Four short links: 9 May 2019
Adversarial Examples, War Crimes, Open Source Firmware, and Better Questions
- Adversarial Examples Are Not Bugs, They Are Features — Adversarial vulnerability is a direct result of our models’ sensitivity to well-generalizing features in the data.
- Tech Companies Are Deleting Evidence of War Crimes (The Atlantic) — By piecing together information that becomes publicly accessible on social media and other sites, internet users can hold the perpetrators accountable—that is, unless algorithms developed by the tech giants expunge the evidence first. Facebook’s automatic content removal tech is removing evidence these investigators use to hold war criminals to account. We live in an age when software designed to get college students laid is critical to prosecuting war criminals.
- Why Open Source Firmware is Important for Security (Jessie Frazelle) — It’s counter-intuitive that the code that we have the least visibility into has the most privileges. This is what open source firmware is aiming to fix.
- Tukey, Design Thinking, and Better Questions (Roger Peng) — In my view, the most useful thing a data scientist can do is to devote serious effort towards improving the quality and sharpness of the question being asked. In my experience as well.