❌

Normal view

Received — 26 March 2026 ⏭

PhySe-RPO: Physics and Semantics Guided Relative Policy Optimization for Diffusion-Based Surgical Smoke Removal

✇cs.AI, q-bio.NC updates on arXiv.org
By: Zining Fang · Chunhui Liu · Bin Xu · Ming Chen · Xiaowei Hu · Cheng Xue
25 March 2026 at 12:00
arXiv:2603.22844v2 Announce Type: new Abstract: Surgical smoke severely degrades intraoperative video quality, obscuring anatomical structures and limiting surgical perception. Existing learning-based desmoking approaches rely on scarce paired supervision and deterministic restoration pipelines, making it difficult to perform exploration or reinforcement-driven refinement under real surgical conditions. We propose PhySe-RPO, a diffusion restoration framework optimized through Physics- and Semantics-Guided Relative Policy Optimization. The core idea is to transform deterministic restoration into a stochastic policy, enabling trajectory-level exploration and critic-free updates via group-relative optimization. A physics-guided reward imposes illumination and color consistency, while a visual-concept semantic reward learned from CLIP-based surgical concepts promotes smoke-free and anatomically coherent restoration. Together with a reference-free perceptual constraint, PhySe-RPO produces results that are physically consistent, semantically faithful, and clinically interpretable across synthetic and real robotic surgical datasets, providing a principled route to robust diffusion-based restoration under limited paired supervision.

Three Creates All: You Only Sample 3 Steps

✇cs.AI, q-bio.NC updates on arXiv.org
By: Yuren Cai · Guangyi Wang · Zongqing Li · Li Li · Zhihui Liu · Songzhi Su
25 March 2026 at 12:00
arXiv:2603.22375v1 Announce Type: cross Abstract: Diffusion models deliver high-fidelity generation but remain slow at inference time due to many sequential network evaluations. We find that standard timestep conditioning becomes a key bottleneck for few-step sampling. Motivated by layer-dependent denoising dynamics, we propose Multi-layer Time Embedding Optimization (MTEO), which freeze the pretrained diffusion backbone and distill a small set of step-wise, layer-wise time embeddings from reference trajectories. MTEO is plug-and-play with existing ODE solvers, adds no inference-time overhead, and trains only a tiny fraction of parameters. Extensive experiments across diverse datasets and backbones show state-of-the-art performance in the few-step sampling and substantially narrow the gap between distillation-based and lightweight methods. Code will be available.

Retrieval-Augmented Generation with Covariate Time Series

✇cs.AI, q-bio.NC updates on arXiv.org
By: Kenny Ye Liang · Zhongyi Pei · Huan Zhang · Yuhui Liu · Shaoxu Song · Jianmin Wang
25 March 2026 at 12:00
arXiv:2603.04951v2 Announce Type: replace Abstract: While RAG has greatly enhanced LLMs, extending this paradigm to Time-Series Foundation Models (TSFMs) remains a challenge. This is exemplified in the Predictive Maintenance of the Pressure Regulating and Shut-Off Valve (PRSOV), a high-stakes industrial scenario characterized by (1) data scarcity, (2) short transient sequences, and (3) covariate coupled dynamics. Unfortunately, existing time-series RAG approaches predominantly rely on generated static vector embeddings and learnable context augmenters, which may fail to distinguish similar regimes in such scarce, transient, and covariate coupled scenarios. To address these limitations, we propose RAG4CTS, a regime-aware, training-free RAG framework for Covariate Time-Series. Specifically, we construct a hierarchal time-series native knowledge base to enable lossless storage and physics-informed retrieval of raw historical regimes. We design a two-stage bi-weighted retrieval mechanism that aligns historical trends through point-wise and multivariate similarities. For context augmentation, we introduce an agent-driven strategy to dynamically optimize context in a self-supervised manner. Extensive experiments on PRSOV demonstrate that our framework significantly outperforms state-of-the-art baselines in prediction accuracy. The proposed system is deployed in Apache IoTDB within China Southern Airlines. Since deployment, our method has successfully identified one PRSOV fault in two months with zero false alarm.

