“2023-07-24”版本间的差异

来自cslt Wiki
跳转至: 导航搜索
 
(6位用户的7个中间修订版本未显示)
第31行: 第31行:
 
||  
 
||  
 
* School project support
 
* School project support
* Verify big model
+
* Verify big KWS model
 
** pretrian-20 epoch:th: 0.7 => 90.96% / 3 (ACC / FA)
 
** pretrian-20 epoch:th: 0.7 => 90.96% / 3 (ACC / FA)
** pretrain-20 epoch:  th: 0.7 => 89.86% / 5  (ACC / FA)
+
** pretrain-20 epoch fine-tune with batch_bce :  th: 0.7 => 89.86% / 5  (ACC / FA)
 
* Write paper
 
* Write paper
 
** test on Clean data set (VS clean baseline)
 
** test on Clean data set (VS clean baseline)
第63行: 第63行:
 
|Pengqi Li
 
|Pengqi Li
 
||   
 
||   
*
+
* Quantitative channel and content relationship[https://z1et6d3xtb.feishu.cn/docx/TjVRdGmq2oRQNmxXlRoclFrfngh]
 
||
 
||
 
*
 
*
第75行: 第75行:
 
|Wan Lin
 
|Wan Lin
 
||  
 
||  
*
+
* experiment [https://z1et6d3xtb.feishu.cn/docx/NbBLdn71NoMoLvxolPlcPHj2nEc] & paper [https://z1et6d3xtb.feishu.cn/docx/QT3kdtajRo9576xLgWPcpcnRnBk]
 
||
 
||
 
*
 
*
第86行: 第86行:
 
|Tianhao Wang
 
|Tianhao Wang
 
||  
 
||  
*
+
* huawei project [https://z1et6d3xtb.feishu.cn/docx/So4sd56lQokQJXxlwljcpoW8n2e]
 
||
 
||
 
*
 
*
第107行: 第107行:
 
|Jiaying Wang
 
|Jiaying Wang
 
||  
 
||  
*
+
* paper reading
 +
* Verification of L(B), L(D)[https://z1et6d3xtb.feishu.cn/docx/UipjdRHLpoGPK7xAj89cHIIknfg?theme=LIGHT&contentTheme=DARK]
 
||
 
||
 
*  
 
*  
第118行: 第119行:
 
|Xiaolou Li
 
|Xiaolou Li
 
||  
 
||  
*  
+
* CNC-AV code collating and Write public account tweets
 
||
 
||
 
*  
 
*  
第138行: 第139行:
 
|Wenqiang Du
 
|Wenqiang Du
 
||  
 
||  
*
+
* Data process
 +
* Training fewshot  model
 
||
 
||
 
*
 
*

2023年7月24日 (一) 10:58的最后版本

People This Week Next Week Task Tracking (DeadLine)
Dong Wang
  • Zhongdian Yanfa, material support
  • Two presentations prepared
  • Blingbling: general training method for KWS; speed up for MMeg net
Lantian Li
  • APT-group
    • Binary, BCE, WBA, DataAug+DANN, CNC-AV
  • HUAWEI VPR project
  • ICASSP review
Ying Shi
  • School project support
  • Verify big KWS model
    • pretrian-20 epoch:th: 0.7 => 90.96% / 3 (ACC / FA)
    • pretrain-20 epoch fine-tune with batch_bce : th: 0.7 => 89.86% / 5 (ACC / FA)
  • Write paper
    • test on Clean data set (VS clean baseline)
    • test on noisy data set (VS data augmentation)
    • test 3/4 mix (training on 2 mix)
    • extreme test
Chen Chen
  • Review
  • data pre-process pipeline
  • run VSR baseline
  • tsy [1]
  • prepare VSR/VTS baseline
Pengqi Li
  • Quantitative channel and content relationship[2]
Wan Lin
Tianhao Wang
  • huawei project [5]
Zhenyu Zhou
  • Paper writing[6]
  • Aug+Dann[7]
  • sota_experiment[8]
  • DA method project code(DANN,MMD,WBDA)
Jiaying Wang
  • paper reading
  • Verification of L(B), L(D)[9]
Xiaolou Li
  • CNC-AV code collating and Write public account tweets
Junhui Chen
  • Warm up experiments verifying[10]
  • ICCIP: uniform sample[11]
Wenqiang Du
  • Data process
  • Training fewshot model
Yang Wei
  • Huilan stuff