Pydata2017 11-29
20 likes5,488 views
Yuta Kashino is the CEO of BakFoo, Inc. and gave a presentation at PyData Tokyo about TensorFlow, PyTorch, and deep learning frameworks. He discussed features of TensorFlow like eager execution, debugging tools, and hardware support. He also covered PyTorch and compared it to TensorFlow and Chainer, noting its Pythonic APIs and define-by-run approach. He highlighted many popular deep learning models, libraries, and resources available for PyTorch.
Pydata2017 11-29
- 2. Yuta Kashino ( ) BakFoo, Inc. CEO Astro Physics /Observational Cosmology Zope / Python Realtime Data Platform for Enterprise Prototyping
- 3. arXiv PyCon2015 Python PyCon2016 PyCon2017 DNN PPL Edward @yutakashino Yuta Kashino ( )
- 5. … TensorFlow eager Dynamic: Define By Run Static: Define AND Run
- 6. TensorFlow eager Dynamic: Define By Run
- 7. …
- 8. TensorFlow - API : tf.contrib - : ecosystem(spark, hadoop, kubenates) - : tfdbg, tensorboard - : Feature Columns - : OpenMPI+ , Horovod(uber) - : Lite, Mobile - : XLA - : TPU - : GCP
- 9. : DL - - - - - AI - Google, NVIDIA, Intel, AMD, Apple, Huwei, Baidu, MS, FB… - Graphcore, Groq, DeePhi, BitMain…
- 10. DL - - - - … - - - AI …
- 13. PyTorch - Torch 7 Chainer - - Adam Paszke - Sam Gross - Soumith Chintala - NYU FBAIR - 2017 2 v0.1.6
- 14. PyTorch - 2017 2 v0.1.6 Google Trend
- 15. PyTorch - TF/Keras … Google Trend
- 18. - GPU(CUDA) - Define by Run: - Pythonic + python http://pytorch.org/tutorials/advanced/numpy_extensions_tutorial.html
- 19. autograd.Variable - - data, grad (, grad_fn) - chainer Variable - autograd.Variable Chainer fork - backward() data grad grad_fn
- 20. autograd.Variable data grad grad_fn x data grad grad_fn y data grad grad_fn z
- 21. autograd … - NLP DyNet 10 - Python addmm - ATen - v.0.3.0 autograd ATen https://twitter.com/kchonyc/status/899832376955744257 https://github.com/zdevito/aten https://goo.gl/Tujudr
- 22. https://github.com/ilkarman/DeepLearningFrameworks
- 26. GPU - CPU → GPU: Variable cuda() - GPU → CPU: cpu() chainer
- 27. Chainer - Chainer - GPU Chainer - Chainer - torch.optim = chainer.optimizers - … - pytorch/tnt = chainer.training - Pythonic … https://github.com/pytorch/tnt
- 28. Chainer
- 30. - PyTorch - PyTorch - Awesome-Pytorch-list, The Incredible PyTorch https://github.com/bharathgs/Awesome-pytorch-list https://github.com/ritchieng/the-incredible-pytorch
- 32. Capsule Nets - Hinton CapculeNets 2017 10 26 - 10 29 TensorFlow/Keras/PyTorch https://github.com/timomernick/pytorch-capsule https://github.com/gram-ai/capsule-networks https://arxiv.org/abs/1710.09829 https://github.com/naturomics/CapsNet-Tensorflow https://github.com/XifengGuo/CapsNet-Keras
- 33. / - - CNN/RNN: torch.nn, torchvision.models - GAN - https://github.com/Cadene/pretrained-models.pytorch https://github.com/szagoruyko/functional-zoo https://github.com/wiseodd/generative-models https://github.com/znxlwm/pytorch-generative-model-collections https://github.com/aaron-xichen/pytorch-playground https://github.com/sdhnshu/pytorch-model-zoo
- 34. DAG - PyTorch DAG - pyro - gpytorch - CUDA FFT pytorch_fft - sequence-to-sequence fairseq-py https://github.com/uber/pyro https://github.com/jrg365/gpytorch https://github.com/locuslab/pytorch_fft https://github.com/facebookresearch/fairseq-py
- 36. - PyTorchZeroToAll - Practical Deep Learning with PyTorch on Udemy - fast.ai https://www.udemy.com/practical-deep-learning-with-pytorch http://www.fast.ai/2017/09/08/introducing-pytorch-for-fastai/ https://goo.gl/kt3uV2
- 37. - conda
- 39. DL
- 41. BakFoo, Inc. NHK NMAPS: +
- 43. BakFoo, Inc.
- 44. BakFoo, Inc. : SNS +