File tree Expand file tree Collapse file tree
Expand file tree Collapse file tree Original file line number Diff line number Diff line change @@ -284,7 +284,7 @@ <h2>研究方向</h2>
284284 < span class ="card-tag tag-training "> Agent Training Framework</ span >
285285 < h3 > Agent 训练框架</ h3 >
286286 < p class ="card-desc ">
287- 面向 LLM Agent 的强化学习训练基础设施。Agent-R1 将 MDP 框架延伸至多轮工具调用场景,构建端到端 RL 训练框架;Claw-R1 进一步将真实 Agent Runtime 与 RLVR 深度结合,实现「部署即训练」的新范式 。
287+ 面向 LLM Agent 的强化学习训练工具套件 。
288288 </ p >
289289 < div class ="card-meta ">
290290 < div class ="meta-row ">
@@ -315,7 +315,7 @@ <h3>Agent 训练框架</h3>
315315 < span class ="card-tag tag-runtime "> Agent Runtime Framework</ span >
316316 < h3 > Agent 推理框架</ h3 >
317317 < p class ="card-desc ">
318- LLM Agent 推理与运行基础设施:Science-Star 以 ReAct 引擎驱动单/多智能体科学任务,在 HLE / GAIA 等权威基准上取得领先结果 。
318+ 面向复杂科学任务的 LLM Agent 推理与运行基础设施。
319319 </ p >
320320 < div class ="card-meta ">
321321 < div class ="meta-row ">
@@ -345,7 +345,7 @@ <h3>Agent 推理框架</h3>
345345 < span class ="card-tag tag-rag "> Agentic RAG</ span >
346346 < h3 > Agentic RAG</ h3 >
347347 < p class ="card-desc ">
348- 在检索增强生成 (RAG)方向上的系列工作:在 KDD Cup 2024 CRAG 竞赛中获得银牌(Task 2&3 第二名);提出 Multi-Source Knowledge Pruning 方法(中稿 CIKM 2025);Mind2Report 以无训练工作流从海量网络信息合成专家级商业报告,配套 QRC-Eval 评测 。
348+ 检索增强生成 (RAG)与 Deep Research 方向的系列工作 。
349349 </ p >
350350 < div class ="card-meta ">
351351 < div class ="meta-row ">
@@ -375,7 +375,7 @@ <h3>Agentic RAG</h3>
375375 < span class ="card-tag tag-apps "> Agent Applications</ span >
376376 < h3 > Agent 应用</ h3 >
377377 < p class ="card-desc ">
378- 将 Agentic 方法落地于实际应用场景的三项研究工作:Cast-R1 探索时间序列预测中的工具增强顺序决策;TableMind++ 构建不确定性感知的表格推理 Programmatic Agent;PaperScout 实现面向学术文献检索的自主 Agent 。
378+ 将 Agentic 方法落地于时序预测、表格推理、学术检索等实际场景的系列工作 。
379379 </ p >
380380 < div class ="card-meta ">
381381 < div class ="meta-row ">
You can’t perform that action at this time.
0 commit comments