Getting away from the “Hardware Lottery“ with Sara Hooker

In this Salon, Google Brain researcher Sara Hooker breaks down two of her recent papers. First, she and Charles discuss “The Hardware Lottery“ and how it perniciously shapes algorithms research and slows progress. Then, she presents her joint work with Chirag Agarwal on detecting outliers and difficult examples using the variance of the gradient as a proxy for network uncertainty. 🥇 “The Hardware Lottery“: 🤔 “Estimating Example Difficulty using Variance of Gradients“: https