are collected from both real-world and simulated outdoor environments for training and evaluation. We further develop ActiveFly, a closed-loop UAV agent that integrates visual-language reasoning with fine-grained control, and deploy it on a physical UAV platform. Experiments with representative VLMs and VLA models show that current UAV agents still struggle with behavior planning, viewpoint adjustment, and robust task completion in active perception. These results establish ActiveFly-Bench as a new testbed for embodied aerial intelligence.

---

### From Patent Expiry to Business Pathways: AI Workflows for Activating Innovation Archives

Authors: Sidney Shapiro, Mark Price
Published: 2026-07-11
Categories: cs.IR, cs.CL, cs.CY
arXiv: https://arxiv.org/abs/2607.10179v1

Patent databases represent one of the largest public archives of technical knowledge, yet much of this knowledge remains difficult to identify, interpret, and reuse once patent rights expire or lapse. This paper proposes an AI-enabled framework for discovering expired and lapsing patents, identifying technology trends, and translating patent disclosures into business pathways. We use pathways to mean structured commercialization routes such as SaaS products, services, licensing packages, consulting playbooks, training offerings, data products, or internal process tools. The framework treats patent expiry as both a business signal and an archival transition, not primarily as a legal problem. Legal status remains important, but it is one risk-screening input alongside customer need, implementation feasibility, channel access, and market timing. We describe a system architecture that combines patent metadata, maintenance-fee records, legal-status indicators, semantic search, patent-family analysis, market signals, and generative AI workflows. A proof of concept parses all 378 records in an official weekly CIPO ST.96 archive, identifies 20 expired, lapsed, or near-expiry candidates, tests the stability of the transparent scoring model, and uses a locally hosted Qwen3.6 model to populate structured review packets. The evaluation demonstrates reproducible ingestion, stable rankings under weight perturbation, and schema-conformant model output, while also exposing incomplete legal-status coverage and the need for register and expert review. We argue that AI can function as a discovery and translation layer for dormant technical knowledge, but that such systems must explicitly represent legal uncertainty, data limitations, and commercialization risk.

---

### On the Efficiency of LoRA Fine-Tuning for Vision-Language-Action Models in Industrial Robotic Manipulation

Authors: Finn Ferchau, Daniel Pommer, Cristian Axenie
Published: 2026-07-11
Categories: cs.RO, cs.LG
arXiv: https://arxiv.org/abs/2607.10172v1

Deploying billion-parameter Vision-Language-Action (VLA) models on industrial hardware requires fine-tuning to bridge the embodiment gap. Full Fine-Tuning (FFT) provides maximal plasticity but requires data centre-grade GPUs. We present a systematic study of Low-Rank Adaptation (LoRA) for $π_0$, a flow-matching VLA, evaluated on four precision assembly tasks with a UR5e robotic manipulator. Across a sweep of LoRA ranks (r=8 to 256), allocation strategies, and component-freezing ablations, we find no statistically significant advantage of FFT over certain LoRA configurations. Performance saturates at r=32, and uniform allocation across the Vision-Language-Model (VLM) backbone and action expert proves sufficient. Freezing the VLM or restricting the vision encoder to LoRA significantly degrades performance, indicating that embodiment adaptation requires both semantic and visual plasticity. These results suggest that LoRA at r=32 with full vision encoder fine-tuning is a practical approach, reducing static peak VRAM from 36.2 to 10.8 GiB (parameters and optimizer states, activation memory excluded) without detectable performance loss.

---

### Beyond Euclidean Clipping: Overcoming Exploration Collapse in LLM RL via Riemannian Isometric Policy Optimization

Authors: Zhicheng Cai, Xinyuan Guo, Hanlin Wu, Mingxuan Wang, Wei-Ying Ma, Ya-Qin Zhang, Hao Zhou
Published: 2026-07-11
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2607.10169v1

Reinforcement learning (RL) has become a dominant paradigm for enhancing LLMs' reasoning capabilities. However, RL algorithms with PPO-Clip are inherently limited by exploration collapse. Subsequent works remain primarily heuristic and fail to identify the essential cause of PPO-Clip's failure. This work reveals the fundamental flaw of PPO-Clip: it implicitly measures policy discrepancy using Euclidean metric, which is theoretically inconsistent with the intrinsic geometry on the policy Riemannian manifold. This geometric mismatch results in overly conservative updates in low-probability regions while aggressive in high-probability regions, ultimately collapsing exploration. To correct this geometric flaw, we propose Riemannian Isometric Policy Optimization (RIPO), which guarantees isometric policy updates on the Riemannian manifold, effectively balancing exploration and exploitation. We further show that RIPO achieves a favorable bias-variance trade-off, which stabilizes optimization. Extensive experiments demonstrate that RIPO significantly surpasses existing LLM RL algorithms across seven competition-level benchmarks (up to 60% improvement over GRPO on AIME24).

---

### Transcript-Free Lightweight Detection of Alzheimer's Disease from Spontaneous Speech Using Handcrafted MFCC-Dominant Acoustic Biomarkers

Authors: Rashin Gholijani Farahani, Azam Bastanfard
Published: 2026-07-11
Categories: cs.SD, cs.AI
arXiv: https://arxiv.org/abs/2607.10168v1

It is still hard to find Alzheimer's disease (AD) early, especially when neuroimaging is expensive or tools that depend on language are not available. Spontaneous speech provides a non-invasive signal; however, numerous current methodologies depend on transcripts/ASR or computationally intensive deep models. We offer a simple, audio-only baseline for detecting AD using 176 Cookie Theft recordings from the DementiaBank Pitt corpus (88 AD, 88 controls). WebRTC voice activity detection (VAD) is used to separate speech from non-speech. We take out 99 hand-crafted acoustic-temporal features, including pause and fluency statistics, spectral/prosodic descriptors, and MFCC summaries with Δ and ΔΔ. Evaluation is performed using a stringent speaker-independent GroupShuffleSplit,documenting performance across 30 iterations. A lightweight SVM with an RBF kernel gets an average AUC of 0.674 across runs. For example, a single split has an AUC of 0.742 and an accuracy of 0.657. We also present an exploratory compact-feature analysis utilizing a Top-20 subset ranked by Random Forest importance; since selection is not nested within training splits, these results may be overly optimistic and are not employed for primary conclusions (AUC 0.719). The results indicate that transcript-free spectro-temporal and fluency-related cues can facilitate speaker-independent Alzheimer's disease screening from raw audio, establishing a practical foundation for deployment-oriented research.

