In today's lecture, we prove strong convex duality with a single picture. www.argmin.net/p/ends-in-a-...
A geometric argument for strong duality
In honor of the 10th anniversary of the iPhone 6. I wrote about technological stagnation in an era of proclaimed technological acceleration. www.argmin.net/p/where-did-...
The iphone6 and the end of history.
After a day of struggling, I converged on an introduction to Lagrangian duality that I find intuitive and satisfying. YMMV, of course. www.argmin.net/p/looking-ab...
I've finally found a satisfying derivation of Lagrangian duality.
What’s the most intuitive way to introduce optimization duality and why students should care? Here’s this morning’s (failed) attempt. www.argmin.net/p/duality
From Volume 3: The Subliminal Verses
Do you remember Libra?
After struggling to understand the concept for a quarter century, I finally realized that the technical term "overfitting" describes correlation, not causation.
machine learning and statistical prediction people, what are your favorite examples of overfitting?
By examining how we might prove optimality in convex programming, we dip our toes into duality theory. www.argmin.net/p/membership...
Slouching our way to duality theory
for the second question, we want replication so we can engineer. we want things that "work" so we can do stuff. and hence replication is only important if you want to do something with your science. you know, like care for sick people or build rockets...
for the first question, it's really worth examining recent history. in the social sciences, the recent replication crisis is the reckoning with the delusion that social science should model itself after natural science.