33< head >
44 < meta charset ="UTF-8 " />
55 < meta name ="viewport " content ="width=device-width, initial-scale=1.0 " />
6- < title > Agent 应用 — Cast-R1 · TableMind++ · PaperScout</ title >
7- < meta name ="description " content ="中国科大认知智能全国重点实验室在 Agent 应用领域的三项研究 :时间序列预测(Cast-R1)、表格推理(TableMind++)、学术文献检索(PaperScout)。 " />
6+ < title > Agent 应用 — Cast-R1 · TableMind++ · PaperScout · AnomaMind </ title >
7+ < meta name ="description " content ="中国科大认知智能全国重点实验室在 Agent 应用领域的四项研究 :时间序列预测(Cast-R1)、表格推理(TableMind++)、学术文献检索(PaperScout)、时间序列异常检测(AnomaMind )。 " />
88 < style >
99 * , * ::before , * ::after { box-sizing : border-box; margin : 0 ; padding : 0 ; }
1010 html { scroll-behavior : smooth; }
115115 .pn-1 { background : # eafaf1 ; color : # 1e8449 ; border : 1px solid # a9dfbf ; }
116116 .pn-2 { background : # eef4ff ; color : # 2471a3 ; border : 1px solid # b8d4f5 ; }
117117 .pn-3 { background : # fff0ee ; color : # c0392b ; border : 1px solid # f5c6c2 ; }
118+ .pn-4 { background : # f5eef8 ; color : # 7d3c98 ; border : 1px solid # d2b4de ; }
118119 .paper-card h3 { font-size : 1.2rem ; font-weight : 800 ; color : # 111 ; line-height : 1.4 ; margin-bottom : 8px ; }
119120 .paper-authors { font-size : 0.88rem ; color : # 777 ; margin-bottom : 12px ; line-height : 1.6 ; }
120121 .paper-abstract { font-size : 0.93rem ; color : # 444 ; line-height : 1.75 ; margin-bottom : 18px ; }
185186 < div class ="cat-badge "> Agent Applications</ div >
186187 < h1 > Agent < span class ="hl "> 应用</ span > 研究</ h1 >
187188 < p class ="subtitle ">
188- 将 Agentic 方法落地于实际场景的三项研究 :< br >
189- 时间序列预测 · 表格推理 · 学术文献检索
189+ 将 Agentic 方法落地于实际场景的四项研究 :< br >
190+ 时间序列预测 · 表格推理 · 学术文献检索 · 时间序列异常检测
190191 </ p >
191192 < p class ="affiliation "> 中国科学技术大学 · 认知智能全国重点实验室</ p >
192193 < div class ="btn-row ">
@@ -203,6 +204,7 @@ <h1>Agent <span class="hl">应用</span>研究</h1>
203204 < a href ="#cast-r1 "> Cast-R1</ a >
204205 < a href ="#tablemind "> TableMind++</ a >
205206 < a href ="#paperscout "> PaperScout</ a >
207+ < a href ="#anomamind "> AnomaMind</ a >
206208 </ div >
207209</ nav >
208210
@@ -214,11 +216,12 @@ <h1>Agent <span class="hl">应用</span>研究</h1>
214216 本方向将 Agentic RL 与工具增强方法应用于具体场景。
215217 < strong > Cast-R1</ strong > 将 RL 引入时间序列预测的顺序决策;
216218 < strong > TableMind++</ strong > 构建不确定性感知的表格推理 Agent;
217- < strong > PaperScout</ strong > 实现面向学术文献检索的自主 Agent,并提出过程感知的序列级策略优化方法。
219+ < strong > PaperScout</ strong > 实现面向学术文献检索的自主 Agent,并提出过程感知的序列级策略优化方法;
220+ < strong > AnomaMind</ strong > 将时间序列异常检测重构为工具增强的顺序决策过程,融合自反思与强化学习。
218221 </ div >
219222
220- < h2 class ="section-title "> 三项研究概览 </ h2 >
221- < div class ="card-grid ">
223+ < h2 class ="section-title "> 四项研究概览 </ h2 >
224+ < div class ="card-grid " style =" grid-template-columns: repeat(2, 1fr); " >
222225 < div class ="card ">