---

### EmoStyle: Affective Conditioning of Style-Specialist Experts for Emotional Image Generation

Authors: Dexiang Hong, Yijie Guo, Weidong Chen, Xinyan Liu, Zixuan Zou, Zhendong Mao, Yongdong Zhang
Published: 2026-07-11
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2607.10165v1

Emotion-aware artistic image generation requires an image to match the input prompt, follow the specified artistic style, and convey the target emotion. In this challenge, the main difficulty is that the visual and affective attributes available in the training data are not explicitly provided at test time. Without these attributes, the generator has to decide not only what to depict, but also how the target emotion should be expressed through color, lighting, brushwork, composition, line, and layout. This creates a control gap between the available test prompt and the fine-grained conditions needed for emotion-aware artistic generation. To bridge this gap, we propose EmoStyle, a Z-Image-based framework that converts the input prompt into a structured generation state. An LLM reasoner first predicts affective cues (valence-arousal, dominant emotion, and therapeutic-effect labels) and an aspect-ratio decision. Instead of using these predictions only as additional prompt text, we encode the affective fields into an affective condition vector and inject it into the denoising blocks through AdaLN-style modulation. This allows the inferred control variables to directly guide the generation of intermediate features. Since emotional expression is also style-dependent, we further train a dedicated LoRA adapter for each artistic style bucket and select the corresponding expert during inference, enabling the same affective cues to be rendered with bucket-specific priors for color, texture, brushwork, and composition. Finally, a lightweight VLM-guided candidate selection step ranks the generated images based on prompt alignment, style consistency, emotional expression, and visual quality. In Track 1 of the AffectiveArt Challenge 2026, our USTC\_PI\_LAB\_TEAM submission achieved first place.

---

### Hearing Like Humans? Sound Symbolism and Perceptual Alignment in Speech Language Models

Authors: Yun-Shao Tsai, Chun-Wei Chen, Chee-En Yu, Yi-Cheng Lin, Hung-yi Lee
Published: 2026-07-11
Categories: eess.AS, cs.CL
arXiv: https://arxiv.org/abs/2607.10162v1

Sound symbolism, the human tendency to map speech sounds to perceptual qualities such as roundness or sharpness, arises primarily from the acoustics of speech rather than spelling. Whether Speech Language Models (SLMs) share this tendency remains open, as prior evaluations rely on text or images rather than real speech. We study it using genuine human speech recordings, comparing model judgments against human data across the auditory, crossmodal, and visual components of the effect. We find that SLMs' auditory judgments align poorly with human perception and miss the acoustic cues, such as spectral tilt, that drive human intuitions, and open-weight models cannot reliably link a heard sound to its corresponding shape. With a visual-only control ruling out shape perception, the weakness localizes to how speech is represented, suggesting that perceptual alignment depends not on stronger vision but on speech representations that capture the cues humans hear.

---

### UNIT: Unleash Large Language Models Potential for Graph Continual Learning

Authors: Tairan Huang, Yili Wang, Beibei Hu, Yiting Shi, Qiutong Li, Changlong He, Jianliang Gao
Published: 2026-07-11
Categories: cs.AI
arXiv: https://arxiv.org/abs/2607.10159v1

In real-world multimodal web scenarios, graph-structured data often arrives in a streaming manner, making graph continual learning a crucial paradigm for continuously modeling such evolving structures. However, existing graph continual learning methods still face two fundamental challenges. 1) semantic-structural separation, where the graph-based methods excel at modeling topological relationships but neglect deep semantics. 2) imbalanced knowledge transfer, where existing models fail to effectively leverage general knowledge gained from early tasks to benefit subsequent new tasks. To address above issues, we propose a novel framework, \textbf{UN}leash Large Language Models PotentIal for Graph ConTinual Learning (UNIT). By fine-tuning large language model only on the first task, we bridge the distributional gap between the pre-trained LLM corpus and the target task dataset to enhance the adaptability of LLMs for graph-structured tasks. Meanwhile, we propose an uncertain-aware anchor generation mechanism to effectively preserve representative knowledge across tasks, avoiding the neglect of universal knowledge learned from previous tasks. Additionally, we introduce structural confluence modeling to explicitly integrates graph topology information into semantic information, enhancing the collaborative capabilities between semantic understanding and structural modeling. Extensive experiments demonstrate that our proposed method achieves state-of-the-art performance in the graph continual learning task.

---

### REVA-PO: Stabilizing Reinforcement Learning for Chest X-ray Report Generation

Authors: Li Guo, Anas M. Tahir, Z. Jane Wang
Published: 2026-07-11
Categories: cs.CV
arXiv: https://arxiv.org/abs/2607.10147v2

Automated chest X-ray report generation has recently benefited from reinforcement learning (RL) and large language models. However, RL training often suffers from instability or limited exploration due to fixed Kullback-Leibler (KL) regularization and a static reference policy that accumulates KL pressure over time. We propose Response-Weighted and Validation-Anchored Policy Optimization (REVA-PO), a RL framework that stabilizes long-term training via Response-Weighted Regularization (RER) and Validation-Anchored Policy Reset (VAPR). RER dynamically adjusts per-response KL weights based on advantage and reference-policy entropy, relaxing constraints for high-quality responses while tightening them for low-quality ones. Complementarily, VAPR periodically synchronizes the reference and current policies to the best validation checkpoint, resetting accumulated regularization pressure to expand the viable exploration space. To ensure a robust starting point, we employ a three-stage pipeline consisting of warm-up training, classifier-guided supervised fine-tuning, and RL. Extensive evaluations on MIMIC-CXR and IU-Xray demonstrate that REVA-PO sets new state-of-the-art benchmarks in both linguistic quality and clinical accuracy. Notably, BLEU-4 improves by 5.1% on MIMIC-CXR and 3.6% on IU-Xray, while CheXpert F1 and RadGraph F1 scores increase by 4.5% and 12.8%, respectively, over prior leading methods. The code is publicly available at https://github.com/LiGuo12/REVA_PO/.

---

### REVA-PO: Stabilizing Reinforcement Learning for Chest X-ray Report Generation

Authors: Li Guo, Anas M. Tahir, Z. Jane Wang
Published: 2026-07-11
Categories: cs.CV
arXiv: https://arxiv.org/abs/2607.10147v1

Automated chest X-ray report generation has recently benefited from reinforcement learning (RL) and large language models. However, RL training often suffers from instability or limited exploration due to fixed Kullback-Leibler (KL) regularization and a static reference policy that accumulates KL pressure over time. We propose Response-Weighted and Validation-Anchored Policy Optimization (REVA-PO), a RL framework that stabilizes long-term training via Response-Weighted Regularization (RER) and Validation-Anchored Policy Reset (VAPR). RER dynamically adjusts per-response KL weights based on advantage and reference-policy entropy, relaxing constraints for high-quality responses while tightening them for low-quality ones. Complementarily, VAPR periodically synchronizes the reference and current policies to the best validation checkpoint, resetting accumulated regularization pressure to expand the viable exploration space. To ensure a robust starting point, we employ a three-stage pipeline consisting of warm-up training, classifier-guided supervised fine-tuning, and RL. Extensive evaluations on MIMIC-CXR and IU-Xray demonstrate that REVA-PO sets new state-of-the-art benchmarks in both linguistic quality and clinical accuracy. Notably, BLEU-4 improves by 5.1% on MIMIC-CXR and 3.6% on IU-Xray, while CheXpert F1 and RadGraph F1 scores increase by 4.5% and 12.8%, respectively, over prior leading methods. The code is publicly available at https://github.com/LiGuo12/REVA_PO/.

