On-the-cheap Machine Learning, revisited

A short update on my On the Cheap Machine Learning blog post. After almost a year of use, I’m still pretty happy with the setup. The hardware has worked well. I haven’t done as much independent ML research as I had hoped, but I have contributed many hours of night-time GPU cycles to the Leela Zero open-source Go-game-AI project. I don’t think I would change anything about the build, and there’s nothing about it I want to upgrade yet.

However, in the past year a new option has appeared for on-the-cheap machine learning: Google’s Colaboratory project. Colaboratory is a free web-based IDE for writing machine learning applications. What’s especially cool about it is that comes with access to a cloud-based GPU. The GPU they provide is the NVIDIA K80, which is not the fastest GPU, but it’s still plenty fast for experimenting with machine learning. [Disclosure: I work for Google, but not in any groups related to Google Colaboratory.]

Colaboratory puts machine learning within the reach of anyone with a modern web browser, even if that browser is on a device (like a laptop, tablet, or even phone) that doesn’t have a high-end GPU.

I find myself using Colaboratory instead of my own desktop computer simply because my son’s often using my desktop computer for schoolwork, YouTube and playing games.

Comments

Popular posts from this blog

GLES Quake - a port of Quake to the Android platform

Using a Mac keyboard with Ubuntu Linux

A Taipei-Torrent postmortem: Writing a BitTorrent client in Go