223226 < div class ="card-icon "> 📈</ div >
224227 < h4 > Cast-R1</ h4 >
@@ -234,6 +237,11 @@ <h4>TableMind++</h4>
234237 < h4 > PaperScout</ h4 >
235238 < p > 学术论文检索自主 Agent,提出 PS-SPO 策略优化。arXiv 2601.10029</ p >
236239 </ div >
240+ < div class ="card ">
241+ < div class ="card-icon "> 🔍</ div >
242+ < h4 > AnomaMind</ h4 >
243+ < p > 时间序列异常检测的 Agentic 框架,工具增强推理与自反思。arXiv 2602.13807</ p >
244+ </ div >
237245 </ div >
238246 </ div >
239247</ section >
@@ -268,6 +276,16 @@ <h3>Cast-R1: Learning Tool-Augmented Sequential Decision Policies for Time Serie
268276 < a class ="paper-btn pb-out " href ="../ "> ← 返回主页</ a >
269277 </ div >
270278 </ div >
279+
280+ < h2 class ="section-title " style ="margin-top: 48px; "> 引用</ h2 >
281+ < div class ="citation-box "> @misc{tao2026castr1,
282+ title={Cast-R1: Learning Tool-Augmented Sequential Decision Policies for Time Series Forecasting},
283+ author={Tao, Xiaoyu and Cheng, Mingyue and Jiang, Chuang and Gao, Tian and Zhang, Huanjian and Liu, Yaguo},
284+ year={2026},
285+ eprint={2602.13802},
286+ archivePrefix={arXiv},
287+ primaryClass={cs.LG}
288+ }</ div >
271289 </ div >
272290</ section >
273291
@@ -302,6 +320,16 @@ <h3>TableMind++: An Uncertainty-Aware Programmatic Agent for Tool-Augmented Tabl
302320 < a class ="paper-btn pb-out " href ="../ "> ← 返回主页</ a >
303321 </ div >
304322 </ div >
323+
324+ < h2 class ="section-title " style ="margin-top: 48px; "> 引用</ h2 >
325+ < div class ="citation-box "> @misc{cheng2026tablemind,
326+ title={TableMind++: An Uncertainty-Aware Programmatic Agent for Tool-Augmented Table Reasoning},
327+ author={Cheng, Mingyue and Yu, Shuo and Jiang, Chuang and Tao, Xiaoyu and Mao, Qingyang and Ouyang, Jie and Liu, Qi and Chen, Enhong},
328+ year={2026},
329+ eprint={2603.07528},
330+ archivePrefix={arXiv},
331+ primaryClass={cs.AI}
332+ }</ div >
305333 </ div >
306334</ section >
307335
@@ -338,26 +366,6 @@ <h3>PaperScout: An Autonomous Agent for Academic Paper Search with Process-Aware
338366 </ div >
339367
340368 < h2 class ="section-title " style ="margin-top: 48px; "> 引用</ h2 >
341- < h3 class ="sub-title "> Cast-R1</ h3 >
342- < div class ="citation-box "> @misc{tao2026castr1,
343- title={Cast-R1: Learning Tool-Augmented Sequential Decision Policies for Time Series Forecasting},
344- author={Tao, Xiaoyu and Cheng, Mingyue and Jiang, Chuang and Gao, Tian and Zhang, Huanjian and Liu, Yaguo},
345- year={2026},
346- eprint={2602.13802},
347- archivePrefix={arXiv},
348- primaryClass={cs.LG}
349- }</ div >
350-
351- < h3 class ="sub-title "> TableMind++</ h3 >
352- < div class ="citation-box "> @misc{cheng2026tablemind,
353- title={TableMind++: An Uncertainty-Aware Programmatic Agent for Tool-Augmented Table Reasoning},
354- author={Cheng, Mingyue and Yu, Shuo and Jiang, Chuang and Tao, Xiaoyu and Mao, Qingyang and Ouyang, Jie and Liu, Qi and Chen, Enhong},
