❌

Normal view

Received — 27 May 2026 ⏭

Tiny Brains, Giant Impact: Uncovering the Keystone Neurons of LLM with Just a Few Prompts

✇cs.AI, q-bio.NC updates on arXiv.org
By: Xiangtian Ji · Yuxin Chen · Zhengzhou Cai · Xiang Wang · An Zhang · Tat-Seng Chua
26 May 2026 at 12:00
arXiv:2605.24846v1 Announce Type: cross Abstract: Large language models (LLMs) display strong comprehensive abilities, yet the internal mechanisms that support these behaviors remain insufficiently understood. In this work, we show that across a wide range of open-weight Transformers, a subset of neurons remains consistently highly activated during inference across tasks of multiple capability dimensions. By probing along the cross-task activation strength, an extremely sparse subset is isolated, whose removal causes a collapse in model behavior, which we term keystone neurons. Our analysis reveals that keystone neurons are a stable and intrinsic neuron subset of the model that is largely established during pretraining. The parameters associated with these neurons are tightly calibrated during the training process, and their precise values are critical for the capabilities of the model. Building on these insights, we propose a supervised fine-tuning approach that updates only keystone neurons, achieving task gains comparable to or even better than full-parameter fine-tuning while better preserving performance in other capability dimensions, despite modifying a much smaller number of parameters.

AgentArk: Distilling Multi-Agent Intelligence into a Single LLM Agent

✇cs.AI, q-bio.NC updates on arXiv.org
By: Yinyi Luo · Yiqiao Jin · Weichen Yu · Mengqi Zhang · Srijan Kumar · Xiaoxiao Li · Weijie Xu · Xin Chen · Jindong Wang
26 May 2026 at 12:00
arXiv:2602.03955v3 Announce Type: replace Abstract: While large language model (LLM) multi-agent systems achieve superior reasoning performance through iterative debate, practical deployment is limited by their high computational cost and error propagation. This paper proposes AgentArk, a novel framework to distill multi-agent dynamics into the weights of a single model, effectively transforming explicit test-time interactions into implicit model capabilities. This equips a single agent with the intelligence of multi-agent systems while remaining computationally efficient. Specifically, we investigate three hierarchical distillation strategies across various models, tasks, scaling, and scenarios: reasoning-enhanced fine-tuning; trajectory-based augmentation; and process-aware distillation. By shifting the burden of computation from inference to training, the distilled models preserve the efficiency of one agent while exhibiting strong reasoning and self-correction performance of multiple agents. They further demonstrate enhanced robustness and generalization across diverse reasoning tasks. We hope this work can shed light on future research on efficient and robust multi-agent development. Our code is at https://github.com/AIFrontierLab/AgentArk.

Interpreting Omics Data Analysis with Large Language Models for Disease Target and Drug Discovery

✇(Multiomics OR Omics) AND (Pancreatic)
By: Zixi Xu · Weihang Chen · Wuyu Ren · Tianqi Xu · Somadina Amaechina · Raad Khan · Yixin Chen · Michael Province · Philip Payne · Fuhai Li
18 May 2026 at 18:00

bioRxiv [Preprint]. 2026 May 5:2026.04.30.721768. doi: 10.64898/2026.04.30.721768.

ABSTRACT

In biomedical scientific discovery, synthesizing prior knowledge from the literature is an essential component of interpreting numerical omics data analyses for disease target identification and drug discovery. Large language models (LLMs) alone can rapidly retrieve disease mechanisms from biomedical text, but text-only outputs are general and unreliable for target and drug prioritization without cohort-specific quantitative evidence. Herein, we propose a provenance-aware Text-to-Target framework that couples schema-constrained multi-model LLM retrieval with numeric omics data analysis. The key design is a modality-aware fusion step: candidates are partitioned into overlap-supported anchors, retrieval-only hidden hubs, and network-emergent novelty nodes, then propagated into staged hypothesis and strategy generation under topology constraints. We evaluate the model in Alzheimer's disease (AD) and pancreatic ductal adenocarcinoma (PDAC). In PDAC, the workflow produced a balanced 75-gene candidate universe and a 23-strategy portfolio, with significant DepMap support at both target level and strategy level. In AD, stricter candidate controls yielded a compact 34-gene universe and 14 strategies; under an expanded CRISPRbrain registry, both target-level axes were significant, with strong strategy-level enrichment. Across both diseases, final strategies preserved full provenance closure to the candidate pool, enabling end-to-end auditability from retrieval artifacts to validation outputs. These results support a transferable discovery architecture in which omics evidence constrains biological activity, LLM retrieval expands mechanistic search space, and network-aware fusion preserves interpretability. The framework provides a reproducible basis for dual-disease target prioritization and motivates continuous literature-mechanism concordance with agentic evidence-refresh loops.

PMID:42146439 | PMC:PMC13174328 | DOI:10.64898/2026.04.30.721768

GPNMB Drives Brain Metastasis by Sculpting a Pathological Endothelial-Immune Interactome

✇Omics In Lung
By: Xuefei Liu · Jun Tan · Chun Wu · Guanyin Huang · Yixin Cheng · Jianyang Hu · Binyu Zhang · Mao Zhao · Boxi Zhao · Jingru Lian · Shuqian Zheng · Lin Zeng · Meng Xu · Yang Xu · Shan Zeng · Hao Yu · Hui Yang · Zhixiang Zuo · Chuanyu Liu · Weineng Feng · Weinan Guo · Chunying Li · Sai-Lan Liu · Qing Liu · Feiqiu Wen · Xin Hong
13 April 2026 at 18:00

Cancer Discov. 2026 Apr 15. doi: 10.1158/2159-8290.CD-25-1663. Online ahead of print.

ABSTRACT

Brain metastases (BM) remain a devastating disease with dismal prognosis. How circulating tumor cells (CTCs) penetrate the blood brain barrier (BBB) and reprogram the brain microenvironment remain unclear. Using spatially resolved multi-omic profiling of CTCs and brain metastases, integrated with experimental and clinical analyses, we identified Glycoprotein Non-Metastatic Melanoma Protein B (GPNMB) as a CTC-secreted driver of vascular disruption and brain colonization. CBX3 upregulation induced GPNMB expression, which bound endothelial EGFR, triggering CBL-mediated ubiquitination and degradation. Attenuated EGFR signaling suppressed FTO and disrupted endothelial junctions via YTHDF2-dependent TJP1 m6A methylation. Remarkably, GPNMB-induced BBB remodeling promoted immune infiltration via CXCL12-CXCR4 axis, and induced time course-dependent T cell exhaustion within the brain microenvironment. Clinically, elevated CBX3⁺GPNMB⁺ CTCs and plasma CXCL12 were significantly associated with BM progression in lung cancer and melanoma. Therapeutically, dual blockade of GPNMB and PD1 enhanced anti-BM efficacy in mice, unveiling GPNMB as a promising target for precision immunotherapy.

PMID:41973996 | DOI:10.1158/2159-8290.CD-25-1663

Received — 8 April 2026 ⏭

HAG: Hierarchical Demographic Tree-based Agent Generation for Topic-Adaptive Simulation

✇cs.AI, q-bio.NC updates on arXiv.org
By: Rongxin Chen · Tianyu Wu · Bingbing Xu · Jiatang Luo · Xiucheng Xu · Huawei Shen
7 April 2026 at 12:00
arXiv:2601.05656v3 Announce Type: replace Abstract: High-fidelity agent initialization is crucial for credible Agent-Based Modeling across diverse domains. A robust framework should be Topic-Adaptive, capturing macro-level joint distributions while ensuring micro-level individual rationality. Existing approaches fall into two categories: static data-based retrieval methods that fail to adapt to unseen topics absent from the data, and LLM-based generation methods that lack macro-level distribution awareness, resulting in inconsistencies between micro-level persona attributes and reality. To address these problems, we propose HAG, a Hierarchical Agent Generation framework that formalizes population generation as a two-stage decision process. Firstly, utilizing a World Knowledge Model to infer hierarchical conditional probabilities to construct the Topic-Adaptive Tree, achieving macro-level distribution alignment. Then, grounded real-world data, instantiation and agentic augmentation are carried out to ensure micro-level consistency. Given the lack of specialized evaluation, we establish a multi-domain benchmark and a comprehensive PACE evaluation framework. Extensive experiments show that HAG significantly outperforms representative baselines, reducing population alignment errors by an average of 37.7% and enhancing sociological consistency by 18.8%.

SRSF10 promotes cisplatin resistance in bladder cancer via BIN1 Exon 12 retention and ANXA1 activation

✇Oncogene - Issue - nature.com science feeds
By: Mengqing Xiao · Yanni Ma · Ganghua Zhang · Guanjun Chen · Yuxing Zhu · Xiaoming Liu · Yaxin Cheng · Ziyang Feng · Ke Cao
6 April 2026 at 08:00

Oncogene, Published online: 06 April 2026; doi:10.1038/s41388-026-03735-7

SRSF10 promotes cisplatin resistance in bladder cancer via BIN1 Exon 12 retention and ANXA1 activation
Received — 4 April 2026 ⏭

SHOE: Semantic HOI Open-Vocabulary Evaluation Metric

✇cs.AI, q-bio.NC updates on arXiv.org
By: Maja Noack · Qinqian Lei · Taipeng Tian · Bihan Dong · Robby T. Tan · Yixin Chen · John Young · Saijun Zhang · Bo Wang
3 April 2026 at 12:00
arXiv:2604.01586v1 Announce Type: cross Abstract: Open-vocabulary human-object interaction (HOI) detection is a step towards building scalable systems that generalize to unseen interactions in real-world scenarios and support grounded multimodal systems that reason about human-object relationships. However, standard evaluation metrics, such as mean Average Precision (mAP), treat HOI classes as discrete categorical labels and fail to credit semantically valid but lexically different predictions (e.g., "lean on couch" vs. "sit on couch"), limiting their applicability for evaluating open-vocabulary predictions that go beyond any predefined set of HOI labels. We introduce SHOE (Semantic HOI Open-Vocabulary Evaluation), a new evaluation framework that incorporates semantic similarity between predicted and ground-truth HOI labels. SHOE decomposes each HOI prediction into its verb and object components, estimates their semantic similarity using the average of multiple large language models (LLMs), and combines them into a similarity score to evaluate alignment beyond exact string match. This enables a flexible and scalable evaluation of both existing HOI detection methods and open-ended generative models using standard benchmarks such as HICO-DET. Experimental results show that SHOE scores align more closely with human judgments than existing metrics, including LLM-based and embedding-based baselines, achieving an agreement of 85.73% with the average human ratings. Our work underscores the need for semantically grounded HOI evaluation that better mirrors human understanding of interactions. We will release our evaluation metric to the public to facilitate future research.

Lifting Unlabeled Internet-level Data for 3D Scene Understanding

✇cs.AI, q-bio.NC updates on arXiv.org
By: Yixin Chen · Yaowei Zhang · Huangyue Yu · Junchao He · Yan Wang · Jiangyong Huang · Hongyu Shen · Junfeng Ni · Shaofei Wang · Baoxiong Jia · Song-Chun Zhu · Siyuan Huang
3 April 2026 at 12:00
arXiv:2604.01907v1 Announce Type: cross Abstract: Annotated 3D scene data is scarce and expensive to acquire, while abundant unlabeled videos are readily available on the internet. In this paper, we demonstrate that carefully designed data engines can leverage web-curated, unlabeled videos to automatically generate training data, to facilitate end-to-end models in 3D scene understanding alongside human-annotated datasets. We identify and analyze bottlenecks in automated data generation, revealing critical factors that determine the efficiency and effectiveness of learning from unlabeled data. To validate our approach across different perception granularities, we evaluate on three tasks spanning low-level perception, i.e., 3D object detection and instance segmentation, to high-evel reasoning, i.e., 3D spatial Visual Question Answering (VQA) and Vision-Lanugage Navigation (VLN). Models trained on our generated data demonstrate strong zero-shot performance and show further improvement after finetuning. This demonstrates the viability of leveraging readily available web data as a path toward more capable scene understanding systems.