---

### IdeaTrail: Full-Process Agent Trajectories for Scientific Ideation

Authors: Hengquan Guo
Published: 2026-07-11
Categories: cs.AI
arXiv: https://arxiv.org/abs/2607.10144v3

Scientific ideation unfolds over multiple stages, including literature search, paper reading, tool use, claim checking, cross-paper synthesis, brainstorming, rejection of weak directions, and iterative writing. Yet most existing resources capture isolated components or final artifacts rather than the process connecting them. We introduce IdeaTrail, a dataset of 1,170 multi-turn trajectories for scientific ideation and proposal generation. Each trajectory follows a research process from evidence gathering to either idea selection or proposal construction, jointly recording tool use, acquired evidence, intermediate artifacts, and reasoning. IdeaTrail is synthesized from human-selected research papers and proposal artifacts through a Generator--Advisor loop. The Generator produces the visible sequence of actions, observations, and artifact edits, while the Advisor uses the full generation context to check grounding, causal order, naturalness, and leakage from hidden targets. This reverse-to-forward design keeps trajectories aligned with real scientific artifacts while retaining the uncertainty, evidence use, and staged convergence characteristic of research practice. IdeaTrail provides both reusable process supervision and a general recipe for constructing scientific-research-agent data.

---

### FlowPainter: Inpainting Optical Flow via Confidence-Guided Completion

Authors: Yuang Meng, Chenyang Wu, Xianshun Liu, Chun-Le Guo, Zichen Liang, Lina Lei, Jie Liang, Hui Zeng, Chongyi Li, Lei Zhang
Published: 2026-07-11
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2607.10140v1

Existing optical flow methods broadly follow two paradigms: iterative optimization and diffusion-based estimation. Iterative methods, exemplified by RAFT, achieve high accuracy through recurrent refinement, but remain challenged by large displacements and complex motion. Diffusion-based methods introduce generative modeling and show promise in such ambiguous regions. However, existing diffusion models usually denoise the entire dense flow field from Gaussian noise, including simple regions where reliable motion can already be estimated by a lightweight network. This increases the denoising burden and may cause slow convergence and unstable training. To address this issue, we introduce FlowPainter, a diffusion-based optical flow framework that reformulates dense-flow generation as confidence-guided soft inpainting. FlowPainter employs a lightweight confidence-aware network to predict a rough flow and a pixel-wise confidence mask, distinguishing reliable simple regions from uncertain hard regions. The resulting simple-flow prior is used for confidence-based initialization and further injected into iterative denoising through confidence-gated residual guidance. With dynamically decaying guidance strength, FlowPainter stabilizes early denoising while preserving the flexibility of the diffusion model for late-stage detail refinement. Extensive experiments on public benchmarks, including Sintel, KITTI, and Spring, show that FlowPainter achieves strong accuracy under comparable training settings and converges more efficiently than existing diffusion-based optical flow methods, with notable gains on challenging benchmark splits. Our approach offers a practical way to integrate reliable discriminative priors with diffusion-based refinement for optical flow estimation. Our code is publicly available at https://github.com/mya012/FlowPainter.

---

### LLMs as a Jury: Cross-Model Consensus Can Outperform Process Reward Models for LLM Reasoning

Authors: Ning Liu
Published: 2026-07-11
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2607.10139v3

Selecting the correct answer from a pool of candidate reasoning chains is the engine of test-time scaling, yet the standard selectors each carry a cost: self-consistency inherits the errors of the single model it resamples, and trained reward models need labeled data and transfer poorly off-distribution. We study a third signal, free at inference time: cross-model consensus, the degree to which independently trained models, each solving the problem once, agree on a final answer. We treat the panel as an LLM-jury, in which the verification signal is the structure of agreement itself, with no model scoring another's work. Across seven benchmarks it selects correct answers better than self-consistency and far better than a model scoring its own candidates: on competition math it closes the entire gap to an oracle selector, while self-scoring closes almost none. The mechanism is error decorrelation: independently trained models err differently, so their wrong answers scatter while the correct one accumulates agreement. We make this precise with a parameter-free law, derived in closed form, that predicts consensus accuracy from three measured panel statistics to a mean absolute error of $0.03$ and exposes the method's ceiling: a shared-error floor where models share a misconception, near zero on math but non-trivial on science. Against four trained verifiers spanning discriminative, outcome, and generative reward models, the free LLM-jury matches the strongest inside their math training domain and is the top selector outside it. Cross-model consensus is thus a verifier we can characterize in advance: a law that says when to trust it, and a floor that marks where it cannot.

---

### LLMs as a Jury: Cross-Model Consensus Can Outperform Process Reward Models for LLM Reasoning

Authors: Ning Liu
Published: 2026-07-11
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2607.10139v2

Selecting the correct answer from a pool of candidate reasoning chains is the engine of test-time scaling, yet the standard selectors each carry a cost: self-consistency inherits the errors of the single model it resamples, and trained reward models need labeled data and transfer poorly off-distribution. We study a third signal, free at inference time: cross-model consensus, the degree to which independently trained models, each solving the problem once, agree on a final answer. We treat the panel as an LLM-jury, in which the structure of agreement, not any model's score of another, is the verification signal. Across seven benchmarks it selects correct answers better than self-consistency and far better than a model scoring its own candidates: on competition math it closes the entire gap to an oracle selector, while self-scoring closes almost none. The mechanism is error decorrelation: independently trained models err differently, so their wrong answers scatter while the correct one accumulates agreement. We make this precise with a parameter-free law, derived in closed form, that predicts consensus accuracy from three measured panel statistics to a mean absolute error of $0.03$ and exposes the method's ceiling: a shared-error floor where models share a misconception, near zero on math but non-trivial on science. Against four trained verifiers spanning discriminative, outcome, and generative reward models, the free LLM-jury matches the strongest inside their math training domain and is the top selector outside it. Cross-model consensus is thus a verifier we can characterize in advance: a law that says when to trust it, and a floor that marks where it cannot.

---

### RDQ: Residual Distribution Quantization for Large Language Models

Authors: Prateek Singh
Published: 2026-07-11
Categories: cs.LG
arXiv: https://arxiv.org/abs/2607.10137v3

Post-training quantization (PTQ) of large language models degrades sharply below 4-bit precision. We identify the root cause as residual stream distributional drift: quantization noise injected at each transformer layer accumulates in the shared residual representation, causing KL divergence from the FP16 baseline to grow super-linearly with depth (Pearson r=0.999 with log-perplexity, p<0.001, confirmed across all tested methods and bit-widths). We discover that 84% of LLaMA-3-8B layers exhibit non-Gaussian residual distributions (KS test, p<=0.05), and that per-layer residual stream variance grows 6,548x across depth. We propose RDQ (Residual Distribution Quantization), a PTQ framework whose central contribution is Cascaded Error Compensation (CEC): a sequential calibration procedure that captures the actual drifted activations each layer receives (computed by running calibration data through already-quantized upstream layers) and fits per-channel AWQ-style scales against those drifted inputs, with scales folded into preceding RMSNorm weights for exact mathematical equivalence at zero inference overhead. RDQ achieves state-of-the-art results on all three tested architectures: LLaMA-3-8B: 7.55 / 5.62 PPL (W3/W4); Qwen-2.5-7B: 7.46 / 6.38 PPL; Mistral-7B: 6.88 / 5.73 PPL. RDQ beats the best published baseline (LeanQuant/SpinQuant) at every model and bit-width combination, with gains up to -46.4% vs. RTN at W3A16 on LLaMA-3-8B. All output is standard group-128 asymmetric quantization, deployable on Qualcomm AIMET, GGUF, and any standard inference stack at zero runtime overhead.

