❌

Normal view

Received — 27 May 2026 ⏭

OASES: Outcome-Aligned Search-Evaluation Co-Training for Agentic Search

✇cs.AI, q-bio.NC updates on arXiv.org
By: Erhan Zhang · Yiqun Chen · Zechun Niu · Wei Yang · Xiaochi Wei · Yan Gao · Yi Wu · Yao Hu · Jiaxin Mao
26 May 2026 at 12:00
arXiv:2604.03675v3 Announce Type: replace Abstract: Agentic search enables language models to solve knowledge-intensive tasks by adaptively acquiring external evidence over multiple steps. Reinforcement learning with verifiable rewards (RLVR) has emerged as a widely adopted training paradigm for search agents, yet outcome-only rewards are sparse and provide limited credit assignment for intermediate search actions. Existing process-reward methods therefore seek to densify supervision through proxy signals, external evaluators, or likelihood-based information gain. However, proxy rewards can deviate from the final outcome objective, while fixed evaluators can become stale as the search policy evolves, leading to unreliable process supervision. To address these challenges, we propose OASES, an Outcome-Aligned Search-Evaluation Supervision framework for agentic search. OASES derives outcome-aligned process rewards by evaluating how well each intermediate search state supports answering the original question. It further co-trains the search policy and the state evaluator on policy, allowing the evaluator to adapt to evolving search behavior and provide more reliable process rewards. Experiments on five multi-hop QA benchmarks show that OASES consistently outperforms strong RL baselines, with further analyses confirming the benefits of outcome-aligned process rewards and search-evaluation co-training.

Precision Thyroid Oncology: A Review of Multi-Omics Biomarkers and Spatiotemporal Technologies

✇MRD
By: Lei Huang · Xinghua Deng · Zhenjun Xi · Xiaofang Huan · Jingxin Mao · Xiaobing Li
25 May 2026 at 18:00

Int J Gen Med. 2026 May 18;19:602509. doi: 10.2147/IJGM.S602509. eCollection 2026.

ABSTRACT

Thyroid cancer (TC), the most prevalent endocrine malignancy worldwide, encompasses a broad spectrum of biological behaviors ranging from indolent microcarcinomas to lethal anaplastic variants. Despite advancements in standard care, critical clinical "bottlenecks" persist, including the diagnostic ambiguity of Bethesda III/IV nodules, the rising prevalence of radioiodine-refractory (RAI-R) differentiated TC, and the dismal survival rates of anaplastic thyroid carcinoma (ATC). The rapid evolution of biomarkers has catalyzed a paradigm shift from traditional anatomical-pathological staging to a sophisticated "Molecular Taxonomy" model, providing the cornerstone for precision oncology. This review systematically delineates the multi-dimensional landscape of TC biomarkers, encompassing genomic and transcriptomic drivers (eg, BRAF, RAS, TERT, RET, NTRK), epigenetic regulators (miRNAs, lncRNAs, circRNAs, and DNA methylation), and the proteomic interface. We highlight the transformative role of Liquid Biopsy 2.0-including ctDNA-based minimal residual disease (MRD) detection and exosomal multi-omics-in enabling non-invasive, longitudinal surveillance. Furthermore, we explore how cutting-edge technologies, such as single-cell sequencing and spatial transcriptomics, are deciphering intratumoral heterogeneity and redefining the "functional invasive front". Clinical translation is addressed through the lens of personalized management: from the use of genomic classifiers (eg, ThyroSeq v3) in preoperative triage to biomarker-guided "de-escalation" or "intensification" of therapy. Finally, we discuss the imperative of addressing ancestry-specific molecular divergence (specifically in Asian cohorts). However, significant challenges remain, including the high cost of multi-omics integration and the lack of standardized protocols for clinical implementation. We conclude by envisioning a future integrated with multimodal AI models, patient-derived organoids (PDOs), and metabolic reprogramming markers, aiming to provide a holistic framework for the "early screening-precise diagnosis-tailored therapy-dynamic monitoring" continuum in thyroid oncology.

PMID:42179850 | PMC:PMC13196814 | DOI:10.2147/IJGM.S602509

Received — 8 April 2026 ⏭

PRAISE: Prefix-Based Rollout Reuse in Agentic Search Training

✇cs.AI, q-bio.NC updates on arXiv.org
By: Erhan Zhang · Yiqun Chen · Zechun Niu · Wei Yang · Xiaochi Wei · Yan Gao · Yi Wu · Yao Hu · Jiaxin Mao
7 April 2026 at 12:00
arXiv:2604.03675v1 Announce Type: new Abstract: In agentic search, large language models (LLMs) are trained to perform multi-turn retrieval and reasoning for complex tasks such as multi-hop question answering (QA). However, current search-based Reinforcement Learning (RL) methods suffer from two core limitations: expensive long-horizon rollouts are under-utilized during training, and supervision is typically available only at the final answer, resulting in severe reward sparsity. We present Prefix-based Rollout reuse for Agentic search with Intermediate Step rEwards (PRAISE), a framework for improving both data efficiency and credit assignment in agentic search training. Given a complete search trajectory, PRAISE extracts prefix states at different search turns, elicits intermediate answers from them, and uses these prefixes both to construct additional training trajectories and to derive step-level rewards from performance differences across prefixes. Our method uses a single shared model for both search policy learning and prefix answer evaluation, enabling joint optimization without extra human annotations or a separate reward model. Experiments on multi-hop QA benchmarks show that PRAISE consistently improves performance over strong baselines.

