❌

Normal view

Received — 2 April 2026 ⏭

Predicting Neuromodulation Outcome for Parkinson's Disease with Generative Virtual Brain Model

✇cs.AI, q-bio.NC updates on arXiv.org
By: Siyuan Du · Siyi Li · Shuwei Bai · Ang Li · Haolin Li · Mingqing Xiao · Yang Pan · Dongsheng Li · Weidi Xie · Yanfeng Wang · Ya Zhang · Chencheng Zhang · Jiangchao Yao
1 April 2026 at 12:00
arXiv:2603.29176v1 Announce Type: new Abstract: Parkinson's disease (PD) affects over ten million people worldwide. Although temporal interference (TI) and deep brain stimulation (DBS) are promising therapies, inter-individual variability limits empirical treatment selection, increasing non-negligible surgical risk and cost. Previous explorations either resort to limited statistical biomarkers that are insufficient to characterize variability, or employ AI-driven methods which is prone to overfitting and opacity. We bridge this gap with a pretraining-finetuning framework to predict outcomes directly from resting-state fMRI. Critically, a generative virtual brain foundation model, pretrained on a collective dataset (2707 subjects, 5621 sessions) to capture universal disorder patterns, was finetuned on PD cohorts receiving TI (n=51) or DBS (n=55) to yield individualized virtual brains with high fidelity to empirical functional connectivity (r=0.935). By constructing counterfactual estimations between pathological and healthy neural states within these personalized models, we predicted clinical responses (TI: AUPR=0.853; DBS: AUPR=0.915), substantially outperforming baselines. External and prospective validations (n=14, n=11) highlight the feasibility of clinical translation. Moreover, our framework provides state-dependent regional patterns linked to response, offering hypothesis-generating mechanistic insights.

AGFT: Alignment-Guided Fine-Tuning for Zero-Shot Adversarial Robustness of Vision-Language Models

✇cs.AI, q-bio.NC updates on arXiv.org
By: Yubo Cui · Xianchao Guan · Zijun Xiong · Zheng Zhang
1 April 2026 at 12:00
arXiv:2603.29410v1 Announce Type: cross Abstract: Pre-trained vision-language models (VLMs) exhibit strong zero-shot generalization but remain vulnerable to adversarial perturbations. Existing classification-guided adversarial fine-tuning methods often disrupt pre-trained cross-modal alignment, weakening visual-textual correspondence and degrading zero-shot performance. In this paper, we propose an Alignment-Guided Fine-Tuning (AGFT) framework that enhances zero-shot adversarial robustness while preserving the cross-modal semantic structure. Unlike label-based methods that rely on hard labels and fail to maintain the relative relationships between image and text, AGFT leverages the probabilistic predictions of the original model for text-guided adversarial training, which aligns adversarial visual features with textual embeddings via soft alignment distributions, improving zero-shot adversarial robustness. To address structural discrepancies introduced by fine-tuning, we introduce a distribution consistency calibration mechanism that adjusts the robust model output to match a temperature-scaled version of the pre-trained model predictions. Extensive experiments across multiple zero-shot benchmarks demonstrate that AGFT outperforms state-of-the-art methods while significantly improving zero-shot adversarial robustness.

LaSM: Layer-wise Scaling Mechanism for Defending Pop-up Attack on GUI Agents

✇cs.AI, q-bio.NC updates on arXiv.org
By: Zihe Yan · Zhuosheng Zhang · Jiaping Gui · Gongshen Liu
1 April 2026 at 12:00
arXiv:2507.10610v2 Announce Type: replace-cross Abstract: Graphical user interface (GUI) agents built on multimodal large language models (MLLMs) have recently demonstrated strong decision-making abilities in screen-based interaction tasks. However, they remain highly vulnerable to pop-up-based environmental injection attacks, where malicious visual elements divert model attention and lead to unsafe or incorrect actions. Existing defense methods either require costly retraining or perform poorly under inductive interference. In this work, we systematically study how such attacks alter the attention behavior of GUI agents and uncover a layer-wise attention divergence pattern between correct and incorrect outputs. Based on this insight, we propose \textbf{LaSM}, a \textit{Layer-wise Scaling Mechanism} that selectively amplifies attention and MLP modules in critical layers. LaSM improves the alignment between model saliency and task-relevant regions without additional training. Extensive experiments across multiple datasets demonstrate that our method significantly improves the defense success rate and exhibits strong robustness, while having negligible impact on the model's general capabilities. Our findings reveal that attention misalignment is a core vulnerability in MLLM agents and can be effectively addressed through selective layer-wise modulation. Our code can be found in https://github.com/YANGTUOMAO/LaSM.