---

### RDQ: Residual Distribution Quantization for Large Language Models

Authors: Prateek Singh
Published: 2026-07-11
Categories: cs.LG
arXiv: https://arxiv.org/abs/2607.10137v2

Post-training quantization (PTQ) of large language models degrades sharply below 4-bit precision. We identify the root cause as residual stream distributional drift: quantization noise injected at each transformer layer accumulates in the shared residual representation, causing KL divergence from the FP16 baseline to grow super-linearly with depth (Pearson r=0.999 with log-perplexity, p<0.001, confirmed across all tested methods and bit-widths). We discover that 84% of LLaMA-3-8B layers exhibit non-Gaussian residual distributions (KS test, p<=0.05), and that per-layer residual stream variance grows 6,548x across depth. We propose RDQ (Residual Distribution Quantization), a PTQ framework whose central contribution is Cascaded Error Compensation (CEC): a sequential calibration procedure that captures the actual drifted activations each layer receives (computed by running calibration data through already-quantized upstream layers) and fits per-channel AWQ-style scales against those drifted inputs, with scales folded into preceding RMSNorm weights for exact mathematical equivalence at zero inference overhead. RDQ achieves state-of-the-art results on all three tested architectures: LLaMA-3-8B: 7.55 / 5.62 PPL (W3/W4); Qwen-2.5-7B: 7.46 / 6.38 PPL; Mistral-7B: 6.88 / 5.73 PPL. RDQ beats the best published baseline (LeanQuant/SpinQuant) at every model and bit-width combination, with gains up to -46.4% vs. RTN at W3A16 on LLaMA-3-8B. All output is standard group-128 asymmetric quantization, deployable on Qualcomm AIMET, GGUF, and any standard inference stack at zero runtime overhead.

---

### LeRoPE: Learnable RoPE Frequencies Improve Language Modeling

Authors: Petros Karypis, Sean O'Brien, Shreyas Kadekodi, Rui Zhu, Julian McAuley
Published: 2026-07-11
Categories: cs.LG
arXiv: https://arxiv.org/abs/2607.10134v1

Rotary Positional Encodings (RoPE) are currently the most popular positional encodings used in modern language models. RoPE rotates two-dimensional chunks of query and key vectors, operating as a function of their relative positional offset. The position-wise rates of rotation in RoPE typically follow a geometric sequence specified by a fixed base-frequency hyperparameter. Prior work has improved performance by either increasing this parameter to slow rotation or by applying RoPE to only a subset of QK dimensions. In this work we modify RoPE by learning a scalar per frequency, treating frequencies as learnable parameters rather than hyperparameters. We validate Learned RoPE by training a ladder of language models from scratch, ranging from 52M to 2.5B parameters. We observe and analyze the emergence of a high-norm, positional LeRoPE band. LeRoPE consistently outperforms RoPE and partial RoPE across all scales, with RoPE requiring 3.4% more compute (FLOPs) to match LeRoPE at the largest scale.

---

### SALT-GNN: Handling Dense Neighborhoods in Anti-Money Laundering Graphs via Statistics-Aware Attention

Authors: Lidia Losavio, Francesco Sovrano, Dario Fenoglio, Martin Gjoreski, Marc Langheinrich
Published: 2026-07-11
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2607.10131v1

Money laundering threatens financial stability and exposes institutions to penalties, motivating automated detection. Because laundering schemes often emerge through relational patterns, graph neural networks (GNNs) are increasingly used for anti-money laundering (AML). Yet AML GNNs are typically evaluated with aggregate metrics such as overall F1 score, which hide an operational issue: high-activity recipient accounts concentrate many incoming transactions, making suspicious signals harder to isolate and costlier to investigate. We introduce a recipient-degree stratified evaluation that reports standard AML metrics across recipient-context density. Across three datasets (HI-Small, HI-Medium, and AMLSim-32k-5%), it reveals consistent degradation in dense recipient contexts, which we trace to three GNN characteristics: two known limitations that AML amplifies, i.e., (1) multiset non-discriminability and (2) cardinality blindness, and (3) an attention-specific effect: in dense neighborhoods, normalized attention attenuates weak but pattern-relevant multi-hop signals. Guided by this diagnosis, we propose SALT-GNN, a lightweight statistics-aware architecture that fuses degree-aware statistical aggregation with attention at each message-passing layer, so distributional and cardinality information shapes the node states used by subsequent attention steps. Ablations support fusion placement as a key factor in dense-context performance. On HI-Small and HI-Medium, SALT-GNN uses up to 77% fewer parameters than task-specific graph-transformer baselines while improving dense-context F1 score by 3-6 points; on AMLSim-32k-5%, it improves highest-degree F1 score by 16-20 points. The gains hold for both Transformer- and GAT-style attention, indicating that the benefit comes from where statistical and attentional evidence is fused rather than from a specific attention operator.

---

### TextGaze: Prompting Gaze Target Estimation with Textual Scene Cues

Authors: Junhui She, Fei Wang, Kun Li, Yiqi Nie, Yuxin Liu, Zhangling Duan, Xun Yang
Published: 2026-07-11
Categories: cs.CV
arXiv: https://arxiv.org/abs/2607.10130v1

Gaze target estimation aims to infer the position of a person's gaze within a scene. Within mainstream design logic, multi-branch methods require extra supervision and annotations, while streamlined designs prioritize low-level visual saliency over true gaze intent. The former leads to a high annotation burden and hinders domain transfer, whereas the latter causes misalignment between predicted attention and actual gaze targets. To address this issue, we propose TextGaze, a unified cross-modal architecture that leverages a Large Vision-Language Model (LVLM) as scalable semantic guidance to balance the two design paradigms. The model extracts visual features from a frozen encoder and utilizes an LVLM to obtain gaze-aligned textual cues. We design a transformer-based fusion module with hierarchical text supervision to preserve task semantics. Lightweight decoding heads enable the joint prediction of gaze heatmaps and in-/out-of-frame status. We evaluate our method on four mainstream datasets, and the results show competitive performance across key metrics with robust cross-dataset generalisation without extra fine-tuning. Overall, we provide a streamlined alternative to traditional designs and highlight the potential of LVLMs as accessible auxiliary guidance for gaze estimation.

