laura_g

What is it specifically about the 1970/80s that causes this dip? Was there an explosion of this academic writing around that era or something else to have this effect?

There are a lot of things religion provides that modern society is really dismal at providing: community, care, socialization across class/career lines, relationships with people who live near you, a support system for sick/hurting people, and other benefits

The thing is, none of those are innately tied to religion. We've abandoned churches for what I feel are largely good reasons, but we haven't found something else to fill that void in community and care. We're more insular, more online, less of us know our neighbours or really have a stake in our community welfare in the same way. I don't think the solution is going back to religion.

I love this idea. I often find getting a big group together for a BBQ or something is great for mixing social groups, cross-pollinating friends, and seeing people I don't often see - the downside is I don't tend to get much 1 on 1 time to really talk with friends, and often people can't make a specific time (especially as we get older). I also have a fair amount of friends who are introverted or have social anxiety and don't feel comfortable in larger groups, and this feels like it would benefit that too.

I'm not sure why page count instead of word or character count is used - feels like there could easily be external factors such as page size or typeface that are effecting this.

This would be membership inference attacks - https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=7958568...

Because NeurIPS doesn't publish papers on stop sign removal yet :P

I completely agree, but that's a very big "if". I'm not terribly familiar with autonomous vehicle driving systems, but my passing understanding is that there are multiple components working together that help make predictions, and these systems do not rely on any single point of failure.

The classic example of a sticker on a stop sign is, in my view, more of a dramatization than a real threat surface. Designing an adversarial perturbation on a sticker that can cause misclassifications from particular angles and lighting conditions is possible, but that alone won't cause a vehicle to ignore traffic situations, pedestrians, and other contextual information.

Plus, if I wanted to trick a self driving vehicle into not stopping at an intersection, it would be much easier and cheaper for me to just take the stop sign down :)

The literature has pretty consistently shown that adversarial examples can be found with only black box access (even with truncated prediction vectors), robustness methods are primarily a cat-and-mouse game between attackers and defenders, and the existence of adversarial examples is likely inevitable (https://arxiv.org/pdf/1809.02104.pdf).

The big question that remains is - so what? There's exceedingly few use cases where the existence of adversarial examples causes a security threat. There's a lot of research value in understanding adversarial examples and what that tells us about how models learn, generalize, and retain information, but I am not convinced that these attacks pose a threat remotely close to the amount of attention given.