Robust transcriptomic hallmarks targeting intratumor heterogeneity in intrahepatic cholangiocarcinoma

✇(Multiomics OR Omics) AND (Pancreatic)
By: Youpei Lin · Lihua Peng · Haichao Zhao · Penghui Lin · Lv Chen · Zunyun Gong · Jian Lin · Dongbing Liu · Xupeng Yang · Xiangze Song · Mao Zhang · Shuyi Ji · Junhua Rao · Huijuan Luo · Yifei Cheng · Xiaofang Chen · Guohe Song · Hao Hou · Dongning Rao · Tiancheng Zhang · Weiren Liu · Shu Zhang · Gaoyang Xu · Juan Zhang · Liyun Xiao · Yan Liang · Yang Liu · Yuhan Chen · Zhirou Cui · Xiaoying Wang · Jian Zhou · Qiang Gao · Kui Wu · Jia Fan
31 March 2026 at 18:00

Cell Rep Med. 2026 Mar 30:102708. doi: 10.1016/j.xcrm.2026.102708. Online ahead of print.

ABSTRACT

Intratumor heterogeneity (ITH) undermines transcriptome-based stratification in intrahepatic cholangiocarcinoma (iCCA). Here, we integrate multi-omics data from multi-region, single-region, and single-cell RNA sequencing cohorts to systematically characterize gene expression ITH. We uncover that immune and stromal heterogeneity are primary drivers of ITH, leading to misclassification of a median 27.8% of tumors by existing subtyping systems. To overcome this, we identify a low-intratumor-heterogeneity/high-intertumor-variability (LIHV) gene set and develop an ITH-insensitive classification system defining five subgroups: inflammatory (SI), metabolic (SII), atypical (SIII-1), immune-silent (SIII-2), and neurodegenerative (SIII-3). These subgroups exhibit distinct clinical outcomes, molecular features, immune landscapes, and therapeutic vulnerabilities. GPRC5A and VTCN1 serve as robust immunohistochemical biomarkers for SI and SIII tumors, while serum CEA and CA19-9 identify inflammatory iCCA. Therapeutically, HSP90 inhibition synergizes with anti-PD1 in inflammatory iCCA, whereas combined anti-PD1 and anti-TIM3 suppresses neurodegenerative iCCA. Collectively, our study provides a robust molecular framework and actionable therapeutic strategies for iCCA.

PMID:41916296 | DOI:10.1016/j.xcrm.2026.102708

Dynamic Targetable Extracellular Vesicle Surface Proteins Monitor Depth of Response to CAR T Therapy

✇MRD
By: Chen Zhao · Lei Qiu · Ning An · Yong Ju · Jacqueline Ziqian Yang · Hyoyong Kim · Ryan Y Zhang · Junseok Lee · Jingjing He · Liang-Yan Chen · Yaya Xu · Yue Ma · Emily Ren · Sophie Pan · Pranay Sashikanth · Jina Kim · Di Wang · Peiling Zhang · Jianlin Hu · Yang Gao · Yuhan Bao · Yanhua Tu · Li Wang · Zhili Wang · Zhicheng Zhang · Renjun Pei · Edwin M Posadas · Sungyong You · Wanxing Chai-Ho · Tasha L Lin · Sarah M Larson · Dinesh S Rao · Hsian-Rong Tseng · Na Sun · Chunrui Li · Yazhen Zhu
27 March 2026 at 18:00

Res Sq [Preprint]. 2026 Mar 18:rs.3.rs-8913641. doi: 10.21203/rs.3.rs-8913641/v1.

