How do you use machine learning to control a drone without worrying it’ll crash or do something weird?
This group says that by separating out the deterministic behavior and using ML only on the stochastic component, they can give formal guarantees for safety and robustness.
They fuse learning and control theory to derive error bounds on their (learned) controllers, ensuring the drone’s trajectory remains within a specified set.
And they show a demo that the math works via a drone that skims the ground.
In addition to this continuous correction (keeping errors within a known ball in trajectory space) they do a discrete correction that forces learning to only occur subject to specific cleverly chosen invariants.
I love this kind of paper b/c the experiments show the math works.
They use this open source programmable drone[1] for their experiments, the Bitcraze Crazyflie 2.1. You can buy it online, then script a whole fleet with the associated Crazyswarm package [2].
- logical argument (list premises, derive lemmas, show graphs, drive to conclusion)
- listicle (thematically related bullets in any order)
- story (characters, plot, conflict, novelistic suspense)
What other structures do you find helpful?
The style of a piece is often implicit. It’s the kind of thing GPT-3 infers automatically.
But it can be useful to state explicitly. For example, you don’t want to write a research paper like a mystery novel. Put the conclusion in the headline, not after 200 pages of suspense.
Here’s another one that’s useful for academic talks:
- tell them what you are going to tell them
- tell them
- then tell them what you told them
Also, put the basic material up front so everyone gets something out of it, and the advanced material towards the end.
American anarchy is already becoming visible on the borders of empire. Western ideologies of chaos have eroded state capacity.
Not just recently, but for decades. What was neoconservatism after all but chaos? No plan but permanent war in the Middle East. wisdomofcrowds.live/the-coming-sto…
Just as the UK handed things off to the culturally adjacent US, the dark horse scenario is that democratic India and its diaspora need to become the global champion of liberal values, relative to China, if America descends into anarchy.
This is a possible scenario by ~2040: Centralized China vs International India. Surveillance state vs decentralized digital diaspora. Two rising powers, with very different cultures.
The US may be like the declining British Empire, whose dominance in 1913 was gone by 1945.
The conventional wisdom is that we are gearing up for a multi-decade conflict between the US and China.
But what if the US gets KO’d in the first round, by military defeat abroad and inflation at home, followed by civil conflict and/or a national divorce?
The various versions of the elephant graph show the phenomenon of the ascending and descending class.
Roughly speaking, most of the world along with the global elite has been economically gaining in recent years, while the Western middle class has not. brookings.edu/research/whats…
Sean McMeekin's latest tour-de-force makes a strong case that Stalin should be thought of as the main actor in WW2 — not Hitler or the US. amazon.com/dp/B08F6Z72Q4
One angle he goes into that I hadn't seen elsewhere is that the Soviets *wanted* the Japanese and Americans to fight.
Makes sense, given the Russo-Japanese War of 1905 and the 1939 Soviet/Japanese war in Manchuria! But not that widely discussed. thediplomat.com/2012/08/the-fo…
He unearths some amazing quotes, which of course make sense, but are a new angle on things.
Even under Lenin, the Soviets *wanted* the Japanese to fight the Americans, and the Euros to fight each other. Then they'd roll in and install communism. Which is kind of what happened...