Something that I realized this year is that I like pondering the future -- looking way out and imagining how the world will be different, what things people will be doing, etc. This probably stems from my interest in technology.
They Did ItDecember 22, 2015
Whew, my anxiety has been soothed and replaced with excitement. They did it! They actually landed the rocket, perfectly.
Being an empathetic soul, I was touched to read of grown men crying at the sight of this accomplishment. One guy said he was sitting in O'Hare Airport crying as he watched the event on the TV there.
Perhaps my reflection here will be on the feeling of unity and encouragement. As a human being, the feeling of being united with our brothers and sisters, sharing joy and deep encouragement, is a profound experience.
SpaceX LaunchDecember 21, 2015
For people intrigued by giant things that blast off into space, tonight (or tomorrow, possibly) is a tense time. SpaceX is returning to flight after a six months hiatus following the rocket that exploded a minute into flight in June. Not only that, this will be their first attempt at landing the rocket back on land, and if it's a successful landing, it will be a historic day -- the first commercial rocket to lob something into orbit, and then return and land on land.
There is quite a bit of anxiety in the air.
People like me realize that, while not likely, the primary mission could be a failure, either due to the rocket exploding, or due to some secondary malfunction. Even if that succeeds, there's still quite a bit of anxiety about the landing. We all want, very badly, for the landing to be a success. And if all of that wasn't enough, there's anxiety about whether the rocket will launch tonight. Anything, such as the weather, or a technical glitch, or even something else, could postpone the launch. Even if it is postponed until tomorrow, it might be postponed again, even into January.
Anxiety, and lots of it.
The anxiety about the rocket exploding is higher than it would normally be for a return-to-flight, because SpaceX, for the first time ever, is flying a modified rocket. Not only is the rocket modified, but they are for the first time ever trying a launch with a super-super-cooled variety of fuel. Anyone with an inclination of an engineer will see changes like that as an invitation for unknown things to happen. And unknown things in the world of rockets, where everything has to happen 99.9% right (or kaboom) brings with it anxiety.
Another contributor to the anxiety is that there is an atmosphere of expectation around the landing being a success. Well, expectation is perhaps a bit strong. But consider that back in June, my internal feeling for a successful landing was still around 40%... I thought it more likely that the rocket would again crash into the drone ship and go kaboom. (but also knew that it could very well work) Today, however, my internal sense is about an 80% likelihood for successful landing. And anytime you start to feel that something is likely, but not certain, you start to feel anxious, because you've told your internal self "I expect it will work, but it might not". Isn't that the very definition of anxiety? (kind of?) No one likes their expectations dashed.
On that 80% figure -- why am I feeling more confident? Well, SpaceX has had 6 months to think long and hard about the landing. That's actually quite a bit of time to tweak things, testing things, etc. Secondly, it's on land, giving them a much wider margin for error, and when you're doing something this insanely difficult, a 50% or 100% or 200% wider margin for error is huge in terms of odds of success. Just imagine the rocket, at the last minute, computing "oh crap, trying to land within 40 feet of the intended target would require aggressive tilting that brings with it low odds of success... oh, but hey, my landing area is way bigger, so forget landing within 30 feet of my target, I'll just go with 100 feet from my target". Even if the rocket were to miss the concrete and land on the packed gravel, the odds of it staying upright are probably decent. Finally, I'm going on SpaceX's posture. They (Elon) are being so bold to invite people to watch the landing from the causeway, and my understanding is that, unlike the low-probability barge landings, they'll be live-streaming the landing. I suspect they wouldn't be live streaming it if the odds of success were still below 50%. I imagine their internal sense is that the odds are > 80%. Ok, one final thought on the odds of successful landing -- they have *already* numerous times successfully done soft touchdowns, they just happened to be on the water, rather than on the barge or land. That indicates that bringing the rocket down to a soft vertical landing is pretty "easy" at this point, and the real challenge is trying to get it within 30 feet, especially if that target is moving somewhat (the barge). Perhaps one of the big unknown variables is how far from the pad SpaceX has programmed their rocket to considering an acceptable landing spot. ie. If landing within 150 feet of the target is deemed low probability by the rocket in the final seconds, will it make a compromise and accept a landing spot that is perhaps 200 feet away from center? If they have that much flexibility, I might think the odds of success to be as high as 85-90% for the landing.
Right, so lots of anxiety floating around.
I'm going to get pessimistic for a moment: Part of me expects the odds of mission failure to be much higher than people might think. The reason for that is mainly around the modifications that SpaceX has made around fuel densification, etc. It is also due to the fact that SpaceX failed three times in a row when they were building the Falcon 1. What's the point? The point is that it is somewhat in SpaceX's DNA to fail spectacularly. They are pushing the envelope, and they're pushing it hard. And when you're doing things at the very edge of what's possible, rapidly iterating and experimenting, watch out -- kaboom is going to be more likely than as compared to the guys who are taking the slow conservative approach. I actually think SpaceX is going about things the right way, so long as they don't get
too carried away with their aggressive envelope pushing. Flying this fuel densified version of their rocket on return to flight -- is that too aggressive or is it reasonable? My gut is telling me it might be a bit too aggressive. If they're successful tonight with it, and successful the next few flights with it, then they will have gotten away with it. I truly hope that's the case. But if the primary mission fails, it is going to be an almost impossibly difficult blow to SpaceX and to Elon Musk. I do worry about his emotional state. Interviews he's given make it apparent just how difficult it was for the rocket to explode in June. What would a second consecutive mission failure do to his psyche?
Perhaps the three failures of Falcon 1, and the insane pressure of 2008, will be useful experience for Elon... now that his organizations are much, much larger, knowing how to chart such difficult waters may be a huge asset.
But let's hope my pessimism is unfounded. We're all hoping for a big party tonight. (or tomorrow)
Using AI to Optimize ParametersNovember 16, 2015
http://arxiv.org/pdf/1402.1694v4.pdfI came across this interesting paper today, which describes a way of doing Markov Chain Monte Carlo in a way that speeds up optimizing the parameters of a probabilistic model up to 200x. And of course, no big surprise, it's about using intelligent approximation.
This brings to mind to possibility that one day the algorithm we use to do optimization may actually use "AI". What I mean by that is that exploring a curve in hyperspace to find the minimum is in some sense like many other problems: You can look around and collect clues as to the characteristics of the curvature, and use those clues to build a kind of mental model of the dynamics at play, and then ultimately to use what you know about the space to intelligently explore it as quickly as you can.
Imagine a "neural net" of sorts that had been trained on millions of example optimization problems, and was able to very efficiently build an internal model of the topology of a hyper-surface, using that to calculate the optimum parameters... it might be hundreds, or thousands of times faster than naive human attempts that use exact algorithms... and for problems that are just mind-blowingly complex and viewed as semi-intractable today, it might be trillions of times faster.
An amusing side realization is that optimization is a core requirement for building AIs / neural nets in the first place, and so it's conceivable that there could be a recursive benefit to using AIs to do optimization... back to our good old exponential improvements in technology game: Your now supercharged ability to do optimization allows you to train an even more capable neural net, which allows an even faster optimization algorithm, which allows even more capable neural nets, and back and forth you go.
older >>