ABSTRACT

Extracellular vesicles (EVs) represent a promising liquid biopsy platform in multiple myeloma (MM). We developed an MM EV Surface Protein Assay to quantify and dynamically monitor four MM EV subpopulations defined by targetable MM surface proteins (BCMA, CD38, GPRC5D, and CD319) across 336 serial blood samples from 45 relapsed/refractory MM (RRMM) patients treated with anti-BCMA chimeric antigen receptor (CAR) T-cell therapy. All four MM EV subpopulations significantly decreased in 43 patients with initial response, while BCMA+, GPRC5D+, and CD319+ MM EVs increased in 19 patients with progression, and antigen escape was detected by BCMA+ MM EVs. MM EV subpopulations differentiated minimal residual disease (MRD) status and complemented MRD for detecting early relapse before clinical progression. Notably, CD319+ MM EVs were early predictors of progression-free and overall survival in MRD-negative patients. This assay enables noninvasive monitoring of deep response, progression, and antigen escape, and stratifies survival in MRD-negative patients with RRMM.

PMID:41890853 | PMC:PMC13015583 | DOI:10.21203/rs.3.rs-8913641/v1

Received — 26 March 2026 ⏭

OmniDiT: Extending Diffusion Transformer to Omni-VTON Framework

✇cs.AI, q-bio.NC updates on arXiv.org
By: Weixuan Zeng · Pengcheng Wei · Huaiqing Wang · Boheng Zhang · Jia Sun · Dewen Fan · Lin HE · Long Chen · Qianqian Gan · Fan Yang · Tingting Gao
25 March 2026 at 12:00
arXiv:2603.19643v2 Announce Type: replace-cross Abstract: Despite the rapid advancement of Virtual Try-On (VTON) and Try-Off (VTOFF) technologies, existing VTON methods face challenges with fine-grained detail preservation, generalization to complex scenes, complicated pipeline, and efficient inference. To tackle these problems, we propose OmniDiT, an omni Virtual Try-On framework based on the Diffusion Transformer, which combines try-on and try-off tasks into one unified model. Specifically, we first establish a self-evolving data curation pipeline to continuously produce data, and construct a large VTON dataset Omni-TryOn, which contains over 380k diverse and high-quality garment-model-tryon image pairs and detailed text prompts. Then, we employ the token concatenation and design an adaptive position encoding to effectively incorporate multiple reference conditions. To relieve the bottleneck of long sequence computation, we are the first to introduce Shifted Window Attention into the diffusion model, thus achieving a linear complexity. To remedy the performance degradation caused by local window attention, we utilize multiple timestep prediction and an alignment loss to improve generation fidelity. Experiments reveal that, under various complex scenes, our method achieves the best performance in both the model-free VTON and VTOFF tasks and a performance comparable to current SOTA methods in the model-based VTON task.
Received — 17 March 2026 ⏭

Unraveling the Link Between Azathioprine and Acute Pancreatitis: Integrating Network Toxicology, Machine Learning, and Mendelian Randomization

✇(Multiomics OR Omics) AND (Lung OR gastric OR Hepatocellular)
By: Zhijun Xie · Hang Lei · Pengcheng Zhang · Zhi Li · Wenfu Tang
15 March 2026 at 18:00

CPT Pharmacometrics Syst Pharmacol. 2026 Mar;15(3):e70178. doi: 10.1002/psp4.70178.