Efficient Reasoning with Balanced Thinking

✇cs.AI, q-bio.NC updates on arXiv.org
By: Yulin Li · Tengyao Tu · Li Ding · Junjie Wang · Huiling Zhen · Yixin Chen · Yong Li · Zhuotao Tian
3 April 2026 at 12:00
arXiv:2603.12372v3 Announce Type: replace Abstract: Large Reasoning Models (LRMs) have shown remarkable reasoning capabilities, yet they often suffer from overthinking, expending redundant computational steps on simple problems, or underthinking, failing to explore sufficient reasoning paths despite inherent capabilities. These issues lead to inefficiencies and potential inaccuracies, limiting practical deployment in resource-constrained settings. Existing methods to mitigate overthinking, such as suppressing reflective keywords or adjusting reasoning length, may inadvertently induce underthinking, compromising accuracy. Therefore, we propose ReBalance, a training-free framework that achieves efficient reasoning with balanced thinking. ReBalance leverages confidence as a continuous indicator of reasoning dynamics, identifying overthinking through high confidence variance and underthinking via consistent overconfidence. By aggregating hidden states from a small-scale dataset into reasoning mode prototypes, we compute a steering vector to guide LRMs' reasoning trajectories. A dynamic control function modulates this vector's strength and direction based on real-time confidence, pruning redundancy during overthinking, and promoting exploration during underthinking. Extensive experiments conducted on four models ranging from 0.5B to 32B, and across nine benchmarks in math reasoning, general question answering, and coding tasks demonstrate that ReBalance effectively reduces output redundancy while improving accuracy, offering a general, training-free, and plug-and-play strategy for efficient and robust LRM deployment. Project page and code are available at https://rebalance-ai.github.io .

Distinctive respiratory toxicity induced by hypoxanthine metabolic disorder from polystyrene microplastics and nanoplastics at environmentally relevant doses: multi-omics insights and experimental validation

✇Omics In Lung
By: Xiaole Xu · Mengnan He · Meiqi Lan · Lan Qiu · Jiajing Wang · Chenghui Zhong · Liqiu Qiu · Wenfeng Lu · Jingwen Fang · Shaojuan Wang · Zhizun Deng · Yingxin Chen · Ping Guo · Yun Zhou · Lieyang Fan
1 April 2026 at 18:00

Environ Int. 2026 Mar 28;210:110212. doi: 10.1016/j.envint.2026.110212. Online ahead of print.

ABSTRACT

Microplastics (MPs) and nanoplastics (NPs) are pervasive environmental contaminants, raising concerns about their potential to cause inflammation, oxidative stress, and lung injury through respiratory toxicity. Due to their smaller size, larger surface area, and greater reactivity, NPs may pose a greater risk than MPs, yet size-dependent toxicity mechanisms remain unclear. This study investigates the distinct early molecular initiating events and toxicological effects of 1 μm polystyrene MPs (PS-MPs) and 20 nm polystyrene NPs (PS-NPs). Based on the internal exposure dose estimated from Py-GC/MS analysis, in vitro exposure concentrations were set at 0, 62.5, 125, 250, 500, and 1000 μg/mL. Multi-omics sequencing and integrative analysis identify specific proteomic and metabolomic alterations. Molecular dynamics simulations and co-immunoprecipitation assays elucidate binding interactions between PS-NPs-induced proteins and metabolic enzymes. In vitro and in vivo experiments reveal a greater accumulation of PS-NPs through endocytosis compared to PS-MPs; while pronounced histopathological damage with inflammatory response in mice lungs were only induced by PS-NPs, rather than PS-MPs. Compared to control group, PS-MPs partly caused proteomic or metabolomic perturbations, while PS-NPs induced significant differential expression of more extensive proteins and metabolites. PS-NPs exposure specifically upregulates insulin-like growth factor 2 receptor (IGF2R) expression and reduces Hypoxanthine levels when compared with PS-MPs. IGF2R directly interacts with Hypoxanthine-guanine phosphoribosyl transferase (HPRT), a key enzyme in Hypoxanthine metabolism, causing its disruption. This study provides important insights into the comparative toxic effects between PS-NPs with PS-MPs, especially the unique toxicological mechanisms of PS-NPs, thereby advancing the understanding of airborne plastic pollutant risks and supporting future regulatory assessments.

