2017 05 31

wangkuiyi

• Follow up https://github.com/PaddlePaddle/Paddle/pull/2217
• PyTorch survey: https://github.com/wangkuiyi/deeplearning/wiki/PyTorch
• Memory management survey: https://github.com/wangkuiyi/deeplearning/wiki/MemoryManagement
• Improve building system: https://github.com/PaddlePaddle/Paddle/pull/2326
• Errands: https://github.com/PaddlePaddle/Paddle/issues/2289

helinwang

• RecordIO C wrapper: https://github.com/PaddlePaddle/Paddle/pull/2294
• Fixes
  • cclient/test cmake remove hardcoded string: https://github.com/PaddlePaddle/Paddle/pull/2263
  • Client library of parameter server design doc update: https://github.com/PaddlePaddle/Paddle/pull/2277
  • move go folder to repo/go https://github.com/PaddlePaddle/Paddle/pull/2311
  • fix cclient/test build on mac https://github.com/PaddlePaddle/Paddle/pull/2329
• PR review:
  • Make generic.cmake no longer depend on util.cmake: https://github.com/PaddlePaddle/Paddle/pull/2326#pullrequestreview-41379344
  • File manager https://github.com/PaddlePaddle/cloud/pull/112#pullrequestreview-41037521
  • File manager https://github.com/PaddlePaddle/cloud/pull/112#pullrequestreview-41337145
  • Optimizer https://github.com/PaddlePaddle/Paddle/pull/2190#pullrequestreview-41372065

Yancey1989(yanxu)

• Paddle Cloud

  Start up beta stage in internal cluster.

  • Support v2 PaddlePadle Job: https://github.com/PaddlePaddle/cloud/pull/99/files
  • [WIP]Prepare demo trainer package:
    • https://github.com/PaddlePaddle/cloud/issues/110
    • https://github.com/PaddlePaddle/Paddle/pull/2320/files
  • Bugfixs:
    • Fix unsupport mulitple users: https://github.com/PaddlePaddle/cloud/pull/113/files
    • Fix invalid format for the active dc in configuration file: https://github.com/PaddlePaddle/cloud/pull/98

• Code Review:

  • File Manager: https://github.com/PaddlePaddle/cloud/pull/112

luotao

• change PaddlePaddle Book CC License: #322
• update the rnn-related info in module details of current implementation on PaddlePaddle

Gang Liao

• Add Paddle CI system documentation https://github.com/PaddlePaddle/Paddle/wiki/PaddlePaddle-CI-on-TeamCity

gongweibao

• Paddle Cloud
  • PR : https://github.com/PaddlePaddle/cloud/pull/112
    This PR implements ls rm mkdir cp commands.

Yu Yang

• Write a skeleton for refactorization brainstorm documentation.

  • Writing the current Paddle modules and design documentation.
    • Thanks @luotao for helping complete the computation of RNN part.
    • This documentation could be complete in today.
  • I will write a documentation about the registration implementation of Layers or Operations in many Neural Network frameworks, including how to implementation, what's the advantages and disadvantages. The comparison is including MXNet, PyTorch, Caffe2 and Tensorflow. It would be complete in this week.

• Discussing the major modules that a neural network framework should contains.

  • It should be written in the wiki page during next week.

fengjiayi

• survey mshadow, especially the part about expression template and lazy operation
  • https://github.com/dmlc/mshadow
• learn Paddle code of TensorEvaluate and TensorAssign

Liu Yiqun

• Support the native build of NVIDIA DRIVE PX2. It is split to two PRs
  • add architecture special modifications, using the aarch64 information: https://github.com/PaddlePaddle/Paddle/pull/2303
  • implement some scalar kernels (do not use simd extended instructions), and repair the organization of simd-related codes: https://github.com/PaddlePaddle/Paddle/pull/2299
• List the failed unit tests in two issues:
  • https://github.com/PaddlePaddle/Paddle/issues/2304
  • https://github.com/PaddlePaddle/Paddle/issues/2305
• Fixed typo of setting openblas compiling options: https://github.com/PaddlePaddle/Paddle/pull/2287

zhaolongxing

• pruning: https://github.com/PaddlePaddle/Paddle/issues/2284 the demo is in process
• Fixes
  • the resnet is inconsistent with a normal one: https://github.com/PaddlePaddle/book/pull/325

Caoying

• review
  • https://github.com/PaddlePaddle/models/pull/57#pullrequestreview-40436498
  • https://github.com/PaddlePaddle/models/pull/57#pullrequestreview-41126977
• Read some paper on ASR and ASR decoding.

wanghaoshuang

• test image classification model on gpu cluster.[WIP]
• mxnet(nnvm) survey for refactor of paddle.[WIP]

Dang Qingqing

• Using PaddlePaddle to evaluate traffic lights detection base on SSD method.

