3 ms·
Why machine learning hyped if easily fooled?
I asked a similar question regarding deep learning, and one answer stated this problem is not unique to DL, but is common to many ML approaches.
https://news.ycombinator.com/item?id=14850096
Given the widespread adoption of ML, shouldn't this be of great concern? Yet I do not see much discussion of ML's shortcomings and being easy to manipulate. Seems an easy to exploit attack vector, could be the next buffer overflow equivalent.
- smt88 9y ago1. Hype is often about how something will be, not how it is today. Examples include self-driving cars, VR, and even the original iPhone (which didn't have third-party apps when it was released). 2. For some uses of ML/DL, it doesn't matter if it's easily fooled. There will either not be an opportunity to feed it tricky data, or it just isn't a problem if it makes some mistakes.