PMID:41921402 | DOI:10.1016/j.envint.2026.110212

Received — 2 April 2026 ⏭

Learning to Generate Formally Verifiable Step-by-Step Logic Reasoning via Structured Formal Intermediaries

✇cs.AI, q-bio.NC updates on arXiv.org
By: Luoxin Chen · Yichi Zhou · Huishuai Zhang
1 April 2026 at 12:00
arXiv:2603.29500v1 Announce Type: new Abstract: Large language models (LLMs) have recently demonstrated impressive performance on complex, multi-step reasoning tasks, especially when post-trained with outcome-rewarded reinforcement learning Guo et al. 2025. However, it has been observed that outcome rewards often overlook flawed intermediate steps, leading to unreliable reasoning steps even when final answers are correct. To address this unreliable reasoning, we propose PRoSFI (Process Reward over Structured Formal Intermediates), a novel reward method that enhances reasoning reliability without compromising accuracy. Instead of generating formal proofs directly, which is rarely accomplishable for a modest-sized (7B) model, the model outputs structured intermediate steps aligned with its natural language reasoning. Each step is then verified by a formal prover. Only fully validated reasoning chains receive high rewards. The integration of formal verification guides the model towards generating step-by-step machine-checkable proofs, thereby yielding more credible final answers. PRoSFI offers a simple and effective approach to training trustworthy reasoning models.

DF-ACBlurGAN: Structure-Aware Conditional Generation of Internally Repeated Patterns for Biomaterial Microtopography Design

✇cs.AI, q-bio.NC updates on arXiv.org
By: Rongjun Dong · Xin Chen · Morgan R Alexander · Karthikeyan Sivakumar · Reza Omdivar · David A Winkler · Grazziela Figueredo
1 April 2026 at 12:00
arXiv:2603.28776v1 Announce Type: cross Abstract: Learning to generate images with internally repeated and periodic structures poses a fundamental challenge for machine learning and computer vision models, which are typically optimised for local texture statistics and semantic realism rather than global structural consistency. This limitation is particularly pronounced in applications requiring strict control over repetition scale, spacing, and boundary coherence, such as microtopographical biomaterial surfaces. In this work, biomaterial design serves as a use case to study conditional generation of repeated patterns under weak supervision and class imbalance. We propose DF-ACBlurGAN, a structure-aware conditional generative adversarial network that explicitly reasons about long-range repetition during training. The approach integrates frequency-domain repetition scale estimation, scale-adaptive Gaussian blurring, and unit-cell reconstruction to balance sharp local features with stable global periodicity. Conditioning on experimentally derived biological response labels, the model synthesises designs aligned with target functional outcomes. Evaluation across multiple biomaterial datasets demonstrates improved repetition consistency and controllable structural variation compared to conventional generative approaches.

Pathogenesis and immune regulation of rheumatoid arthritis-associated interstitial lung disease: from basic research to clinical implications

✇Omics In Lung
By: Yaxin Cheng · Yu Shan · Yixin Zheng · Fuyu Zhao · Yiming Shi · Chenyang Song · Yunshen Li · Jianan Zhao · Cen Chang · Yuguang Wang · Dongyi He
30 March 2026 at 18:00

Front Immunol. 2026 Mar 13;17:1770348. doi: 10.3389/fimmu.2026.1770348. eCollection 2026.

