Group Meeting | 2025-10-13

Date Oct 13, 2025
Event Group meeting

4 papers were presented at this group meeting.

Presenters
@Yongsheng Lu @Zhiwei Huang @Yuhan Dong @Jinshan Guo

Papers and research themes

Papers presented at this group meeting
Paper and research focusPresenterPublicationResources

TS-Reasoner Aligning Time Series Foundation Models with LLM Reasoning

该论文提出 TS-REASONER 模型，旨在解决时间序列基础模型缺乏推理能力、大语言模型难解读时间依赖与数值的问题，通过 “预训练 TSFM + 预训练 LLM+TS-to-Text 适配器” 架构实现跨模态对齐。在基准测试中，不仅优于多种 LLM、VLM 及时间序列 LLM，还展现出高数据效率

@Yongsheng LuarXiv 2025-10Not available

Path planning techniques for unmanned aerial vehicles: A review, solutions, and challenges

This survey comprehensively reviews path planning techniques for Unmanned Aerial Vehicles (UAVs). It analyzes existing methods based on criteria such as path length, optimality, completeness, cost-efficiency, time-efficiency, energy-efficiency, robustness, and collision avoidance. The paper also discusses coverage, connectivity, and security aspects in UAV networks, identifies challenges, and suggests future research directions.

@Zhiwei HuangComputer Communications 2019-10Not available

Less is More: Improving LLM Alignment via Preference Data Selection

本文针对大语言模型对齐过程中的数据冗余问题，提出结合margin最大化与贝叶斯聚合(BeeS)的数据选择策略，在显著减少数据量的同时提升模型对齐性能与泛化能力。

@Yuhan DongarXiv 2025-06Not available

Optimizing Context-Based Location Extraction by Tuning Open-Source LLMs with RAG

本研究创新性地将检索增强生成（RAG）框架与开源大语言模型（LLM）相结合，旨在解决从新闻文本中精确提取复杂上下文地理位置的挑战；其核心创新点在于，该方法显著超越了传统命名实体识别和标准提示调优策略，在苏丹冲突数据集上实现了F1分数超过0.9的准确，为灾害监测、冲突分析等领域提供了高效、准确的自动化地理信息提取方案

@Jinshan GuoInternational Journal of Digital Earth 2025-07Not available
Group Meeting Literature Review