Xpertbench: Expert Level Tasks with Rubrics-Based Evaluation

✇cs.AI, q-bio.NC updates on arXiv.org
By: Xue Liu · Xin Ma · Yuxin Ma · Yongchang Peng · Duo Wang · Zhoufutu Wen · Ge Zhang · Kaiyuan Zhang · Xinyu Chen · Tianci He · Jiani Hou · Liang Hu · Ziyun Huang · Yongzhe Hui · Jianpeng Jiao · Chennan Ju · Yingru Kong · Yiran Li · Mengyun Liu · Luyao Ma · Fei Ni · Yiqing Ni · Yueyan Qiu · Yanle Ren · Zilin Shi · Zaiyuan Wang · Wenjie Yue · Shiyu Zhang · Xinyi Zhang · Kaiwen Zhao · Zhenwei Zhu · Shanshan Wu · Qi Zhao · Wenhao Huang
7 April 2026 at 12:00
arXiv:2604.02368v3 Announce Type: replace Abstract: As Large Language Models (LLMs) exhibit plateauing performance on conventional benchmarks, a pivotal challenge persists: evaluating their proficiency in complex, open-ended tasks characterizing genuine expert-level cognition. Existing frameworks suffer from narrow domain coverage, reliance on generalist tasks, or self-evaluation biases. To bridge this gap, we present XpertBench, a high-fidelity benchmark engineered to assess LLMs across authentic professional domains. XpertBench consists of 1,346 meticulously curated tasks across 80 categories, spanning finance, healthcare, legal services, education, and dual-track research (STEM and Humanities). These tasks are derived from over 1,000 submissions by domain experts--including researchers from elite institutions and practitioners with extensive clinical or industrial experience--ensuring superior ecological validity. Each task uses detailed rubrics with mostly 15-40 weighted checkpoints to assess professional rigor. To facilitate scalable yet human-aligned assessment, we introduce ShotJudge, a novel evaluation paradigm that employs LLM judges calibrated with expert few-shot exemplars to mitigate self-rewarding biases. Our empirical evaluation of state-of-the-art LLMs reveals a pronounced performance ceiling: even leading models achieve a peak success rate of only ~66%, with a mean score around 55%. Models also exhibit domain-specific divergence, showing non-overlapping strengths in quantitative reasoning versus linguistic synthesis.. These findings underscore a significant "expert-gap" in current AI systems and establish XpertBench as a critical instrument for navigating the transition from general-purpose assistants to specialized professional collaborators.

Threshold Modulation for Online Test-Time Adaptation of Spiking Neural Networks

✇cs.AI, q-bio.NC updates on arXiv.org
By: Kejie Zhao · Wenjia Hua · Aiersi Tuerhong · Luziwei Leng · Yuxin Ma · Qinghai Guo
7 April 2026 at 12:00
arXiv:2505.05375v3 Announce Type: replace-cross Abstract: Recently, spiking neural networks (SNNs), deployed on neuromorphic chips, provide highly efficient solutions on edge devices in different scenarios. However, their ability to adapt to distribution shifts after deployment has become a crucial challenge. Online test-time adaptation (OTTA) offers a promising solution by enabling models to dynamically adjust to new data distributions without requiring source data or labeled target samples. Nevertheless, existing OTTA methods are largely designed for traditional artificial neural networks and are not well-suited for SNNs. To address this gap, we propose a low-power, neuromorphic chip-friendly online test-time adaptation framework, aiming to enhance model generalization under distribution shifts. The proposed approach is called Threshold Modulation (TM), which dynamically adjusts the firing threshold through neuronal dynamics-inspired normalization, being more compatible with neuromorphic hardware. Experimental results on benchmark datasets demonstrate the effectiveness of this method in improving the robustness of SNNs against distribution shifts while maintaining low computational cost. The proposed method offers a practical solution for online test-time adaptation of SNNs, providing inspiration for the design of future neuromorphic chips. The demo code is available at github.com/NneurotransmitterR/TM-OTTA-SNN.
Received — 11 March 2026 ⏭

Looking Back and Forth: Cross-Image Attention Calibration and Attentive Preference Learning for Multi-Image Hallucination Mitigation

✇cs.AI, q-bio.NC updates on arXiv.org
By: Xiaochen Yang · Hao Fang · Jiawei Kong · Yaoxin Mao · Bin Chen · Shu-Tao Xia
10 March 2026 at 12:00
arXiv:2603.07048v1 Announce Type: cross Abstract: Although large vision-language models (LVLMs) have demonstrated remarkable capabilities, they are prone to hallucinations in multi-image tasks. We attribute this issue to limitations in existing attention mechanisms and insufficient cross-image modeling. Inspired by this, we propose a structured hallucination mitigation framework involving Cross-Image Attention calibration and Preference Learning (CAPL). CAPL explicitly enhances inter-image interactions at the architectural level while reinforcing reliance on genuine cross-image evidence during training, thereby improving the model's perception and modeling of cross-image associations. Specifically, we (i) introduce a selectable image token interaction attention mechanism to establish fine-grained cross-image entity alignment and information flow; (ii) design a cross-image modeling-based preference optimization strategy that contrasts reasoning outcomes under full inter-image interaction and those obtained when images are mutually invisible, encouraging the model to ground its predictions in authentic visual evidence and mitigating erroneous inferences driven by textual priors. Experimental results demonstrate that CAPL consistently improves performance across multiple model architectures, achieving stable gains on both multi-image hallucination and general benchmarks. Notably, performance on single-image visual tasks remains stable or slightly improves, indicating strong generalization capability.
There are no more articles
❌