Metaphor-based Jailbreak Attacks on Text-to-Image Models

✇cs.AI, q-bio.NC updates on arXiv.org
By: Chenyu Zhang · Lanjun Wang · Yiwen Ma · Wenhui Li · Yi Tu · An-An Liu
25 March 2026 at 12:00
arXiv:2512.10766v2 Announce Type: replace-cross Abstract: Text-to-image (T2I) models commonly incorporate defense mechanisms to prevent the generation of sensitive images. Unfortunately, recent jailbreak attacks have shown that adversarial prompts can effectively bypass these mechanisms and induce T2I models to produce sensitive content, revealing critical safety vulnerabilities. However, existing attack methods implicitly assume that the attacker knows the type of deployed defenses, which limits their effectiveness against unknown or diverse defense mechanisms. In this work, we reveal an underexplored vulnerability of T2I models to metaphor-based jailbreak attacks (MJA), which aims to attack diverse defense mechanisms without prior knowledge of their type by generating metaphor-based adversarial prompts. Specifically, MJA consists of two modules: an LLM-based multi-agent generation module (LMAG) and an adversarial prompt optimization module (APO). LMAG decomposes the generation of metaphor-based adversarial prompts into three subtasks: metaphor retrieval, context matching, and adversarial prompt generation. Subsequently, LMAG coordinates three LLM-based agents to generate diverse adversarial prompts by exploring various metaphors and contexts. To enhance attack efficiency, APO first trains a surrogate model to predict the attack results of adversarial prompts and then designs an acquisition strategy to adaptively identify optimal adversarial prompts. Extensive experiments on T2I models with various external and internal defense mechanisms demonstrate that MJA achieves stronger attack performance while using fewer queries, compared with six baseline methods. Additionally, we provide an in-depth vulnerability analysis suggesting that metaphor-based adversarial prompts evade safety mechanisms by inducing semantic ambiguity, while sensitive images arise from the model's probabilistic interpretation of concealed semantics.

Vision-DeepResearch: Incentivizing DeepResearch Capability in Multimodal Large Language Models

✇cs.AI, q-bio.NC updates on arXiv.org
By: Wenxuan Huang · Yu Zeng · Qiuchen Wang · Zhen Fang · Shaosheng Cao · Zheng Chu · Qingyu Yin · Shuang Chen · Zhenfei Yin · Lin Chen · Zehui Chen · Xu Tang · Yao Hu · Shaohui Lin · Philip Torr · Feng Zhao · Wanli Ouyang
25 March 2026 at 12:00
arXiv:2601.22060v3 Announce Type: replace-cross Abstract: Multimodal large language models (MLLMs) have achieved remarkable success across a broad range of vision tasks. However, constrained by the capacity of their internal world knowledge, prior work has proposed augmenting MLLMs by ``reasoning-then-tool-call'' for visual and textual search engines to obtain substantial gains on tasks requiring extensive factual information. However, these approaches typically define multimodal search in a naive setting, assuming that a single full-level or entity-level image query and few text query suffices to retrieve the key evidence needed to answer the question, which is unrealistic in real-world scenarios with substantial visual noise. Moreover, they are often limited in the reasoning depth and search breadth, making it difficult to solve complex questions that require aggregating evidence from diverse visual and textual sources. Building on this, we propose Vision-DeepResearch, which proposes one new multimodal deep-research paradigm, i.e., performs multi-turn, multi-entity and multi-scale visual and textual search to robustly hit real-world search engines under heavy noise. Our Vision-DeepResearch supports dozens of reasoning steps and hundreds of engine interactions, while internalizing deep-research capabilities into the MLLM via cold-start supervision and RL training, resulting in a strong end-to-end multimodal deep-research MLLM. It substantially outperforming existing multimodal deep-research MLLMs, and workflows built on strong closed-source foundation model such as GPT-5, Gemini-2.5-pro and Claude-4-Sonnet. The code will be released in https://github.com/Osilly/Vision-DeepResearch.