---

### Energy-guided Recursive Model

Authors: Yifei Zhao, Ying Tang
Published: 2026-07-11
Categories: cs.LG, stat.ML
arXiv: https://arxiv.org/abs/2607.10128v3

Recursive models show promise on reasoning and language tasks, yet their test-time scaling lacks a principled criterion for selecting trajectories or determining recurrent depth. We introduce \textbf{Energy-guided Recursive Model (ERM)}, which uses Hopfield-type memories of valid local and global structures to assign intrinsic energies to candidate trajectories. These energies guide candidate selection and suggest an effective range of recurrent depths, implying that deeper recurrence does not necessarily improve reasoning accuracy. They also enable sampling methods such as parallel tempering to improve exploration. For reasoning tasks, ERM achieves optimal solutions on Sudoku ($98.97\%$), Pencil Puzzle Bench (PPBench, $88.04\%$) and Maze ($99.30\%$), reaching the best accuracy in recursive modeling. On language modeling, ERM reduces RedPajama-V2 perplexity by $1.74\%$ with marginal inference overhead. The results support energy guidance as a practical framework for improving test-time scaling in recursive models.

---

### Energy-guided Recursive Model

Authors: Yifei Zhao, Ying Tang
Published: 2026-07-11
Categories: cs.LG, stat.ML
arXiv: https://arxiv.org/abs/2607.10128v2

Recursive reasoning models address structured problems by repeatedly updating latent states of small neural networks. However, their test-time scaling lacks a principled inference mechanism: increasing depth or stochastic breadth generates more trajectories without a clear criterion for selection, and existing methods predominantly rely on additional q-heads or heuristic voting. Here, we develop the Energy-guided Recursive Model (ERM), which introduces an intrinsic selection principle based on explicit Hopfield energies. ERM leverages Hopfield-type memories of valid local or global structures to define the selector over candidate trajectories. The resulting energy seamlessly integrates with energy-based techniques such as parallel tempering to enhance sampling efficiency and ranking. With $D=64$ recurrent steps and $K=128$ candidates, ERM reaches optimal solutions on Sudoku ($98.97\%$), Pencil Puzzle Bench (PPBench, $88.04\%$) and Maze ($99.30\%$), improving upon recent Probabilistic Tiny Recursive Model and Equilibrium Reasoners. These results suggest that incorporating explicit energy functions into recursive reasoning offers a principled path toward more effective inference.

---

### GAE: Graph-Augmented Evolution for Scientific Discovery via Reinforcement Optimization

Authors: Xuanzhou Chen, Taoli Cheng
Published: 2026-07-11
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2607.10127v1

Evolutionary program search guided by Large Language Models (LLMs) has emerged as a powerful paradigm for automated scientific discovery. However, current approaches are fundamentally constrained by three bottlenecks: structurally blind parent selection, sparse whole-program evaluation rewards, and static mutation operators that fail to adapt during search. We present GAE (Graph-Augmented Evolution), a framework that resolves these limitations through a tightly coupled, three-pillar architecture. First, a relational graph neural network (GNN) parses programs into typed computation graphs, producing structure-aware embeddings. Second, an RL-optimized meta-controller leverages these embeddings to replace blind evolutionary sampling with a directed policy, dynamically selecting optimal parents and mutation directions based on reward history. Third, an online GRPO fine-tuning loop continuously updates the LLM mutation operator at test-time using group-normalized evaluation rewards, directly aligning the model's generation distribution with high-fitness structural edits. We evaluate GAE on a challenging scientific discovery task: symbolic regression for complex nonlinear oscillator systems. By transforming stochastic search into a directed, self-improving trajectory, GAE efficiently discovers closed-form physical equations, consistently matching or outperforming static LLM-driven baselines and achieving state-of-the-art out-of-distribution performance.

---

### ML in a Box: Analyzing Containerization Practices in Open Source ML Projects

Authors: Faten Jebari, Emna Ksontini, Amine Barrak, Wael Kessentini
Published: 2026-07-11
Categories: cs.SE, cs.AI
arXiv: https://arxiv.org/abs/2607.10126v1

Containerization has become increasingly essential in the machine learning (ML) domain, providing reproducibility, portability, and environment consistency. While prior studies have analyzed Dockerfile structures and best practices, none have examined ML projects in depth to reveal how the iterative nature of ML workflows influences container footprint, build performance, and caching behavior. We present the first large scale empirical study of 1,993 ML related Dockerfiles, combining quantitative analysis of container roles in ML projects and build dynamics with a qualitative investigation of refactoring practices. Results show that containers serve distinct roles across training, inference, and infrastructure. Containers are typically large, averaging 10.27 GB in size, and require long build times of about 8.84 minutes. We find that 44.4% of commits trigger rebuilds, primarily due to context file changes (96.4%), with experimentation being the main motive behind those commits that initiate rebuilds. Despite partial cache reuse, 71% of rebuild work is wasted on redundant computation. From stable projects, we identify 7 recurring ML-specific Dockerfile refactoring patterns that improve build efficiency and reduce container footprint.

---

### A Large-Scale Dataset of MCP Implementations on GitHub

Authors: Benny Toeppe, Amine Barrak, Emna Ksontini
Published: 2026-07-11
Categories: cs.SE, cs.AI
arXiv: https://arxiv.org/abs/2607.10123v1

The rapid emergence of the Model Context Protocol (MCP) has introduced a new standard for connecting large language models to external tools and services. Despite its rapid adoption in open-source development, systematic understanding of how MCP is implemented, structured, and maintained remains limited. This study presents the first large-scale, evidence-based dataset of real-world MCP implementation collected directly from GitHub. Using a hybrid pipeline that integrates the GitHub REST and GraphQL APIs with custom Python verification scripts, 3,238 candidate repositories were discovered, filtered, and validated through multi-stage evidence checks. Each verified project was classified by operational role (e.g., client, server, gateway) and exported in a reproducible JSONL schema. A manual review of a representative subset confirmed an overall precision of 83% at a 95% confidence level, and additionally revealed a set of repositories functioning primarily as educational samples, tutorials, or demonstration templates. A targeted exclusion rule was then applied to remove these non-operational repositories, resulting in a final dataset of 2,297 validated MCP projects. The analysis shows that Python and TypeScript dominate MCP development, with hybrid architectures emerging as the most common design pattern. By emphasizing transparent verification strategies, structured evidence tagging, and reproducible data organization, this work establishes a foundational benchmark for studying real-world MCP ecosystems and supports future research on integration, connectivity, and compatibility across the broader developer community.

---

### WeaveEarth: Structured Evidence Construction and Reasoning for Training-Free UHR Remote Sensing Understanding

Authors: Xianzhi Ma, Shujun Wang, Xiaohan Li, Hao Liu, Changhua Pei, Jianhui li
Published: 2026-07-11
Categories: cs.CV
arXiv: https://arxiv.org/abs/2607.10120v1

Ultra-High-Resolution (UHR) remote sensing image understanding requires Vision-Language Models (VLMs) to capture both the global scene layout and sparse yet task-critical local details under limited computational budgets. Existing methods mainly follow two paradigms. One is passive perception, which relies on resolution expansion or token compression and may therefore discard fine-grained details. The other is active perception, which depends on multi-round zooming and search, but suffers from high latency, contextual fragmentation, and error accumulation. We argue that a more effective path toward UHR understanding lies not in accessing more, but in organizing better. To this end, we propose WeaveEarth, a training-free framework that reformulates UHR understanding as a problem of structured evidence construction and reasoning under global context constraints. Specifically, WeaveEarth first employs Global-Aware Evidence Construction to select a compact, low-redundancy, and spatially complementary Minimal Support Evidence Set. It then introduces Structured Evidence Reasoning, which weaves local evidence, spatial metadata, and relative topology into a unified reasoning interface, thereby enhancing the VLM's ability to perform global-local joint reasoning. Extensive experiments show that WeaveEarth consistently outperforms strong baselines and existing UHR methods across multiple UHR remote sensing benchmarks and multiple frozen VLM backbones. Code is available at https://github.com/XianZhi-Ma/WeaveEarth.