ABSTRACT

Interstitial lung disease (ILD) is one of the most common extra-articular manifestations of rheumatoid arthritis (RA). Some patients with RA-ILD may develop progressive pulmonary fibrosis, leading to severe impairment of lung function and respiratory failure, which impacts quality of life and can even be life-threatening. This review identified genetic susceptibility, environmental factors, and immune dysregulation as key contributors to the etiology and pathogenesis of RA-ILD. We highlight that autoantibodies, adaptive immune abnormalities, and tertiary lymphoid organ formation significantly drive pulmonary inflammation and fibrosis, while pro-inflammatory cytokines and epithelial-mesenchymal transition (EMT) further contribute to lung tissue injury. Current treatment options, including glucocorticoids, immunosuppressants, and antifibrotic agents such as nintedanib and pirfenidone, are often limited by substantial side effects. Additionally, emerging therapies like JAK inhibitors, CAR-T cells, and the upcoming phosphodiesterase-4B inhibitor, nerandomilast, show promise, but no curative treatment exists to date. Future research could focus on multi-omics technologies and conducting multicenter clinical trials to establish therapeutic targets and advance precision medicine for RA-ILD.

PMID:41909710 | PMC:PMC13021622 | DOI:10.3389/fimmu.2026.1770348

Received — 26 March 2026 ⏭

Reasoning over Semantic IDs Enhances Generative Recommendation

✇cs.AI, q-bio.NC updates on arXiv.org
By: Yingzhi He · Yan Sun · Junfei Tan · Yuxin Chen · Xiaoyu Kong · Chunxu Shen · Xiang Wang · An Zhang · Tat-Seng Chua
25 March 2026 at 12:00
arXiv:2603.23183v1 Announce Type: cross Abstract: Recent advances in generative recommendation have leveraged pretrained LLMs by formulating sequential recommendation as autoregressive generation over a unified token space comprising language tokens and itemic identifiers, where each item is represented by a compact sequence of discrete tokens, namely Semantic IDs (SIDs). This SID-based formulation enables efficient decoding over large-scale item corpora and provides a natural interface for LLM-based recommenders to leverage rich world knowledge. Meanwhile, breakthroughs in LLM reasoning motivate reasoning-enhanced recommendation, yet effective reasoning over SIDs remains underexplored and challenging. Itemic tokens are not natively meaningful to LLMs; moreover, recommendation-oriented SID reasoning is hard to evaluate, making high-quality supervision scarce. To address these challenges, we propose SIDReasoner, a two-stage framework that elicits reasoning over SIDs by strengthening SID--language alignment to unlock transferable LLM reasoning, rather than relying on large amounts of recommendation-specific reasoning traces. Concretely, SIDReasoner first enhances SID-language alignment via multi-task training on an enriched SID-centered corpus synthesized by a stronger teacher model, grounding itemic tokens in diverse semantic and behavioral contexts. Building on this enhanced alignment, SIDReasoner further improves recommendation reasoning through outcome-driven reinforced optimization, which guides the model toward effective reasoning trajectories without requiring explicit reasoning annotations. Extensive experiments on three real-world datasets demonstrate the effectiveness of our reasoning-augmented SID-based generative recommendation. Beyond accuracy, the results highlight the broader potential of large reasoning models for generative recommendation, including improved interpretability and cross-domain generalization.
Received — 17 March 2026 ⏭

Efficient Reasoning with Balanced Thinking

✇cs.AI, q-bio.NC updates on arXiv.org
By: Yulin Li · Tengyao Tu · Li Ding · Junjie Wang · Huiling Zhen · Yixin Chen · Yong Li · Zhuotao Tian
16 March 2026 at 12:00
arXiv:2603.12372v1 Announce Type: new Abstract: Large Reasoning Models (LRMs) have shown remarkable reasoning capabilities, yet they often suffer from overthinking, expending redundant computational steps on simple problems, or underthinking, failing to explore sufficient reasoning paths despite inherent capabilities. These issues lead to inefficiencies and potential inaccuracies, limiting practical deployment in resource-constrained settings. Existing methods to mitigate overthinking, such as suppressing reflective keywords or adjusting reasoning length, may inadvertently induce underthinking, compromising accuracy. Therefore, we propose ReBalance, a training-free framework that achieves efficient reasoning with balanced thinking. ReBalance leverages confidence as a continuous indicator of reasoning dynamics, identifying overthinking through high confidence variance and underthinking via consistent overconfidence. By aggregating hidden states from a small-scale dataset into reasoning mode prototypes, we compute a steering vector to guide LRMs' reasoning trajectories. A dynamic control function modulates this vector's strength and direction based on real-time confidence, pruning redundancy during overthinking, and promoting exploration during underthinking. Extensive experiments conducted on four models ranging from 0.5B to 32B, and across nine benchmarks in math reasoning, general question answering, and coding tasks demonstrate that ReBalance effectively reduces output redundancy while improving accuracy, offering a general, training-free, and plug-and-play strategy for efficient and robust LRM deployment. Code is available at https://github.com/yu-lin-li/ReBalance .