ABSTRACT

Azathioprine (AZA), a widely used immunosuppressant, can induce acute pancreatitis (AP), yet the underlying molecular mechanisms remain unclear. This study employed an integrative multiomics strategy-combining network toxicology, machine learning, Mendelian randomization (MR), and molecular docking-to elucidate the biological basis of AZA-induced AP. AZA-associated genes were first identified through bioinformatics databases and analyzed using protein-protein interaction networks and GO/KEGG functional enrichment. Least absolute shrinkage and selection operator (LASSO) regression and support vector machine recursive feature elimination (SVM-RFE) were applied to prioritize key differentially expressed genes for diagnostic modeling. MR was then used to examine potential causal links between gene expression and AP risk, followed by molecular docking to assess AZA-protein interactions. Sixty-eight candidate genes related to AZA-induced AP were identified. Enrichment analyses indicated involvement in lipid metabolic regulation, inflammatory pathways, and energy homeostasis. Machine learning highlighted seven key genes-CES1, CTSK, JAK1, NR3C2, PLIN5, WEE1, and RORA-as central to AP development. MR analysis further demonstrated that decreased expression of CES1 and CTSK may mediate AZA-related AP susceptibility. Docking simulations revealed strong, specific binding between AZA and both CES1 and CTSK. Overall, this study identifies CES1 and CTSK as genetically protective factors and mechanistic mediators in AZA-triggered AP. These findings offer new molecular insights into the genomic and biochemical pathways underlying this adverse drug reaction.

PMID:41832938 | DOI:10.1002/psp4.70178

LINC-AC092535.5 regulates MICAL2 mRNA level to inhibit p53-mediated ferroptosis in nasopharyngeal carcinoma

✇Oncogene - Issue - nature.com science feeds
By: Siyu Zhang · Wenyue Chen · Ji Yang · Zhen Ge · Yan Kong · Haiou Yan · Lili Bao · Ying Shan · Zhiyuan Yang · Wenhui Chen · Si Pan · Yiwen You · Bo You · Qicheng Zhang
14 March 2026 at 08:00

Oncogene, Published online: 14 March 2026; doi:10.1038/s41388-026-03714-y

LINC-AC092535.5 regulates MICAL2 mRNA level to inhibit p53-mediated ferroptosis in nasopharyngeal carcinoma
Received — 11 March 2026 ⏭

RECAP: Local Hebbian Prototype Learning as a Self-Organizing Readout for Reservoir Dynamics

✇cs.AI, q-bio.NC updates on arXiv.org
By: Heng Zhang
10 March 2026 at 12:00
arXiv:2603.06639v1 Announce Type: cross Abstract: Robust perception in brains is often attributed to high-dimensional population activity together with local plasticity mechanisms that reinforce recurring structure. In contrast, most modern image recognition systems are trained by error backpropagation and end-to-end gradient optimization, which are not naturally aligned with local computation and local plasticity. We introduce RECAP (Reservoir Computing with Hebbian Co-Activation Prototypes), a bio-inspired learning strategy for robust image classification that couples untrained reservoir dynamics with a self-organizing Hebbian prototype readout. RECAP discretizes time-averaged reservoir responses into activation levels, constructs a co-activation mask over reservoir unit pairs, and incrementally updates class-wise prototype matrices via a Hebbian-like potentiation-decay rule. Inference is performed by overlap-based prototype matching. The method avoids error backpropagation and is naturally compatible with online prototype updates. We illustrate the resulting robustness behavior on MNIST-C, where RECAP remains robust under diverse corruptions without exposure to corrupted training samples.