---

### When Data Imbalance Helps: Robust Generalization Through Shortcut Saturation

Authors: Cheng-Ting Chou, Duc Binh Hoang
Published: 2026-07-11
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2607.10116v2

We study robust generalization under spurious correlations: tasks where a shortcut feature is correlated with the true label in training but anti-correlated in an adversarial held-out split. Varying the spurious ratio $r$ (the fraction of training examples where shortcut = true label) and model capacity, we find a counterintuitive result: data imbalance promotes generalization in sufficiently capable models. On a synthetic task where the true label is sum parity of an integer sequence and the shortcut is the parity of the maximum-valued element, a 2-layer, 2-head transformer generalized (reached $100\%$ adversarial accuracy) in 0% of seeds at $r{=}0.50$ but 77% of seeds at $r{=}0.90$. The effect is absent in 1-layer models, where imbalance instead traps the model on the shortcut. Through mechanistic analysis -- gradient conflict dynamics, circuit evolution, and QK/OV circuit ablations -- we characterize a mechanistic pathway consistent with imbalance promoting generalization.

---

### When Data Imbalance Helps: Robust Generalization Through Shortcut Saturation

Authors: Cheng-Ting Chou, Duc Binh Hoang
Published: 2026-07-11
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2607.10116v1

We study robust generalization under spurious correlations: tasks where a shortcut feature is correlated with the true label in training but anti-correlated in an adversarial held-out split. Varying the spurious ratio $r$ (the fraction of training examples where shortcut = true label) and model capacity, we find a counterintuitive result: data imbalance promotes generalization in sufficiently capable models. On a synthetic task where the true label is sum parity of an integer sequence and the shortcut is the parity of the maximum-valued element, a 2-layer, 2-head transformer generalized (reached $100\%$ adversarial accuracy) in 0% of seeds at $r{=}0.50$ but 77% of seeds at $r{=}0.90$. The effect is absent in 1-layer models, where imbalance instead traps the model on the shortcut. Through mechanistic analysis -- gradient conflict dynamics, circuit evolution, and QK/OV circuit ablations -- we characterize a mechanistic pathway consistent with imbalance promoting generalization.

---

### Cost of Reasoning in non-English Languages: A Case Study on Japanese

Authors: Yuu Jinnai
Published: 2026-07-11
Categories: cs.CL, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2607.10114v1

Reasoning Language Models (RLMs) achieve their strongest performance when they reason in English, the language for which reasoning-oriented training data is most abundant. However, reasoning trace is a clue for model interpretability and safety, and useful in practice for both the model users and for model developers. Thus, it is desirable to be able to develop a model that reasons in a language of the user's choice, while still maintaining strong reasoning performance. To this end, we study the feasibility of training a model that reasons in Japanese. We develop a Japanese-reasoning variant of Qwen-3-Swallow-8B, which is a Japanese LLM continually pretrained from Qwen-3-8B, with GRPO and evaluate it across coding, math, and science benchmarks. The study shows that reasoning-language control is feasible by training a Japanese continually pretrained model with GRPO. However, its performance is at best on par with strong English-reasoning baselines on several benchmarks. We also evaluate the trained model on Japanese cultural benchmarks and observe that the model's performance is worse than the baseline models, suggesting that the reasoning in Japanese does not immediately improve performance on culturally relevant tasks for free.

---

### Dynamic Agent Skills: A Lifecycle Survey and Taxonomy of Evolving Skill Libraries

Authors: Yubo Li
Published: 2026-07-11
Categories: cs.AI
arXiv: https://arxiv.org/abs/2607.10113v1

Large language model agents increasingly store reusable procedures outside the model. These reusable procedures are often called \emph{skills}: they may be code functions, natural-language instructions, SKILL.md packages, workflow graphs, or learned adapters that a future agent can retrieve and invoke. This taxonomy-driven survey asks how such skill libraries change over time. Across a $124$-paper $2023$--$2026$ audit set, we synthesize dynamic skill systems as \emph{lifecycle-managed, verified, evolving artifact stores}: agents collect evidence from interaction, propose skill updates, verify and admit candidates, organize them for retrieval and composition, repair or prune stale entries, and govern sharing through provenance and rollback. We organize the literature around three survey tools. First, a $\text{six}$-sense taxonomy distinguishes the structurally different artifacts called ``skills'' in current papers. Second, an $\text{eight}$-stage lifecycle architecture identifies the recurring design decisions behind evidence acquisition, proposal, verification/admission, storage, retrieval/composition, maintenance, distillation/portability, and governance. Third, a lightweight skill-record schema and $\text{ten}$-operator vocabulary provide common terms for comparing library updates without elevating them into a separate method contribution. Using this structure, we synthesize evidence-graded patterns with explicit caveats: admission and repair are repeatedly important, verifier quality materially affects skill-aware RL, flat retrieval can degrade as libraries grow, and current benchmarks still under-report library trajectories, usage--utility gaps, and safety surfaces. We close with concrete reporting standards and open problems for evaluating dynamic skills as changing libraries rather than static prompt or tool collections.

---

### Minionese: Comprehensive Benchmark and Mechanistic Study of Multilingual LLM Safety

Authors: Chigozirim Ifebi, Brent Kong, Ayushi Mehrotra
Published: 2026-07-11
Categories: cs.CR, cs.AI
arXiv: https://arxiv.org/abs/2607.10112v1

Safety alignment in large language models remains brittle across languages: prompts reliably refused in English can elicit harmful compliance in non-English and low-resource settings. We introduce \textsc{Minionese}, a multilingual jailbreak benchmark spanning 18 languages, 4 resource tiers, and 4 perturbation types (standard translation, code-switching, transliteration, and translationese), paired with a geometric mechanistic analysis of refusal failure across language tiers. We show that each attack type produces a distinct vulnerability profile: transliteration vulnerability is mediated by script identity, code-switching maintains effectiveness through the lowest-resource tier, and a sharp safety regime transition between Tiers 2 and 3 is consistent across all models. Mechanistically, low-resource jailbreaks succeed by routing harmful content through a geometrically misaligned subspace that projects insufficiently onto the refusal directions, leaving the refusal mechanism intact but untriggered. These findings show that English-only safety evaluations are insufficient; they require accounting for script family, perturbation type, and per-language alignment coverage. The benchmark and analysis code is at https://github.com/Brentkong/Minionese-Comprehensive-Benchmark-and-Mechanistic-Study-of-Multilingual-LLM-Safety.git.