CCMamba: Topologically-Informed Selective State-Space Networks on Combinatorial Complexes for Higher-Order Graph Learning

✇cs.AI, q-bio.NC updates on arXiv.org
By: Jiawen Chen · Qi Shao · Mingtong Zhou · Duxin Chen · Wenwu Yu
16 March 2026 at 12:00
arXiv:2601.20518v2 Announce Type: replace-cross Abstract: Topological deep learning has emerged as a powerful paradigm for modeling higher-order relational structures beyond pairwise interactions that standard graph neural networks fail to capture. While combinatorial complexes (CCs) offer a unified topological foundation for the higher-order graph learning, existing topological deep learning methods rely heavily on local message passing and attention mechanisms. These suffer from quadratic complexity and local neighborhood constraints, limiting their scalability and capacity for rank-aware, long-range dependency modeling. To overcome these challenges, we propose Combinatorial Complex Mamba (CCMamba), the first unified Mamba-based neural framework for learning on combinatorial complexes. CCMamba reformulates higher-order message passing as a selective state-space modeling problem by linearizing multi-rank incidence relations into structured, rank-aware sequences. This architecture enables adaptive, directional, and long-range information propagation in linear time bypassing the scalability bottlenecks of self-attention. Theoretically, we further establish that the expressive power of CCMamba is upper-bounded by the 1-dimensional combinatorial complex Weisfeiler-Lehman (1-CCWL) test. Extensive experiments across graph, hypergraph, and simplicial benchmarks demonstrate that CCMamba consistently outperforms existing methods while exhibiting superior scalability and remarkable robustness against over-smoothing in deep architectures.
Received — 14 March 2026 ⏭

UIS-Digger: Towards Comprehensive Research Agent Systems for Real-world Unindexed Information Seeking

✇cs.AI, q-bio.NC updates on arXiv.org
By: Chang Liu · Chuqiao Kuang · Tianyi Zhuang · Yuxin Cheng · Huichi Zhou · Xiaoguang Li · Lifeng Shang
10 March 2026 at 12:00
arXiv:2603.08117v1 Announce Type: new Abstract: Recent advancements in LLM-based information-seeking agents have achieved record-breaking performance on established benchmarks. However, these agents remain heavily reliant on search-engine-indexed knowledge, leaving a critical blind spot: Unindexed Information Seeking (UIS). This paper identifies and explores the UIS problem, where vital information is not captured by search engine crawlers, such as overlooked content, dynamic webpages, and embedded files. Despite its significance, UIS remains an underexplored challenge. To address this gap, we introduce UIS-QA, the first dedicated UIS benchmark, comprising 110 expert-annotated QA pairs. Notably, even state-of-the-art agents experience a drastic performance drop on UIS-QA (e.g., from 70.90 on GAIA and 46.70 on BrowseComp-zh to 24.55 on UIS-QA), underscoring the severity of the problem. To mitigate this, we propose UIS-Digger, a novel multi-agent framework that incorporates dual-mode browsing and enables simultaneous webpage searching and file parsing. With a relatively small $\sim$30B-parameter backbone LLM optimized using SFT and RFT training strategies, UIS-Digger sets a strong baseline at 27.27\%, outperforming systems integrating sophisticated LLMs such as O3 and GPT-4.1. This demonstrates the importance of proactive interaction with unindexed sources for effective and comprehensive information-seeking. Our work not only uncovers a fundamental limitation in current agent evaluation paradigms but also provides the first toolkit for advancing UIS research, defining a new and promising direction for robust information-seeking systems.
Received — 11 March 2026 ⏭

Mean Flow Policy with Instantaneous Velocity Constraint for One-step Action Generation

