About
Operationalizing Attack Research
Implementing the research
This site isn’t a traditional newsletter.
I’m taking a semi-formal approach in my explorations of adversarial research topics with emphasis on how to make them both usable and approachable. Despite author’s best efforts, too often their research is somewhat hidden behind esoteric formulae and off-margin LaTeX - that’s what these pages aim to address.
Ideally, readers without a technical background will find this useful. Conversely, researchers will ideally allow the liberties I’ve taken with various analogies and lossy shortcuts I’ve taken towards explanations of their work.
My thesis advisor explained it with something like this:
We want to take the lessons learned from the blue area, boil off what we can, and synthesize something we can use. Once the topics get to the red area, it’s going to be commoditized into major products and will become totally uninteresting to us.
And yes - I know what you’re thinking. Researchers often release proof-of-concept code, so what’s the point? Most importantly, their code is rarely updated or made to work outside their initial testing dataset. This leaves the uninitiated with a gap - how do we go from the whitepaper to something we can actually use? That’s what I’m exploring here. By the way, their work is cited in each respective article.
These pages are simply my thoughts and understanding of what researchers are doing, and put to the page mostly for my understanding. I hope you find it useful.
No paywalls
Paywalls suck.
Send research papers
There’s something like one ML/AI paper being released every second. Seriously. It’s impossible to separate the wheat from the chaff. So, see an interesting paper that’d be a good fit? Send it my way.
About
Drew Stelly - Computer Science PhD with a few published papers. Find the shortlist at https://cv.stelly.org.
Join
You’re welcome to subscribe via email or RSS. Associated code will be available on my GitHub at https://github.com/cdstelly/.
last updated Fall 2023