  • The mAP(mean Average Precision) seems ok.
  • This work will not go on after disscusing with Yongfeng.

• Developing row convolution operation which is used in DS2.

  • finish CPU version.
  • https://github.com/PaddlePaddle/Paddle/compare/develop...qingqing01:row_conv?expand=1

• PR Veiw

  • https://github.com/PaddlePaddle/models/pull/52
  • https://github.com/PaddlePaddle/book/pull/325

Shaoyong Xu

• 404 error page:
  • http://www.paddlepaddle.org/404.html
• audio data augmentation process
  • https://github.com/PaddlePaddle/Paddle/issues/2227

Xinghai Sun

• DeepSpeech2
  • Model debugging (e.g. pool convergence, unstable training of CTC layer).
  • Refactor audio data preprocessing (https://github.com/PaddlePaddle/models/pull/55/commits)
  • Inference part: prefix-beam-search (CTC-LM) @liuyibing @yangyaming @qingqing.

QiaoLongfei

• ADU traffic light detection. prepare the environment and run SSD training.
• Design for remote parameter updater using cclient (https://github.com/PaddlePaddle/Paddle/pull/2330)
• remove duplicate code (https://github.com/PaddlePaddle/Paddle/pull/2293)

Yibing Liu

• Read papers on Deep speech 2
• Implement CTC beam search decoder
  • https://github.com/PaddlePaddle/models/pull/59
• fixing bug about paddle.init() (c.f. https://github.com/PaddlePaddle/Paddle/issues/2031)

yangyaming

• Bugs
  • Find and fix bugs in smoothL1 and smoothL1Bp
    https://github.com/PaddlePaddle/Paddle/issues/2265
    https://github.com/PaddlePaddle/Paddle/pull/2267
• SSD Model
  • Refactor code
    • MultiBoxLossLayer 90%
    • DetectionOutputLayer 90%
    • DetectionMAPEvaluator 70%
• NTM with addressing mechanism Model
  • Add training and generation script
    
  • https://github.com/PaddlePaddle/models/pull/56

livc

Familiar with 手音 team hardware and software, discuss the plan today

hedaoyuan

• convolution
  https://github.com/hedaoyuan/Function
  https://github.com/PaddlePaddle/Paddle/pull/2282

• review
  https://github.com/PaddlePaddle/Paddle/pull/2299#pullrequestreview-40649103
  https://github.com/PaddlePaddle/Paddle/pull/2303#pullrequestreview-40648833

GuoSheng

• tool for converting models from Caffe to PaddlePaddle, and test the transformed VGG16 model, haven't got the exactly consistent model. PR: https://github.com/PaddlePaddle/models/pull/58

typhoonzero(wuyi)

• paddlecloud
  • paddlecloud client and API server issue fixes
    • https://github.com/PaddlePaddle/cloud/issues/94
    • https://github.com/PaddlePaddle/cloud/issues/89
    • https://github.com/PaddlePaddle/cloud/issues/86
  • pfs, new pserver PR reviews.
• code cleanup discussions
• MPI cluster v2
  • add new clusters

dongzhihong

• develop

  • fix optimizer lib integrate with new pserver https://github.com/PaddlePaddle/Paddle/pull/2190
  • disscusion the trainer library https://github.com/PaddlePaddle/Paddle/pull/2330/

• PR review:

  • pserver RPC@helin https://github.com/PaddlePaddle/Paddle/pull/2243
  • parameter client stub @helin https://github.com/PaddlePaddle/Paddle/pull/2172

juliecbd

• Working on v2 implementation of Learning to Learn by Google https://arxiv.org/abs/1606.04474
• Building Learning Bot code implementation in IDL in various environment