Tuning the sensitivity of mechanosensory receptors through histidine scanning

✇Cell
By: Yuanhao Wang · Yuhan Wang · Wenjie Yuan · Mingyu Fan · Xiaojing Wang · Anhui Wang · Yanling Bao · Yajing Zhang · Jia Chi Tan · Jianglai Wang · Junshuang Liu · Tianqi Huang · Zixuan Han · Biling Pei · Lijuan Chen · Zhengxu Ren · Xueqing Wang · Lanxin Hu · Siqi Wu · Mengke Pang · Sifan Wang · Zihuan Yang · Jiaze Li · Delian Huang · Shuai Shao · Huairui Yuan · Ling Wu · Yinnian Feng · Penghui Zhou · Guohui Li · Bo Sun · Chenqi Xu · Nicholas R.J. Gascoigne · Xiang Zhao
18 February 2026 at 08:00
Histidine scanning represents a broadly applicable technique for the identification of critical interaction sites within TCRs and other mechanosensory receptors to enhance receptor signaling strength and augment therapeutic efficacy via the catch bond mechanism.
Received — 17 March 2026 ⏭

From Text to Forecasts: Bridging Modality Gap with Temporal Evolution Semantic Space

✇cs.AI, q-bio.NC updates on arXiv.org
By: Lehui Li · Yuyao Wang · Jisheng Yan · Wei Zhang · Jinliang Deng · Haoliang Sun · Zhongyi Han · Yongshun Gong
16 March 2026 at 12:00
arXiv:2603.12664v1 Announce Type: cross Abstract: Incorporating textual information into time-series forecasting holds promise for addressing event-driven non-stationarity; however, a fundamental modality gap hinders effective fusion: textual descriptions express temporal impacts implicitly and qualitatively, whereas forecasting models rely on explicit and quantitative signals. Through controlled semi-synthetic experiments, we show that existing methods over-attend to redundant tokens and struggle to reliably translate textual semantics into usable numerical cues. To bridge this gap, we propose TESS, which introduces a Temporal Evolution Semantic Space as an intermediate bottleneck between modalities. This space consists of interpretable, numerically grounded temporal primitives (mean shift, volatility, shape, and lag) extracted from text by an LLM via structured prompting and filtered through confidence-aware gating. Experiments on four real-world datasets demonstrate up to a 29 percent reduction in forecasting error compared to state-of-the-art unimodal and multimodal baselines. The code will be released after acceptance.
Received — 14 March 2026 ⏭

Multi-Omics and Single-Cell Mendelian Randomization Reveal a Potential Role of VNN2 in Lung Adenocarcinoma in Resting Natural Killer Cells

✇Omics In Lung
By: Zhi Chun Xue · Jing Lin · Jian Hui Wu · Zhi Wen Peng · Mei Yan Tang · Peng Liang · Hui Ling Chen · Gui Ju Fang · Qing Xue
13 March 2026 at 18:00

World J Oncol. 2026 Mar 5;17(2):247-255. doi: 10.14740/wjon2689. eCollection 2026 Apr.

ABSTRACT

BACKGROUND: We aimed to evaluate the potential association between genetically predicted vanin-2 (VNN2) expression and lung adenocarcinoma (LUAD) risk, and to explore the immune cell subtype that may underlie this relationship.

METHODS: We integrated whole-blood expression quantitative trait loci (eQTL) data from eQTLGen, plasma protein quantitative trait loci (pQTL) data from deCODE, and LUAD genome-wide association study (GWAS) data from European-ancestry cohorts, together with differential expression analysis using GEPIA2, to identify candidate genes for subsequent single-cell eQTL (sc-eQTL) Mendelian randomization (MR) analysis. For the sc-eQTL analysis, VNN2-associated eQTLs from 14 immune cell types profiled in the OneK1K single-cell eQTL resource were tested for associations with LUAD risk.

RESULTS: Bulk-level MR analysis showed that genetically predicted increases in VNN2 expression and protein levels were significantly associated with a reduced risk of LUAD (eQTL-MR: odds ratio (OR) = 0.964, 95% confidence interval (95% CI), 0.934-0.995; P = 0.024; pQTL-MR: OR = 0.946, 95% CI, 0.921-0.970; P = 2.87 × 10-5). Transcriptomic analyses confirmed significant downregulation of VNN2 in LUAD tumors compared with normal lung tissues. sc-eQTL MR identified the strongest association in resting natural killer (rNK) cells (OR = 0.896, 95% CI, 0.829-0.967; P = 0.005).