Adaptation of Agentic AI: A Survey of Post-Training, Memory, and Skills

✇cs.AI, q-bio.NC updates on arXiv.org
By: Pengcheng Jiang · Jiacheng Lin · Zhiyi Shi · Zifeng Wang · Luxi He · Yichen Wu · Ming Zhong · Peiyang Song · Qizheng Zhang · Heng Wang · Xueqiang Xu · Hanwen Xu · Pengrui Han · Dylan Zhang · Jiashuo Sun · Chaoqi Yang · Kun Qian · Tian Wang · Changran Hu · Manling Li · Quanzheng Li · Hao Peng · Sheng Wang · Jingbo Shang · Chao Zhang · Jiaxuan You · Liyuan Liu · Pan Lu · Yu Zhang · Heng Ji · Yejin Choi · Dawn Song · Jimeng Sun · Jiawei Han
10 March 2026 at 12:00
arXiv:2512.16301v3 Announce Type: replace Abstract: Large language model (LLM) agents are moving beyond prompting alone. ChatGPT marked the rise of general-purpose LLM assistants, DeepSeek showed that on-policy reinforcement learning with verifiable rewards can improve reasoning and tool use, and OpenClaw highlights a newer direction in which agents accumulate persistent memory and reusable skills. Yet the research landscape remains fragmented across post-training, retrieval, memory, and skill systems. This survey studies these developments under a single notion of \emph{adaptation}: improving an agent, its tools, or their interaction after pretraining. We organize the field with a four-paradigm framework spanning agent adaptation and tool adaptation. On the agent side, A1 (tool-execution-signaled) and A2 (agent-output-signaled) improve the agent itself through supervised fine-tuning, preference optimization, and reinforcement learning with verifiable rewards. On the tool side, T1 (agent-agnostic) provides reusable pre-trained modules any agent can call, while T2 (agent-supervised) uses the agent's outputs to train memory systems, skill libraries, or lightweight subagents. Using this framework, we review post-training methods, adaptive memory architectures, and agent skills; compare their trade-offs in cost, flexibility, and generalization; and summarize evaluation practices across deep research, software development, computer use, and drug discovery. We conclude by outlining open problems in agent-tool co-adaptation, continual learning, safety, and efficient deployment.

CaMeLs Can Use Computers Too: System-level Security for Computer Use Agents

✇cs.AI, q-bio.NC updates on arXiv.org
By: Hanna Foerster · Tom Blanchard · Kristina Nikoli\'c · Ilia Shumailov · Cheng Zhang · Robert Mullins · Nicolas Papernot · Florian Tram\`er · Yiren Zhao
10 March 2026 at 12:00
arXiv:2601.09923v2 Announce Type: replace Abstract: AI agents are vulnerable to prompt injection attacks, where malicious content hijacks agent behavior to steal credentials or cause financial loss. The only known robust defense is architectural isolation that strictly separates trusted task planning from untrusted environment observations. However, applying this design to Computer Use Agents (CUAs) -- systems that automate tasks by viewing screens and executing actions -- presents a fundamental challenge: current agents require continuous observation of UI state to determine each action, conflicting with the isolation required for security. We resolve this tension by demonstrating that UI workflows, while dynamic, are structurally predictable. We introduce Single-Shot Planning for CUAs, where a trusted planner generates a complete execution graph with conditional branches before any observation of potentially malicious content, providing provable control flow integrity guarantees against arbitrary instruction injections. Although this architectural isolation successfully prevents instruction injections, we show that additional measures are needed to prevent Branch Steering attacks, which manipulate UI elements to trigger unintended valid paths within the plan. We evaluate our design on OSWorld, and retain up to 57% of the performance of frontier models while improving performance for smaller open-source models by up to 19%, demonstrating that rigorous security and utility can coexist in CUAs.

Unified Medical Image Segmentation with State Space Modeling Snake

✇cs.AI, q-bio.NC updates on arXiv.org
By: Ruicheng Zhang · Haowei Guo · Kanghui Tian · Jun Zhou · Mingliang Yan · Zeyu Zhang · Shen Zhao
10 March 2026 at 12:00
arXiv:2507.12760v2 Announce Type: replace-cross Abstract: Unified Medical Image Segmentation (UMIS) is critical for comprehensive anatomical assessment but faces challenges due to multi-scale structural heterogeneity. Conventional pixel-based approaches, lacking object-level anatomical insight and inter-organ relational modeling, struggle with morphological complexity and feature conflicts, limiting their efficacy in UMIS. We propose Mamba Snake, a novel deep snake framework enhanced by state space modeling for UMIS. Mamba Snake frames multi-contour evolution as a hierarchical state space atlas, effectively modeling macroscopic inter-organ topological relationships and microscopic contour refinements. We introduce a snake-specific vision state space module, the Mamba Evolution Block (MEB), which leverages effective spatiotemporal information aggregation for adaptive refinement of complex morphologies. Energy map shape priors further ensure robust long-range contour evolution in heterogeneous data. Additionally, a dual-classification synergy mechanism is incorporated to concurrently optimize detection and segmentation, mitigating under-segmentation of microstructures in UMIS. Extensive evaluations across five clinical datasets reveal Mamba Snake's superior performance, with an average Dice improvement of 3\% over state-of-the-art methods.
Received — 5 March 2026 ⏭