---

### CHASE: Cache-Hole-Adapted Skip Exit for Looped State-Space Language Models

Authors: Zhenxuan Yu, Takeshi Kojima, Yutaka Matsuo, Yusuke Iwasawa
Published: 2026-07-11
Categories: cs.AI
arXiv: https://arxiv.org/abs/2607.10110v2

Recent work on looped language models suggests that many reasoning problems benefit from greater computational depth rather than from additional independent parameters. Existing studies, however, focus almost exclusively on Transformer backbones, leaving open whether this principle also applies to state-space language models. We investigate Looped Mamba and Looped Hybrid Mamba-Transformer architectures, which repeatedly apply a shared Mamba (or hybrid) block to introduce explicit finite-depth recurrent computation. On two controlled reasoning tasks-Mano (modular-arithmetic manipulation) and p-hop induction-Looped Mamba consistently outperforms parameter-matched non-looped baselines and, in several settings, matches or exceeds non-looped models of equal effective depth. We then extend the study to language model pre-training under matched iso-parameter and iso-FLOPs protocols, which jointly disentangle the effects of parameter sharing and effective depth: looped models remain competitive on downstream benchmarks with substantially fewer distinct parameters, although deeper non-looped models retain an advantage in validation perplexity under strict iso-FLOPs comparisons. Finally, we adapt Ouro's two-stage exit gate to Looped Mamba for threshold-controlled selection among recurrent-step outputs. Executing such exits on a state-space backbone, however, leaves the recurrent state without its deeper updates, and validation perplexity then degrades severely. We therefore introduce a cache-hole adaptation that aligns continued training with skipped-state inference. At the scales studied, the adapted model keeps perplexity close to full computation and matches or exceeds full-compute exit-state selection on downstream benchmarks while executing roughly half of the recurrent steps, which translates into measured inference speedups once the prefill is compute-bound.

---

### Looped State-Space Language Models with Adaptive Exit-State Selection

Authors: Zhenxuan Yu, Takeshi Kojima, Yutaka Matsuo, Yusuke Iwasawa
Published: 2026-07-11
Categories: cs.AI
arXiv: https://arxiv.org/abs/2607.10110v1

Recent work on looped language models suggests that many reasoning problems benefit from greater computational depth rather than from additional independent parameters. Existing studies, however, focus almost exclusively on Transformer backbones, leaving open whether this principle also applies to state-space language models. We investigate Looped Mamba and Looped Hybrid Mamba-Transformer architectures, which repeatedly apply a shared Mamba (or hybrid) block to introduce explicit finite-depth recurrent computation. On two controlled reasoning tasks-Mano (modular-arithmetic manipulation) and p-hop induction-Looped Mamba consistently outperforms parameter-matched non-looped baselines and, in several settings, matches or exceeds non-looped models of equal effective depth. We then extend the study to language model pre-training under matched iso-parameter and iso-FLOPs protocols, which jointly disentangle the effects of parameter sharing and effective depth: looped models remain competitive on downstream benchmarks with substantially fewer distinct parameters, although deeper non-looped models retain an advantage in validation perplexity under strict iso-FLOPs comparisons. Finally, we adapt Ouro's two-stage exit gate to Looped Mamba for threshold-controlled selection among recurrent-step outputs. Since all recurrent steps are still executed, the selected exit step represents prediction depth rather than reduced wall-clock computation. At the scales studied, adaptive exit-state selection improves downstream performance at intermediate depths, while actual inference-time savings require additional state-handling mechanisms.

---

### Adaptive Model Compression (AMC): Saliency-Driven Resource Allocation for Ultra-Low-Power Transformer Inference

Authors: Jiayin Hu, Kai Yuan, Vanessa Hu, Xuetao Yin, Jianhua Li, Sean Suchter
Published: 2026-07-11
Categories: cs.IR, cs.AI, cs.AR, cs.LG
arXiv: https://arxiv.org/abs/2607.10109v1

Deploying large-scale transformer models on resource-constrained edge devices remains a challenge due to the high energy and memory overhead inherent in static inference, which processes simple and complex tokens with uniform intensity. To address this, we propose Adaptive Model Compression (AMC), a saliency-driven framework that dynamically allocates hardware resources based on token importance. By implementing a multi-tier architecture, our system identifies critical high-saliency information for full-precision processing while aggressively reducing the rank and bit-width of less significant data. Experimental results demonstrate that AMC achieves a 59.2% reduction in system energy and a 2.24x increase in throughput on 45nm CMOS hardware. This approach effectively extends the battery life of mobile devices by utilizing high-definition compute only where necessary, maintaining robust performance with a marginal 3.6% accuracy trade-off.

---

### LLM Watermarking as Big Data Provenance: A Deployment-Oriented Systematization

Authors: Huy Phan, Kieu Dang, Ojaswi Dulal, Aiham AL Shukairi, Abby Shine, Chase Garner, Phung Lai
Published: 2026-07-11
Categories: cs.CR, cs.CL
arXiv: https://arxiv.org/abs/2607.10103v2

As large language models (LLMs) become widely deployed, their outputs can be copied, transformed, and redistributed at scale without reliable evidence of origin, creating risks for trust, accountability, intellectual property (IP) protection, and high-stakes decision-making. LLM watermarking addresses this problem by embedding detectable signals into text during or after generation. However, existing methods vary in design assumptions, threat models, and evaluation criteria, while deployment choices such as watermark placement, detection authority, and key management affect reliability, security, and scalability. This paper systematizes LLM watermarking as provenance infrastructure for large-scale data ecosystems. We organize existing approaches along four deployment dimensions: insertion point, verification authority, operational state, and transformation threat model, and relate them to the big data requirements of Volume, Velocity, Variety, Veracity, and Value. We further introduce a Big Data Watermarking Readiness framework centered on four deployment workloads: online generation, streaming detection, transformation pipelines, and ecosystem governance. The framework connects these workloads to system-level requirements including throughput, false-positive control, robustness, cross-domain reliability, governance, and downstream utility. Our analysis highlights a gap between benchmark performance and deployment readiness: false positives accumulate at scale, repeated transformations weaken watermark signals, computational overhead can limit online deployment, and centralized verification can create governance bottlenecks. We conclude with an evaluation blueprint and research directions for scalable, trustworthy provenance in big data ecosystems.