CONCLUSIONS: Multi-omics and sc-eQTL MR analyses indicated that genetically predicted increases in VNN2 expression were associated with a reduced risk of LUAD, with the most pronounced effect observed in rNK cells. These findings suggest a potential cell type-specific role of VNN2 in LUAD susceptibility and warrant further studies to validate its biological relevance and clinical implications.

PMID:41822323 | PMC:PMC12978397 | DOI:10.14740/wjon2689

Profiling of the mycobiome and metabolome: a comparative study of benign pulmonary nodules and lung adenocarcinoma

✇Pulmonary nodule
By: Weiyi Zhang · Yuan Dong · Fangyan Chen · Fang Wang · Jiahui Li · Chunxi Liu · Tianyi Liu · Li Han · Xiaodong Jia
11 March 2026 at 18:00

Front Cell Infect Microbiol. 2026 Feb 23;16:1732958. doi: 10.3389/fcimb.2026.1732958. eCollection 2026.

ABSTRACT

INTRODUCTION: Lung adenocarcinoma (LUAD), the most common subtype of non-small cell lung cancer, is a form of malignant pulmonary nodule that requires clinical differentiation from benign pulmonary nodules (BPN). The mechanisms underlying the development of LUAD are complex, and effective non-invasive methods for differentiating BPN from LUAD are lacking. This study aimed not only to distinguish BPN from LUAD using gut fungi and serum metabolites, but also to establish an integrated network of gut fungi-metabolite-cytokine interactions.

METHODS: Fecal and serum samples from individuals with BPN and patients with LUAD were subjected to internal transcribed spacer sequencing, ultra-performance liquid chromatography-tandem mass spectrometry, and multiplex Luminex assays to quantify gut fungi, metabolites, and cytokines, respectively.

RESULTS: A significant difference in gut fungal communities was observed between the BPN and LUAD groups. Multiple genera and species were more abundant in LUAD than in BPN. Docosapentaenoic acid n-6 (DPAn-6), indole-3-propionic acid (IPA), and interferon-γ-induced protein 10 (IP-10) were significantly elevated in the LUAD group. The integrated model established using a combination of gut fungi and metabolites demonstrated excellent performance in distinguishing BPN from LUAD. A network of interactions was established among differentially abundant gut fungi, serum metabolites, and cytokines.

CONCLUSION: Our study identifies a novel panel of fungal and metabolite biomarkers for differentiating between BPN and LUAD, and constructs a multi-omics network that provides new insights into investigating the mechanistic role of gut mycobiota dysbiosis in LUAD.

PMID:41809995 | PMC:PMC12968269 | DOI:10.3389/fcimb.2026.1732958

Facile induction of immune tolerance by an interleukin-2–TGFβ surrogate agonist

✇Nature - Issue - nature.com science feeds
By: Qinli Sun · Alison K. Barrett · Masato Ogishi · Huiyun Lyu · Hua Jiang · Honghui Liu · Yang Zhao · Grayson E. Rodriguez · Pingdong Tao · Matthias Obenaus · Karsten D. Householder · Qizhi Tang · Tobias V. Lanz · K. Christopher Garcia
11 March 2026 at 08:00

Nature, Published online: 11 March 2026; doi:10.1038/s41586-026-10208-0

A fusion protein designed to comprise IL-2 and a helminth-derived TGFβ mimic activates IL-2 and TGFβ signalling pathways in IL-2 receptor-expressing T cells and induces stable antigen-specific regulatory T cells in peripheral lymphoid organs.

