Py Torch An Imperative Style High Performance Deep

  • Slides: 20
Download presentation
Py. Torch: An Imperative Style, High. Performance Deep Learning Library Book Title : Advances

Py. Torch: An Imperative Style, High. Performance Deep Learning Library Book Title : Advances in Neural Information Processing Systems 32 (Pg: 8026 --8037) Conference : NIPS Year : 2019 Authors Count: 21 Presented by, Mahendran (CS 19 M 008)

Abstract Deep learning frameworks have often focused on either usability or speed, but not

Abstract Deep learning frameworks have often focused on either usability or speed, but not both. Py. Torch is a machine learning library that shows that these two goals are in fact compatible: it was designed from first principles to support an imperative and Pythonic programming style that supports code as a model, makes debugging easy and is consistent with other popular scientific computing libraries, while remaining efficient and supporting hardware accelerators such as GPUs.

Introduction Usually Py. Torch is used either as: ● a replacement for Num. Py

Introduction Usually Py. Torch is used either as: ● a replacement for Num. Py to use the power of GPUs. ● a deep learning research platform that provides maximum flexibility and speed.

Why Py. Torch? ● ● ● Regular Python program User has full control Compelling

Why Py. Torch? ● ● ● Regular Python program User has full control Compelling performance Most of code developed on C++ Researcher’s favorite Can use your favorite libraries and packages

High level overview of Py. Torch components https: //www. learnopencv. com/pytorch-c-front-end-tensors/

High level overview of Py. Torch components https: //www. learnopencv. com/pytorch-c-front-end-tensors/

Tensors

Tensors

Tensors Py. Torch provides Tensors that can live either on the CPU or the

Tensors Py. Torch provides Tensors that can live either on the CPU or the GPU, and accelerates the computation by a huge amount.

Tensor vs matrix

Tensor vs matrix

Work of Tensor ● ● ● Slicing Indexing Math operations Linear algebra Reductions

Work of Tensor ● ● ● Slicing Indexing Math operations Linear algebra Reductions

Mnist Dataset and dataloader Importing the dataset from torchvision library and dataloader helps in

Mnist Dataset and dataloader Importing the dataset from torchvision library and dataloader helps in putting the inputs in batches.

Data Visualization

Data Visualization

Loss and optimizer

Loss and optimizer

References 1. Py. Torch: An Imperative Style, High-Performance Deep Learning Library, 33 rd conference,

References 1. Py. Torch: An Imperative Style, High-Performance Deep Learning Library, 33 rd conference, NIPS(2019) 2. Example used (https: //github. com/mahendrannarayanan/Pytorch/blob/master/pytorch_2. 1. ipynb) 3. Pytorch Github (https: //github. com/pytorch) 4. Pytorch Website Documentation (https: //pytorch. org/) 5. https: //www. learnopencv. com/pytorch-c-front-end-tensors/