✇cs.AI, q-bio.NC updates on arXiv.org
By: Guojian Zhan · Letian Tao · Pengcheng Wang · Yixiao Wang · Yiheng Li · Yuxin Chen · Hongyang Li · Masayoshi Tomizuka · Shengbo Eben Li
10 March 2026 at 12:00
arXiv:2602.13810v2 Announce Type: replace-cross Abstract: Learning expressive and efficient policy functions is a promising direction in reinforcement learning (RL). While flow-based policies have recently proven effective in modeling complex action distributions with a fast deterministic sampling process, they still face a trade-off between expressiveness and computational burden, which is typically controlled by the number of flow steps. In this work, we propose mean velocity policy (MVP), a new generative policy function that models the mean velocity field to achieve the fastest one-step action generation. To ensure its high expressiveness, an instantaneous velocity constraint (IVC) is introduced on the mean velocity field during training. We theoretically prove that this design explicitly serves as a crucial boundary condition, thereby improving learning accuracy and enhancing policy expressiveness. Empirically, our MVP achieves state-of-the-art success rates across several challenging robotic manipulation tasks from Robomimic and OGBench. It also delivers substantial improvements in training and inference speed over existing flow-based policy baselines.

CRTAM inhibition mitigates toxicity of immune checkpoint inhibitors without antitumor efficacy trade-off

✇Nature Cancer
By: Si-Cong Ma · Zi-Xuan Rong · Zi-Peng Xu · Yan-Pei Zhang · Kui-Mao Zhuang · Hao Sun · Chuan-Hui Cao · Ze-Nan Wu · Hai-Peng Zhang · Qiang Zuo · Jia-Run Lin · Jia-Xin Cheng · Hua-Ting Qu · Duan-Duan Han · Wei Wei · Ke Liu · Xiao-Ting Cai · Ze-Qin Guo · Xue Bai · Li Liu · De-Hua Wu · Zhong-Yi Dong
5 March 2026 at 08:00

Nature Cancer, Published online: 05 March 2026; doi:10.1038/s43018-026-01135-0

Dong and colleagues report that blockade of T cell-expressed cytotoxic and regulatory T cell molecule results in selective mitigation of immune-related toxicities without affecting antitumor efficacy of immune checkpoint inhibitors.
Received — 5 March 2026 ⏭

Towards Self-Robust LLMs: Intrinsic Prompt Noise Resistance via CoIPO

✇cs.AI, q-bio.NC updates on arXiv.org
By: Xin Yang · Letian Li · Abudukelimu Wuerkaixi · Xuxin Cheng · Cao Liu · Ke Zeng · Xunliang Cai · Wenyuan Jiang
5 March 2026 at 13:00
arXiv:2603.03314v1 Announce Type: cross Abstract: Large language models (LLMs) have demonstrated remarkable and steadily improving performance across a wide range of tasks. However, LLM performance may be highly sensitive to prompt variations especially in scenarios with limited openness or strict output formatting requirements, indicating insufficient robustness. In real-world applications, user prompts provided to LLMs often contain imperfections, which may undermine the quality of the model's responses. To address this issue, previous work has primarily focused on preprocessing prompts, employing external tools or even LLMs to refine prompt formulations in advance. However, these approaches overlook the intrinsic robustness of LLMs, and their reliance on external components introduces additional computational overhead and uncertainty. In this work, we propose a Contrastive Learning-based Inverse Direct Preference Optimization (CoIPO) method that minimizes the discrepancy between the label-aligned logits produced by the model under a clean prompt and its noisy counterpart, and conduct a detailed analysis using mutual information theory. We augment the FLAN dataset by constructing paired prompts, each consisting of a clean prompt and its corresponding noisy version for training. Additionally, to evaluate the effectiveness, we develop NoisyPromptBench, a benchmark enhanced and derived from the existing PromptBench. Experimental results conducted on NoisyPromptBench demonstrate that our proposed method achieves a significant improvement in average accuracy over the current state-of-the-art approaches. The source code of CoIPO, pair-wise FLAN datasets, and NoisyPromptBench have already been released on https://github.com/vegetable-yx/CoIPO.
❌