355- year={2026},
356- eprint={2603.07528},
357- archivePrefix={arXiv},
358- primaryClass={cs.AI}
359- }</ div >
360-
361369 < h3 class ="sub-title "> PaperScout</ h3 >
362370 < div class ="citation-box "> @misc{pan2026paperscout,
363371 title={PaperScout: An Autonomous Agent for Academic Paper Search with Process-Aware Sequence-Level Policy Optimization},
@@ -370,6 +378,51 @@ <h3 class="sub-title">PaperScout</h3>
370378 </ div >
371379</ section >
372380
381+ <!-- ═══════════════════ ANOMAMIND ═══════════════════ -->
382+ < section id ="anomamind " class ="alt ">
383+ < div class ="container ">
384+ < h2 class ="section-title "> AnomaMind</ h2 >
385+ < div class ="paper-card " style ="background:#fdf8ff;border-color:#d2b4de; ">
386+ < span class ="paper-number pn-4 "> AnomaMind · arXiv 2602.13807</ span >
387+ < h3 > AnomaMind: Agentic Time Series Anomaly Detection with Tool-Augmented Reasoning</ h3 >
388+ < p class ="paper-authors ">
389+ Xiaoyu Tao · Yuchong Wu · Mingyue Cheng ·
390+ Ze Guo · Tian Gao
391+ < br > < em > 中国科学技术大学 · 认知智能全国重点实验室</ em >
392+ </ p >
393+ < div class ="paper-meta ">
394+ < span class ="paper-tag "> 时间序列异常检测</ span >
395+ < span class ="paper-tag "> 工具增强推理</ span >
396+ < span class ="paper-tag "> 顺序决策</ span >
397+ < span class ="paper-tag "> 自反思机制</ span >
398+ < span class ="paper-tag "> 强化学习</ span >
399+ </ div >
400+ < p class ="paper-abstract ">
401+ 时间序列异常检测在工业监控、金融风控等场景中至关重要,但传统判别式方法依赖静态特征提取,难以适应上下文依赖的多样异常模式。AnomaMind 将异常检测重构为顺序决策过程:Agent 通过多轮工具调用对可疑区间进行由粗到细的定位,并结合动态特征准备与多维证据推理做出诊断。
402+ </ p >
403+ < p class ="paper-abstract ">
404+ AnomaMind 的核心创新包括:< strong > 粗到细定位</ strong > 策略精确识别异常区间;< strong > 多轮工具交互</ strong > 实现自适应特征准备与动态分析;< strong > 自反思机制</ strong > 允许 Agent 迭代修正检测决策;以及结合通用模型与强化学习的< strong > 混合推理</ strong > 范式,实现任务专属的策略优化。
405+ </ p >
406+ < div class ="paper-links ">
407+ < a class ="paper-btn " style ="background:#7d3c98;color:#fff;border-color:#7d3c98; " href ="https://arxiv.org/abs/2602.13807 " target ="_blank " rel ="noopener "> arXiv 2602.13807</ a >
408+ < a class ="paper-btn pb-out " href ="../ "> ← 返回主页</ a >
409+ </ div >
410+ </ div >
411+
412+ < h2 class ="section-title " style ="margin-top: 48px; "> 引用</ h2 >
413+ < h3 class ="sub-title "> AnomaMind</ h3 >
414+ < div class ="citation-box "> @misc{tao2026anomamind,
415+ title={AnomaMind: Agentic Time Series Anomaly Detection with Tool-Augmented Reasoning},
416+ author={Tao, Xiaoyu and Wu, Yuchong and Cheng, Mingyue and Guo, Ze and Gao, Tian},
417+ year={2026},
418+ eprint={2602.13807},
419+ archivePrefix={arXiv},
420+ primaryClass={cs.LG}
421+ }</ div >
422+ </ div >
423+ </ section >
424+
425+
373426<!-- ═══════════════════ FOOTER ═══════════════════ -->
374427< footer >
375428 < div class ="container ">
0 commit comments