MemSifter: Offloading LLM Memory Retrieval via Outcome-Driven Proxy Reasoning

✇cs.AI, q-bio.NC updates on arXiv.org
By: Jiejun Tan · Zhicheng Dou · Liancheng Zhang · Yuyang Hu · Yiruo Cheng · Ji-Rong Wen
5 March 2026 at 13:00
arXiv:2603.03379v1 Announce Type: cross Abstract: As Large Language Models (LLMs) are increasingly used for long-duration tasks, maintaining effective long-term memory has become a critical challenge. Current methods often face a trade-off between cost and accuracy. Simple storage methods often fail to retrieve relevant information, while complex indexing methods (such as memory graphs) require heavy computation and can cause information loss. Furthermore, relying on the working LLM to process all memories is computationally expensive and slow. To address these limitations, we propose MemSifter, a novel framework that offloads the memory retrieval process to a small-scale proxy model. Instead of increasing the burden on the primary working LLM, MemSifter uses a smaller model to reason about the task before retrieving the necessary information. This approach requires no heavy computation during the indexing phase and adds minimal overhead during inference. To optimize the proxy model, we introduce a memory-specific Reinforcement Learning (RL) training paradigm. We design a task-outcome-oriented reward based on the working LLM's actual performance in completing the task. The reward measures the actual contribution of retrieved memories by mutiple interactions with the working LLM, and discriminates retrieved rankings by stepped decreasing contributions. Additionally, we employ training techniques such as Curriculum Learning and Model Merging to improve performance. We evaluated MemSifter on eight LLM memory benchmarks, including Deep Research tasks. The results demonstrate that our method meets or exceeds the performance of existing state-of-the-art approaches in both retrieval accuracy and final task completion. MemSifter offers an efficient and scalable solution for long-term LLM memory. We have open-sourced the model weights, code, and training data to support further research.

See, Think, Act: Teaching Multimodal Agents to Effectively Interact with GUI by Identifying Toggles

✇cs.AI, q-bio.NC updates on arXiv.org
By: Zongru Wu · Rui Mao · Zhiyuan Tian · Pengzhou Cheng · Tianjie Ju · Zheng Wu · Lingzhong Dong · Haiyue Sheng · Zhuosheng Zhang · Gongshen Liu
5 March 2026 at 13:00
arXiv:2509.13615v4 Announce Type: replace Abstract: The advent of multimodal agents facilitates effective interaction within graphical user interface (GUI), especially in ubiquitous GUI control. However, their inability to reliably execute toggle control instructions remains a key bottleneck. To investigate this, we construct a state control benchmark with binary toggle instructions derived from public datasets. Evaluation results of existing agents demonstrate their notable unreliability, particularly when the current toggle state already matches the desired state. To address the challenge, we propose State-aware Reasoning (StaR), a multimodal reasoning method that enables agents to perceive the current toggle state, infer the desired state from the instruction, and act accordingly. Experiments on four multimodal agents demonstrate that StaR can improve toggle instruction execution accuracy by over 30\%. Further evaluations on three public agentic benchmarks show that StaR also enhances general agentic task performance. Finally, evaluations on a dynamic environment highlight the potential of StaR for real-world applications. Code and benchmark: https://github.com/ZrW00/StaR.

Training High-Level Schedulers with Execution-Feedback Reinforcement Learning for Long-Horizon GUI Automation

