❌

Normal view

Received β€” 27 January 2026 ⏭ Favourites

DeepEra: A Deep Evidence Reranking Agent for Scientific Retrieval-Augmented Generated Question Answering

βœ‡cs.AI, q-bio.NC updates on arXiv.org
By: Haotian Chen Β· Qingqing Long Β· Siyu Pu Β· Xiao Luo Β· Wei Ju Β· Meng Xiao Β· Yuanchun Zhou Β· Jianghua Zhao Β· Xuezhi Wang
26 January 2026 at 13:00
arXiv:2601.16478v1 Announce Type: cross Abstract: With the rapid growth of scientific literature, scientific question answering (SciQA) has become increasingly critical for exploring and utilizing scientific knowledge. Retrieval-Augmented Generation (RAG) enhances LLMs by incorporating knowledge from external sources, thereby providing credible evidence for scientific question answering. But existing retrieval and reranking methods remain vulnerable to passages that are semantically similar but logically irrelevant, often reducing factual reliability and amplifying hallucinations.To address this challenge, we propose a Deep Evidence Reranking Agent (DeepEra) that integrates step-by-step reasoning, enabling more precise evaluation of candidate passages beyond surface-level semantics. To support systematic evaluation, we construct SciRAG-SSLI (Scientific RAG - Semantically Similar but Logically Irrelevant), a large-scale dataset comprising about 300K SciQA instances across 10 subjects, constructed from 10M scientific corpus. The dataset combines naturally retrieved contexts with systematically generated distractors to test logical robustness and factual grounding. Comprehensive evaluations confirm that our approach achieves superior retrieval performance compared to leading rerankers. To our knowledge, this work is the first to comprehensively study and empirically validate innegligible SSLI issues in two-stage RAG frameworks.
Received β€” 28 October 2025 ⏭ Favourites

Mixture-of-Experts Meets In-Context Reinforcement Learning

βœ‡cs.AI, q-bio.NC updates on arXiv.org
By: Wenhao Wu Β· Fuhong Liu Β· Haoru Li Β· Zican Hu Β· Daoyi Dong Β· Chunlin Chen Β· Zhi Wang
28 October 2025 at 12:00
arXiv:2506.05426v2 Announce Type: replace-cross Abstract: In-context reinforcement learning (ICRL) has emerged as a promising paradigm for adapting RL agents to downstream tasks through prompt conditioning. However, two notable challenges remain in fully harnessing in-context learning within RL domains: the intrinsic multi-modality of the state-action-reward data and the diverse, heterogeneous nature of decision tasks. To tackle these challenges, we propose T2MIR (Token- and Task-wise MoE for In-context RL), an innovative framework that introduces architectural advances of mixture-of-experts (MoE) into transformer-based decision models. T2MIR substitutes the feedforward layer with two parallel layers: a token-wise MoE that captures distinct semantics of input tokens across multiple modalities, and a task-wise MoE that routes diverse tasks to specialized experts for managing a broad task distribution with alleviated gradient conflicts. To enhance task-wise routing, we introduce a contrastive learning method that maximizes the mutual information between the task and its router representation, enabling more precise capture of task-relevant information. The outputs of two MoE components are concatenated and fed into the next layer. Comprehensive experiments show that T2MIR significantly facilitates in-context learning capacity and outperforms various types of baselines. We bring the potential and promise of MoE to ICRL, offering a simple and scalable architectural enhancement to advance ICRL one step closer toward achievements in language and vision communities. Our code is available at https://github.com/NJU-RL/T2MIR.
Received β€” 24 January 2025 ⏭ Favourites

Multiple time points for detecting circulating tumor DNA to monitor the response to neoadjuvant therapy in breast cancer: a meta-analysis

βœ‡MRD
By: Shuyi Niu Β· Tie Sun Β· Mozhi Wang Β· Litong Yao Β· Tianyi He Β· Yusong Wang Β· Hengjun Zhang Β· Xiang Li Β· Yingying Xu
23 January 2025 at 19:00

BMC Cancer. 2025 Jan 22;25(1):115. doi: 10.1186/s12885-025-13526-0.

ABSTRACT

BACKGROUND: Not all breast cancer (BC) patients can benefit from neoadjuvant therapy (NAT). A poor response may result in patients missing the best opportunity for treatment, ultimately leading to a poor prognosis. Thus, to identify an effective predictor that can assess and predict patient response at early time points, we focused on circulating tumor DNA (ctDNA), which is a vital noninvasive liquid biopsy biomarker. We performed a meta-analysis to explore the predictive value of response by monitoring ctDNA at four time points of NAT using pathologic complete response (pCR) and residual cancer burden (RCB).

METHODS: By searching Embase, PubMed, the Cochrane Library, and the Web of Science until December 24, 2023, we selected studies concerning the relationship between ctDNA and response or prognosis. We analysed the results at the following various time points: baseline (T0), first cycle of NAT (T1), mid-treatment (MT), and end of NAT (EOT). pCR and RCB were used to evaluate the response as the primary endpoint. The secondary endpoint was to investigate the relationship between ctDNA and prognosis. Odds ratios (ORs) and hazard ratios (HRs) were used as effect indicators.

RESULTS: Thirteen reports from twelve studies were eligible for inclusion in this meta-analysis. The results demonstrated that ctDNA negativity was associated with pCR at T1 (OR = 0.34; 95% CI: 0.21-0.57), MT (OR = 0.35; 95% CI: 0.20-0.60), and EOT (OR = 0.38; 95% CI: 0.22-0.66). When RCB was used to evaluate responses, ctDNA negativity was associated with RCB-0/I at the MT (OR = 0.34; 95% CI: 0.21-0.55) and EOT (OR = 0.26; 95% CI: 0.15-0.46). Furthermore, ctDNA positivity at T1 predicted a worse prognosis for patients (HR = 2.73; 95% CI: 1.29-5.75). We also performed a subgroup analysis to more accurately assess the predictive value of ctDNA for triple-negative breast cancer.

CONCLUSIONS: Our meta-analysis suggested that the ctDNA status at the early stage of NAT can predict patient response, which provides evidence for adjusting personalized treatment strategies and improving patient survival.

PROSPERO REGISTRATION NUMBER: CRD42024496465.

PMID:39844103 | PMC:PMC11752932 | DOI:10.1186/s12885-025-13526-0

Received β€” 24 December 2024 ⏭ Favourites

Extracellular vesicle-mediated ferroptosis, pyroptosis, and necroptosis: potential clinical applications in cancer therapy

βœ‡Cell Death Discovery nature.com science feeds
By: Yi-Chi Yang Β· Qian Jiang Β· Ke-Ping Yang Β· Lingzhi Wang Β· Gautam Sethi Β· Zhaowu Ma
12 January 2024 at 08:00

Cell Death Discovery, Published online: 12 January 2024; doi:10.1038/s41420-024-01799-6

Extracellular vesicle-mediated ferroptosis, pyroptosis, and necroptosis: potential clinical applications in cancer therapy
There are no more articles
❌