Risk-adaptive therapy guided by dynamic ctDNA in nasopharyngeal carcinoma

✇Nature - Issue - nature.com science feeds
By: Jiawei Lv · Dan-Xue Zheng · Jin-Hui Liang · Ning Zhang · Zu-Lu Ye · Xu-Dong Xu · Melvin. L. K. Chua · Lu-Lu Zhang · Zi-Ming Du · Zi-Chen Zhang · Wen-Fei Li · Ling-Long Tang · Lei Chen · Yan-Ping Mao · Rui Guo · Yu-Pei Chen · Li Lin · Yuan Zhang · Xu Liu · Cheng Xu · Zhi-Xuan Li · Ling-Xin Xu · Pan-Yang Yang · Kun Chen · Deng Bin · Tian-Sheng Gao · Jian-Ye Yan · Lu-Si Chen · Shao Hui Huang · Hong-Yun Zhao · Shu-Bin Hong · Yu-Sheng Jie · Hui-Ling Huang · Xu-Hua Tang · Jing-Ping Yun · Li-Zhi Liu · Li Tian · Hao-Jiang Li · Ji-Bin Li · Guan-Qun Zhou · Jun Ma · Ying Sun
11 March 2026 at 08:00

Nature, Published online: 11 March 2026; doi:10.1038/s41586-026-10244-w

A clinical trial testing whether monitoring ctDNA clearance during treatment for nasopharyngeal cancer could be used to inform decisions about an individual’s subsequent therapeutic programme shows promising results.
Received — 11 March 2026 ⏭

Breast Cancer Screening Knowledge and Sentiments in Singaporean Women: Mixed Methods Study Using Topic Modeling, Sentiment Analysis, and Structured Questionnaire Data

✇Journal of Medical Internet Research
By: Peh Joo Ho · Zi Lin Lim · Jenny Liu · Nur Khaliesah Mohamed Riza · Ying Jia Chew · Yi Ying Lim · Hui Ling Tan · Su-Ann Goh · Han Boon Oh · Chi Hui Chin · Sing Cheer Kwek · Zhi Peng Zhang · Desmond Luan Seng Ong · Swee Tian Quek · Sujith Wijerathne · Philip Tsau Choong Iau · Mikael Hartman · Jingmei Li
11 March 2026 at 04:15
Background: Mammography screening uptake in Singapore remains below 40% despite campaigns and subsidies. Natural language processing (NLP) can extract nuanced attitudes from free text that fixed response options miss, revealing latent factors influencing breast cancer (BC) screening behavior. Objective: This study characterized women’s attitudes toward mammography using mixed methods data, examined associations between BC awareness and screening willingness, and identified barriers and facilitators through NLP of free-text responses. Methods: We conducted a cross-sectional study within the multicenter cohort in Singapore (October 2021-December 2023). In total, 4169 women aged 35‐59 years (median 48, IQR 43‐54) were recruited via convenience sampling (3 hospitals and 2 polyclinics). Participants completed online structured questionnaires on demographics and screening history, then a BC education quiz with feedback. Participants answering >80% correctly were classified as “BC-aware.” Posteducation, participants reported screening willingness (motivated or neutral) with optional free-text explanations. Logistic regression models (adjusted for study site, age, ethnicity, marital status, housing, and education) examined the associations with willingness. For 3819 English-language respondents, biterm topic modeling identified themes and sentiment analysis quantified emotional tone. Statistical significance: =.05. Results: Overall, 79% (3287/4169) were BC-aware, and 94% (3908/4169) reported increased motivation posteducation. BC-aware women had higher screening motivation than BC-unaware women (adjusted odds ratio [aOR] 2.88, 95% CI 2.19‐3.80;

Agentic Critical Training

✇cs.AI, q-bio.NC updates on arXiv.org
By: Weize Liu · Minghui Liu · Sy-Tuyen Ho · Souradip Chakraborty · Xiyao Wang · Furong Huang
10 March 2026 at 12:00
arXiv:2603.08706v1 Announce Type: new Abstract: Training large language models (LLMs) as autonomous agents often begins with imitation learning, but it only teaches agents what to do without understanding why: agents never contrast successful actions against suboptimal alternatives and thus lack awareness of action quality. Recent approaches attempt to address this by introducing self-reflection supervision derived from contrasts between expert and alternative actions. However, the training paradigm fundamentally remains imitation learning: the model imitates pre-constructed reflection text rather than learning to reason autonomously. We propose Agentic Critical Training (ACT), a reinforcement learning paradigm that trains agents to identify the better action among alternatives. By rewarding whether the model's judgment is correct, ACT drives the model to autonomously develop reasoning about action quality, producing genuine self-reflection rather than imitating it. Across three challenging agent benchmarks, ACT consistently improves agent performance when combined with different post-training methods. It achieves an average improvement of 5.07 points over imitation learning and 4.62 points over reinforcement learning. Compared to approaches that inject reflection capability through knowledge distillation, ACT also demonstrates clear advantages, yielding an average improvement of 2.42 points. Moreover, ACT enables strong out-of-distribution generalization on agentic benchmarks and improves performance on general reasoning benchmarks without any reasoning-specific training data, highlighting the value of our method. These results suggest that ACT is a promising path toward developing more reflective and capable LLM agents.

GraphSkill: Documentation-Guided Hierarchical Retrieval-Augmented Coding for Complex Graph Reasoning

✇cs.AI, q-bio.NC updates on arXiv.org
By: Fali Wang · Chenglin Weng · Xianren Zhang · Siyuan Hong · Hui Liu · Suhang Wang
10 March 2026 at 12:00
arXiv:2603.06620v1 Announce Type: cross Abstract: The growing demand for automated graph algorithm reasoning has attracted increasing attention in the large language model (LLM) community. Recent LLM-based graph reasoning methods typically decouple task descriptions from graph data, generate executable code augmented by retrieval from technical documentation, and refine the code through debugging. However, we identify two key limitations in existing approaches: (i) they treat technical documentation as flat text collections and ignore its hierarchical structure, leading to noisy retrieval that degrades code generation quality; and (ii) their debugging mechanisms focus primarily on runtime errors, yet ignore more critical logical errors. To address them, we propose {\method}, an \textit{agentic hierarchical retrieval-augmented coding framework} that exploits the document hierarchy through top-down traversal and early pruning, together with a \textit{self-debugging coding agent} that iteratively refines code using automatically generated small-scale test cases. To enable comprehensive evaluation of complex graph reasoning, we introduce a new dataset, {\dataset}, covering small-scale, large-scale, and composite graph reasoning tasks. Extensive experiments demonstrate that our method achieves higher task accuracy and lower inference cost compared to baselines\footnote{The code is available at \href{https://github.com/FairyFali/GraphSkill}{\textcolor{blue}{https://github.com/FairyFali/GraphSkill}}.}.

MetaWorld-X: Hierarchical World Modeling via VLM-Orchestrated Experts for Humanoid Loco-Manipulation

✇cs.AI, q-bio.NC updates on arXiv.org
By: Yutong Shen · Hangxu Liu · Penghui Liu · Jiashuo Luo · Yongkang Zhang · Rex Morvley · Chen Jiang · Jianwei Zhang · Lei Zhang
10 March 2026 at 12:00
arXiv:2603.08572v1 Announce Type: cross Abstract: Learning natural, stable, and compositionally generalizable whole-body control policies for humanoid robots performing simultaneous locomotion and manipulation (loco-manipulation) remains a fundamental challenge in robotics. Existing reinforcement learning approaches typically rely on a single monolithic policy to acquire multiple skills, which often leads to cross-skill gradient interference and motion pattern conflicts in high-degree-of-freedom systems. As a result, generated behaviors frequently exhibit unnatural movements, limited stability, and poor generalization to complex task compositions. To address these limitations, we propose MetaWorld-X, a hierarchical world model framework for humanoid control. Guided by a divide-and-conquer principle, our method decomposes complex control problems into a set of specialized expert policies (Specialized Expert Policies, SEP). Each expert is trained under human motion priors through imitation-constrained reinforcement learning, introducing biomechanically consistent inductive biases that ensure natural and physically plausible motion generation. Building upon this foundation, we further develop an Intelligent Routing Mechanism (IRM) supervised by a Vision-Language Model (VLM), enabling semantic-driven expert composition. The VLM-guided router dynamically integrates expert policies according to high-level task semantics, facilitating compositional generalization and adaptive execution in multi-stage loco-manipulation tasks.