✇cs.AI, q-bio.NC updates on arXiv.org
By: Zehao Deng · Tianjie Ju · Zheng Wu · Zhuosheng Zhang · Gongshen Liu
5 March 2026 at 13:00
arXiv:2511.22235v2 Announce Type: replace Abstract: The rapid development of large vision-language model (VLM) has greatly promoted the research of GUI agent. However, GUI agents still face significant challenges in handling long-horizon tasks. First, single-agent models struggle to balance high-level capabilities and low-level execution capability, facing prevalent issues of responsibility coupling and capability conflicts. Second, agents lack awareness of the task state, leading to progress loss in long-horizon tasks. To address these challenges, we propose a staged execution-feedback reinforcement learning algorithm. Unlike training a unified policy model, we focus on training high-level scheduling models. Specifically, we propose and train two agents: a Coordinator, responsible for the strategic planning and task decomposition; and a State Tracker, responsible for context compression and information management to maintain the task's state and coherence. Based on this, we built the Coordinator-Executor-State Tracker (CES) multi-agent framework, which can be integrated with any low-level Executor model, assisting the Executor in solving long-horizon tasks through task scheduling and state management. Experiments on long-horizon task benchmarks demonstrate that CES significantly enhances the system's planning and state management capabilities. Furthermore, analysis confirms that our trained high-level scheduling module is a generalizable, plug-and-play module that significantly enhances the long-horizon capabilities of various Executors. Code can be available at https://github.com/hehehahi4/CES.

Dynamic Adversarial Reinforcement Learning for Robust Multimodal Large Language Models

✇cs.AI, q-bio.NC updates on arXiv.org
By: Yicheng Bao · Xuhong Wang · Qiaosheng Zhang · Chaochao Lu · Xia Hu · Xin Tan
5 March 2026 at 13:00
arXiv:2602.22227v3 Announce Type: replace-cross Abstract: Despite their impressive capabilities, Multimodal Large Language Models (MLLMs) exhibit perceptual fragility when confronted with visually complex scenes. This weakness stems from a reliance on finite training datasets, which are prohibitively expensive to scale and impose a ceiling on model robustness. We introduce \textbf{AOT-SFT}, a large-scale adversarial dataset for bootstrapping MLLM robustness. Building on this, we propose \textbf{AOT (Adversarial Opponent Training)}, a self-play framework that forges MLLM robustness by creating its own training data. Our method orchestrates a co-evolution between an image-editing Attacker and a Defender MLLM, where the Attacker generates a diverse and dynamic curriculum of image manipulations, forcing the Defender to adapt and improve. Extensive experiments demonstrate that AOT enhances the Defender's perceptual robustness and reduces hallucinations, establishing a scalable paradigm for training more reliable MLLMs.

MedFeat: Model-Aware and Explainability-Driven Feature Engineering with LLMs for Clinical Tabular Prediction

✇cs.AI, q-bio.NC updates on arXiv.org
By: Zizheng Zhang · Yiming Li · Justin Xu · Jinyu Wang · Rui Wang · Lei Song · Jiang Bian · David W Eyre · Jingjing Fu
4 March 2026 at 13:00
arXiv:2603.02221v1 Announce Type: cross Abstract: In healthcare tabular predictions, classical models with feature engineering often outperform neural approaches. Recent advances in Large Language Models enable the integration of domain knowledge into feature engineering, offering a promising direction. However, existing approaches typically rely on a broad search over predefined transformations, overlooking downstream model characteristics and feature importance signals. We present MedFeat, a feedback-driven and model-aware feature engineering framework that leverages LLM reasoning with domain knowledge and provides feature explanations based on SHAP values while tracking successful and failed proposals to guide feature discovery. By incorporating model awareness, MedFeat prioritizes informative signals that are difficult for the downstream model to learn directly due to its characteristics. Across a broad range of clinical prediction tasks, MedFeat achieves stable improvements over various baselines and discovers clinically meaningful features that generalize under distribution shift, demonstrating robustness across years and from ICU cohorts to general hospitalized patients, thereby offering insights into real-world deployment. Code required to reproduce our experiments will be released, subject to dataset agreements and institutional policies.

