2025-04-21

来自cslt Wiki
跳转至: 导航搜索
People This Week Next Week Task Tracking (DeadLine)
Dong Wang
  • Refine the framework of AI text book of primary school.
  • Two talks for Tencent.


Lantian Li
  • Refresh audio retrieval proposal
  • Translate ML book
Ying Shi
  • thesis
Zhenghai You
  • Weekly report
  • Huawei project (add cosine loss, get slightly better results)[1]
Junming Yuan
  • Make Jiaoyubu's AI lesson PPTs with Qiang bro(3)
Xiaolou Li
  • Gongan Document Writing
  • Gongan PPT making
  • Making presentation system
Zehua Liu
  • Gongan Document Writing
  • Gongan PPT making
  • Making presentation system
Pengqi Li
  • developed Jiaoyubu's AI lesson PPTs with Qiang bro
Wan Lin
  • Supply experiments, almost finish, nothing strange
  • Start writing paper this week
Tianhao Wang
  • huawei huohua 2-6mix model training (almost done)
Xiaoxue Luo
  • Huawei huohua
    • Replace ground truth with CED topK labels(ground truth SDR:12.2182, CED topk SDR:7.4349)
Zhenyu Zhou
Junhui Chen
  • Exp for NS
  • Start writing paper this week
  • Reading paper
Jiaying Wang
  • Network+ materials preparation with wenqiang
  • Explore FunASR:need to register the speakers in advance,not very suitable
  • Replace the cnn in ASR part and validate the performance
Yu Zhang
  • Multi Agent Investment:
    • weight average result pretty promising (win rate 0.58, 2% annual excess return compare to baseline)
    • prompt praetor to generate detailed debate topic first.
Wenqiang Du
  • Jiaoyubu's AI lesson(6/10)with Junming and Pengqi
Yang Wei
  • Organizing experiment results for text enroll kws.
Turi
  • shared conference experience
  • Some thesis forms
Yue Gu
  • write the draft of introduction, complement some exps, draw method figure
Qi Qu
  • WebRTC frontend audio processing (AEC+NS+AGC) module: compiled as a C++ library for mr536, used to enhance experience of bidirectional calling.
  • Adjusting NPU clock frequency on mr536 to balance performance and heat.
  • Recompiling v851se kernels using glibc toolchains instead of musl to address a compatibility issue.
  • Trying to use KWS models (both audio- and text-enroll) to filter keyword submissions from domestic elderly products.