---

### A Survey on LLM Watermarking: Theory and Deployment

Authors: Huy Phan, Kieu Dang, Ojaswi Dulal, Aiham AL Shukairi, Abby Shine, Chase Garner, Phung Lai
Published: 2026-07-11
Categories: cs.CR, cs.CL
arXiv: https://arxiv.org/abs/2607.10103v1

Large language models (LLMs) are increasingly embedded in high-impact workflows, yet their ability to generate fluent text at scale has amplified risks of provenance ambiguity, model misuse, and large-scale content laundering. LLM watermarking, embedding invisible signatures into model outputs, has emerged as a promising technical layer for attribution, auditing, and downstream trust decisions. However, the literature has grown rapidly and unevenly: existing categorizations often mix orthogonal design choices, making it difficult to compare methods, reason about guarantees, or translate research results into deployable systems. This survey provides a systematic, deployment-oriented review of LLM watermarking. We organize the space by the core questions practitioners must answer: where a watermark is embedded (generation-time vs. training-time, token vs. representation), who can detect it (public vs. private detection authority), what is assumed (access to logits, sampling control, secret keys, model ownership), and which threat models are targeted (paraphrasing, translation, summarization, style transfer, token manipulation, and adaptive removal). We synthesize the main families of techniques-including sampling biasing, code-based schemes, representation- and training-based approaches-and analyze their security-utility trade-offs through the lens of detectability, robustness, and distribution shift. We further review attack and evasion strategies, evaluation protocols and metrics (false positive control, calibration, robustness curves), and open challenges such as cross-model transfer, multi-modal pipelines, collusion, and governance constraints. Finally, we provide practical guidance for selecting watermark designs under real operational requirements and identify research directions needed for reliable, accountable LLM deployment.

---

### DynaFilter: Cloud-driven Dynamic Filtering for Satellite Edge Intelligence

Authors: Ziyang Zhang, Jie Liu, Luca Mottola
Published: 2026-07-11
Categories: cs.CV, cs.LG
arXiv: https://arxiv.org/abs/2607.10098v1

Modern satellite edge systems, including those performing remote sensing tasks such object detection and tracking, are characterized by severely limited bandwidth and intermittent connections, making continuous data transmission to the cloud impractical. Existing edge-cloud systems, however, either require heavy pre-processing before analysis, for instance, full decompression of imagery data, or transmit all compressed data regardless of relevance. To address these challenges, we design DynaFilter, a dynamic filtering technique that enables satellite edge devices to perform selective region-of-interest (RoI) inference directly in the compressed-domain, without full decompression. Our key insight is that low-level compression syntax, specifically DC coefficients/AC energy in JPEG images and motion vectors in video streams, exhibits strong correlations with high-level semantic queries. By establishing a precise mapping between cloud query semantics and multimodal compressed-domain features, DynaFilter enables the edge to identify and transmit only relevant data associated to RoIs. Extensive evaluations show that DynaFilter reduces the total volume of pixel data for decoding and subsequent inference by 1.6x-7.1x for images, and achieves 92.0% bandwidth savings for video streams compared to state-of-the-art baselines. Furthermore, it decreases energy consumption by 43.1-88.6% on target devices and achieves a 1.6x-3.0x speedup in inference latency.

---

### LFD: Enabling Real-World Lensless Face Recognition with a Large-Scale Dataset

Authors: Junho Kim, Salman S. Khan, Sara Wan, Tomi Kuye, Ashok Veeraraghavan
Published: 2026-07-11
Categories: cs.CV, eess.IV
arXiv: https://arxiv.org/abs/2607.10094v1

Face recognition is a ubiquitously used computer vision task that has a wide range of applications ranging from everyday smartphone biometrics to high-stakes security systems. Most face recognition systems rely on traditional cameras, which often suffer from limitations such as bulky form factors, high costs, and limited privacy protection. To address these limitations, lensless cameras have emerged as an alternative. Lensless cameras use thin optical encoders, enabling smaller size, lower cost, and greater design flexibility. These cameras are typically paired with reconstruction algorithms that convert raw captures into recognizable images. However, reconstructed images often contain artifacts, and the reconstruction methods struggle to generalize well to real-world conditions. Furthermore, existing face datasets do not account for the artifacts present in lensless images. To address this issue, we introduce the Lensless Face Dataset (LFD). LFD comprises 21,080 lensless raw measurements, reconstructions, and standard images of faces captured under diverse lighting, angle, and distance. Our key contributions are: (1) Real-world lensless face data: LFD focuses on capturing a diverse face dataset with varying levels of artifacts introduced under different environments; (2) In-the-wild captures: 4,976 images are captured in outdoor settings with varying intensities of natural light and different background patterns; (3) Multiple lensless devices: LFD includes face images collected from three different types of lensless cameras, each with a unique optical encoder. We use this hardware diversity to demonstrate generalization across different lensless cameras. Through comprehensive evaluations and analysis, we show that LFD effectively captures shared features and artifacts across different lensless imaging devices, making it a valuable dataset for advancing lensless face recognition.

---

### EMBRACE: A Multi-task Framework for Comprehensive Quality Assessment in Cleavage-stage Embryo

Authors: Anwar Hussain Sofi, Jung-Hua Wang, Ming-Jer Chen, Tsung-Hsien Lee, Yu-Chiao Yi, Ming-Kuan Lin, Yi-Chung Lai
Published: 2026-07-11
Categories: cs.CV
arXiv: https://arxiv.org/abs/2607.10093v1

Cleavage-stage embryo assessment in in vitro fertilization requires the integrated interpretation of cytoplasmic fragmentation, developmental stage, and blastomere symmetry. However, conventional visual assessment is affected by observer variability, particularly when fragmented regions are small, irregular, or low contrast. This study presents EMBRACE, a multi-task deep learning framework for jointly performing cytoplasmic-fragmentation segmentation, t2/t4 developmental-stage classification, and blastomere-symmetry grading from static cleavage-stage embryo microscopy images. EMBRACE combines a shared ResNet-50 backbone, a concatenation-based multi-scale feature-fusion (C-MSFF) module, a U-Net-style segmentation decoder, and two task-specific classification heads. After predefined inclusion and exclusion criteria, 9,137 annotated embryo images were divided into 7,309 training, 914 validation, and 914 held-out test images. On the held-out test set, EMBRACE achieved a Dice coefficient of 0.781 and an intersection over union of 0.677 for fragmentation segmentation. Developmental-stage classification achieved an accuracy of 0.995, macro-F1 of 0.994, and AUC of 1.000. Blastomere-symmetry grading achieved a balanced accuracy of 0.901, macro-F1 of 0.907, and quadratic weighted kappa of 0.859. These findings support the feasibility of combining spatially inspectable fragmentation localization with embryo-level morphology assessment in a single framework. External and prospective validation is required before clinical deployment.

---

### Efficiently Adapting Spoken Language Models for the Singaporean Context

Authors: Ng Jia Sheng Jason
Published: 2026-07-11
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2607.10092v1

Spoken language models (SLMs) unify speech perception and reasoning, but adapting them to sensitive domains is underexplored, especially when the original training data is inaccessible and the use case demands multilingual, spoken-query interaction. We adapt an open-source SLM to the Singaporean Home Team context across five speech tasks in Singapore's four official languages, combining LoRA fine-tuning, a surrogate text-QA dataset that guards against catastrophic forgetting, and a multi-task objective that adapts the CoBa reweighting scheme to speech. We also build HTD-multilingual-QA, a 504,853 sample multilingual QA dataset in text and spoken form. The resulting HT-Moonstone (5B) matches or outperforms SLMs up to 7x its size on most tasks, attains the best accent and gender recognition among all models evaluated, and loses under 2\% of its original speech QA ability.

---