Microsoft releases Unadversarial examples: Designing objects for robust vision – a complete hands-on guide

In a recent work by Microsoft Research, a new framework is introduced which can address these problems of data models to create “unadversarial objects,” inputs that are optimized particularly for more robust model performance. This newly proposed approach for image recognition/classification methods helps in predicting better in the case of unforeseen corruptions or distribution shifts.

Read More

How the Dunning-Kruger effect can explain why your data science proposals don’t get buy-in

Consider the Dunning-Kruger effect to get your proposals taken seriously.
Many brilliant data science proposals never make it beyond the paper they’re written on. I’d like to start off by painting you a picture. Imagine you’re an experienced data scientist. You work for a small company and report into a team of directors who lead the company and are responsible for all the decisions made. Only proposals that get their buy-in can be implemented.

Read More