Membership Inference Attacks on Tokenizers of Large Language Models

✇cs.AI, q-bio.NC updates on arXiv.org
By: Meng Tong · Yuntao Du · Kejiang Chen · Weiming Zhang · Ninghui Li
10 March 2026 at 12:00
arXiv:2510.05699v3 Announce Type: replace-cross Abstract: Membership inference attacks (MIAs) are widely used to assess the privacy risks associated with machine learning models. However, when these attacks are applied to pre-trained large language models (LLMs), they encounter significant challenges, including mislabeled samples, distribution shifts, and discrepancies in model size between experimental and real-world settings. To address these limitations, we introduce tokenizers as a new attack vector for membership inference. Specifically, a tokenizer converts raw text into tokens for LLMs. Unlike full models, tokenizers can be efficiently trained from scratch, thereby avoiding the aforementioned challenges. In addition, the tokenizer's training data is typically representative of the data used to pre-train LLMs. Despite these advantages, the potential of tokenizers as an attack vector remains unexplored. To this end, we present the first study on membership leakage through tokenizers and explore five attack methods to infer dataset membership. Extensive experiments on millions of Internet samples reveal the vulnerabilities in the tokenizers of state-of-the-art LLMs. To mitigate this emerging risk, we further propose an adaptive defense. Our findings highlight tokenizers as an overlooked yet critical privacy threat, underscoring the urgent need for privacy-preserving mechanisms specifically designed for them.
Received — 5 March 2026 ⏭

Unbiased Dynamic Pruning for Efficient Group-Based Policy Optimization

✇cs.AI, q-bio.NC updates on arXiv.org
By: Haodong Zhu · Yangyang Ren · Yanjing Li · Mingbao Lin · Linlin Yang · Xuhui Liu · Xiantong Zhen · Haiguang Liu · Baochang Zhang
5 March 2026 at 13:00
arXiv:2603.04135v1 Announce Type: cross Abstract: Group Relative Policy Optimization (GRPO) effectively scales LLM reasoning but incurs prohibitive computational costs due to its extensive group-based sampling requirement. While recent selective data utilization methods can mitigate this overhead, they could induce estimation bias by altering the underlying sampling distribution, compromising theoretical rigor and convergence behavior. To address this limitation, we propose Dynamic Pruning Policy Optimization (DPPO), a framework that enables dynamic pruning while preserving unbiased gradient estimation through importance sampling-based correction. By incorporating mathematically derived rescaling factors, DPPO significantly accelerates GRPO training without altering the optimization objective of the full-batch baseline. Furthermore, to mitigate the data sparsity induced by pruning, we introduce Dense Prompt Packing, a window-based greedy strategy that maximizes valid token density and hardware utilization. Extensive experiments demonstrate that DPPO consistently accelerates training across diverse models and benchmarks. For instance, on Qwen3-4B trained on MATH, DPPO achieves 2.37$\times$ training speedup and outperforms GRPO by 3.36% in average accuracy across six mathematical reasoning benchmarks.

Quantum-Inspired Fine-Tuning for Few-Shot AIGC Detection via Phase-Structured Reparameterization

✇cs.AI, q-bio.NC updates on arXiv.org
By: Kaiyang Xing · Han Fang · Zhaoyun Chen · Zhonghui Li · Yang Yang · Weiming Zhang · Guoping Guo
4 March 2026 at 13:00
arXiv:2603.02281v1 Announce Type: cross Abstract: Recent studies show that quantum neural networks (QNNs) generalize well in few-shot regimes. To extend this advantage to large-scale tasks, we propose Q-LoRA, a quantum-enhanced fine-tuning scheme that integrates lightweight QNNs into the low-rank adaptation (LoRA) adapter. Applied to AI-generated content (AIGC) detection, Q-LoRA consistently outperforms standard LoRA under few-shot settings. We analyze the source of this improvement and identify two possible structural inductive biases from QNNs: (i) phase-aware representations, which encode richer information across orthogonal amplitude-phase components, and (ii) norm-constrained transformations, which stabilize optimization via inherent orthogonality. However, Q-LoRA incurs non-trivial overhead due to quantum simulation. Motivated by our analysis, we further introduce H-LoRA, a fully classical variant that applies the Hilbert transform within the LoRA adapter to retain similar phase structure and constraints. Experiments on few-shot AIGC detection show that both Q-LoRA and H-LoRA outperform standard LoRA by over 5% accuracy, with H-LoRA achieving comparable accuracy at significantly lower cost in this task.
❌