An Introduction to PyTorch versus TensorFlow for Deep Learning

Author(s): Tan Pengshi Alvin

Originally published on Towards AI.

A side-by-side comparison of PyTorch and TensorFlow for training and inference of neural networks

This member-only story is on us. Upgrade to access all of Medium.

Image by Lanju Fotografie on Unsplash

It is widely known that PyTorch and TensorFlow are the two most popular and established frameworks in the deep learning community. They are important for achieving two practical functions: they provide an easy-to-use yet customizable boilerplate for coding neural network architectures, and their boilerplate wraps around lower-level C++ codes, optimizing and speeding up computation, especially with GPU resources.

In the era without PyTorch and TensorFlow, neural networks for deep learning must be coded from scratch using the Numpy library, a common Python library for matrix/array operations. With the knowledge of tensor operations and neural network architectures, constructing neural networks can be done with Numpy, but this can be unnecessarily cumbersome and the computation is slow without GPU optimization. Nonetheless building deep learning models from scratch can help us appreciate and develop intuition about how neural networks work.

I have detailed the theoretical foundations and built the architecture of the vanilla neural networks — Multi-Layered Perceptron (MLP) — from scratch in Numpy in the below article, and I invite you to read it before we understand the higher API for constructing neural networks — PyTorch and TensorFlow… Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI