“2025-04-07”版本间的差异

来自cslt Wiki
跳转至: 导航搜索
 
(9位用户的11个中间修订版本未显示)
第67行: 第67行:
 
|Xiaolou Li
 
|Xiaolou Li
 
||
 
||
*  
+
* Reproduction of ASR component in LipVoicer. Finetuning a new pre-trained Mandarin ASR system is challenging due to the complexity of integrating  diffusion steps (or diffusion steps embedding) into a new ASR architecture. Therefore, finetuning the existing checkpoint that was pretrained on English will be a more practical approach.
 
||
 
||
*  
+
* Finish reproduction in Wed.
 +
* Start to audit new 3000 hours data.
 
||
 
||
 
*
 
*
第78行: 第79行:
 
|Zehua Liu
 
|Zehua Liu
 
||
 
||
*
+
*Finish Lip2Speech model Training(Need Vocoder to reconstruct)
 +
*IS25 Rebutall
 
||
 
||
 
*
 
*
第89行: 第91行:
 
|Pengqi Li
 
|Pengqi Li
 
||
 
||
*  
+
* Some(3) PPTs development
 
||
 
||
 
*
 
*
第100行: 第102行:
 
|Wan Lin
 
|Wan Lin
 
||
 
||
*  
+
* Experiments[https://z1et6d3xtb.feishu.cn/docx/MxBNdPbLao0tsoxkBVCcUgUoneh?from=from_copylink]:
 +
** Performance on diff. encoder corresponds to the same encoder
 +
** NS 1layer performs similar to 8layer
 
||
 
||
 
*
 
*
第111行: 第115行:
 
|Tianhao Wang
 
|Tianhao Wang
 
||
 
||
*  
+
* Huawei project things
 
||
 
||
 
*
 
*
第122行: 第126行:
 
|Xiaoxue Luo
 
|Xiaoxue Luo
 
||
 
||
*  
+
* did some testing work of sound separation
 
||
 
||
 
*
 
*
第166行: 第170行:
 
|-
 
|-
 
|Yu Zhang
 
|Yu Zhang
 +
||
 +
Multi Agent Investment:
 +
long term investment seems useful (better than pervious short term only investment), but still not good than buy and hold baseline
 +
* Policy Level Problem:
 +
** feedback is portfolio level (about to choose which portfolio), need to design stock level feedback.
 +
** The current investment strategy group's strategy generation process is too simplistic and needs to incorporate additional reasoning pipelines.
 +
* Group Level Prombel:
 +
** weight average of different policy instead of choose one from them.
 
||
 
||
 
*  
 
*  
||
 
*
 
 
||
 
||
 
*
 
*
第178行: 第188行:
 
|Wenqiang Du
 
|Wenqiang Du
 
||
 
||
*  
+
* Check AI primary handbook's PPT
 +
* Jiaoyubu's AI lesson(4/10)with Junming and Pengqi
 +
* Some project cooperation about AIbabel
 
||
 
||
 
*
 
*
第189行: 第201行:
 
|Yang Wei
 
|Yang Wei
 
||
 
||
*  
+
* Test text enroll kws model with southern accent data [https://b30lttjm7l.feishu.cn/docx/N74UdZHhCoa6cPx4C5wccY4Hn9d?from=from_copylink]
 
||
 
||
 
*
 
*
第207行: 第219行:
 
|Yue Gu
 
|Yue Gu
 
||
 
||
*  
+
* move my code, model and training envs to new cloud service, reproduce some exps
 +
* read several papers and write the question analysis paragraph
 
||
 
||
 
*
 
*

2025年4月7日 (一) 10:59的最后版本

People This Week Next Week Task Tracking (DeadLine)
Dong Wang
Lantian Li
  • Submit AI Handbook (College Edition) (10/10)
Ying Shi
  • Huawei project cooperation
  • thesis
Zhenghai You
  • Some tests on the second iteration model of Huawei TSE project
  • Revised rebuttal
Junming Yuan
  • some works with brother Qiang:
    • Review primary school PPT
    • Make some PPTs
  • some AI practice handbook works
Xiaolou Li
  • Reproduction of ASR component in LipVoicer. Finetuning a new pre-trained Mandarin ASR system is challenging due to the complexity of integrating diffusion steps (or diffusion steps embedding) into a new ASR architecture. Therefore, finetuning the existing checkpoint that was pretrained on English will be a more practical approach.
  • Finish reproduction in Wed.
  • Start to audit new 3000 hours data.
Zehua Liu
  • Finish Lip2Speech model Training(Need Vocoder to reconstruct)
  • IS25 Rebutall
Pengqi Li
  • Some(3) PPTs development
Wan Lin
  • Experiments[1]:
    • Performance on diff. encoder corresponds to the same encoder
    • NS 1layer performs similar to 8layer
Tianhao Wang
  • Huawei project things
Xiaoxue Luo
  • did some testing work of sound separation
Zhenyu Zhou
Junhui Chen
  • Do some mid-term course homework things
  • read paper
Jiaying Wang
Yu Zhang

Multi Agent Investment: long term investment seems useful (better than pervious short term only investment), but still not good than buy and hold baseline

  • Policy Level Problem:
    • feedback is portfolio level (about to choose which portfolio), need to design stock level feedback.
    • The current investment strategy group's strategy generation process is too simplistic and needs to incorporate additional reasoning pipelines.
  • Group Level Prombel:
    • weight average of different policy instead of choose one from them.
Wenqiang Du
  • Check AI primary handbook's PPT
  • Jiaoyubu's AI lesson(4/10)with Junming and Pengqi
  • Some project cooperation about AIbabel
Yang Wei
  • Test text enroll kws model with southern accent data [2]
Turi
Yue Gu
  • move my code, model and training envs to new cloud service, reproduce some exps
  • read several papers and write the question analysis paragraph
Qi Qu
  • Reorganizing codebase for evaluating KWS models (audio-enroll and text-enroll).