Efficient Self-Evaluation for Diffusion Language Models via Sequence Regeneration

✇cs.AI, q-bio.NC updates on arXiv.org
By: Linhao Zhong · Linyu Wu · Wen Wang · Yuling Xi · Chenchen Jing · Jiaheng Zhang · Hao Chen · Chunhua Shen
4 March 2026 at 13:00
arXiv:2603.02760v1 Announce Type: cross Abstract: Diffusion large language models (dLLMs) have recently attracted significant attention for their ability to enhance diversity, controllability, and parallelism. However, their non-sequential, bidirectionally masked generation makes quality assessment difficult, underscoring the need for effective self-evaluation. In this work, we propose DiSE, a simple yet effective self-evaluation confidence quantification method for dLLMs. DiSE quantifies confidence by computing the probability of regenerating the tokens in the entire generated sequence, given the full context. This method enables more efficient and reliable quality assessment by leveraging token regeneration probabilities, facilitating both likelihood estimation and robust uncertainty quantification. Building upon DiSE, we further introduce a flexible-length generation framework, which adaptively controls the sequence length based on the model's self-assessment of its own output. We analyze and validate the feasibility of DiSE from the perspective of dLLM generalization, and empirically demonstrate that DiSE is positively correlated with both semantic coherence and answer accuracy. Extensive experiments on likelihood evaluation, uncertainty quantification, and flexible-length generation further confirm the effectiveness of the proposed DiSE.

Chain of World: World Model Thinking in Latent Motion

✇cs.AI, q-bio.NC updates on arXiv.org
By: Fuxiang Yang · Donglin Di · Lulu Tang · Xuancheng Zhang · Lei Fan · Hao Li · Chen Wei · Tonghua Su · Baorui Ma
4 March 2026 at 13:00
arXiv:2603.03195v1 Announce Type: cross Abstract: Vision-Language-Action (VLA) models are a promising path toward embodied intelligence, yet they often overlook the predictive and temporal-causal structure underlying visual dynamics. World-model VLAs address this by predicting future frames, but waste capacity reconstructing redundant backgrounds. Latent-action VLAs encode frame-to-frame transitions compactly, but lack temporally continuous dynamic modeling and world knowledge. To overcome these limitations, we introduce CoWVLA (Chain-of-World VLA), a new "Chain of World" paradigm that unifies world-model temporal reasoning with a disentangled latent motion representation. First, a pretrained video VAE serves as a latent motion extractor, explicitly factorizing video segments into structure and motion latents. Then, during pre-training, the VLA learns from an instruction and an initial frame to infer a continuous latent motion chain and predict the segment's terminal frame. Finally, during co-fine-tuning, this latent dynamic is aligned with discrete action prediction by jointly modeling sparse keyframes and action sequences in a unified autoregressive decoder. This design preserves the world-model benefits of temporal reasoning and world knowledge while retaining the compactness and interpretability of latent actions, enabling efficient visuomotor learning. Extensive experiments on robotic simulation benchmarks show that CoWVLA outperforms existing world-model and latent-action approaches and achieves moderate computational efficiency, highlighting its potential as a more effective VLA pretraining paradigm. The project website can be found at https://fx-hit.github.io/cowvla-io.

See, Think, Act: Teaching Multimodal Agents to Effectively Interact with GUI by Identifying Toggles

✇cs.AI, q-bio.NC updates on arXiv.org
By: Zongru Wu · Rui Mao · Zhiyuan Tian · Pengzhou Cheng · Tianjie Ju · Zheng Wu · Lingzhong Dong · Haiyue Sheng · Zhuosheng Zhang · Gongshen Liu
4 March 2026 at 13:00
arXiv:2509.13615v3 Announce Type: replace Abstract: The advent of multimodal agents facilitates effective interaction within graphical user interface (GUI), especially in ubiquitous GUI control. However, their inability to reliably execute toggle control instructions remains a key bottleneck. To investigate this, we construct a state control benchmark with binary toggle instructions derived from public datasets. Evaluation results of existing agents demonstrate their notable unreliability, particularly when the current toggle state already matches the desired state. To address the challenge, we propose State-aware Reasoning (StaR), a multimodal reasoning method that enables agents to perceive the current toggle state, infer the desired state from the instruction, and act accordingly. Experiments on four multimodal agents demonstrate that StaR can improve toggle instruction execution accuracy by over 30\%. Further evaluations on three public agentic benchmarks show that StaR also enhances general agentic task performance. Finally, evaluations on a dynamic environment highlight the potential of StaR for real-world applications. Code and benchmark: https://github.com/ZrW00/StaR.
❌