Creativity isn't painting a photographic self-portrait. Creativity is, like, painting your own self over the course of your disease and so painting the disease.
>> ways to understand NN's are useful, if for nothing other than debugging, but I don't think the community agrees at all over whether explainability is necessary to use it.
I have to be able to debug anything I work with; debugging is necessary.
>> "the neural network parsing the LIDAR did not classify the stop sign correctly, so the driving algorithm did not stop"
It's important to look into why the stop sign was mis-classified, even if the classifier is a subcomponent.
I would go further: AI in a system that is in charge of safety critical system needs to be understood perfectly by its engineers. Right now, the community is trying to get by with shrugging and saying "dunno, but look at the reliability". In the long run this is not going to be good enough. We will need tools to dissect trained systems and build complete explanations of why it works (or doesn't).
It seems to me that understanding these deep learning models will be very tough since the reason why ml was necessary in the first place was the sheer complexity of the solution being too much for generations of programmers to solve
Understanding an existing solution is very different from trying to find it. We get taught things at school that seem simple and obvious to us now, but were extremely hard to discover or invent.
I see these training approaches as a tool that makes machines search the space of potential solutions for certain types of problems more efficient than humans. The next logical step is to try and understand these solutions and improve upon them.
Consider adversarial input. Right now we cannot tell which classes of adverserial inputs can exist for a given NN. We can only try to find representative examples. If you had a good enough understanding of how the NN works internally, you have a chance to derive the full set of adversarial inputs or - maybe -prove their absence.
It means that if you are using your phone while driving, you are at just as much increased risk of incurring an accident as if you had not slept the previous night. Right, folks?
Where I live, buzzwords probably a longer shelf-life than normal. Also, industry tends to become aware of buzzwords a few years later than everyone else so the dynamic is a little different there.
He is well-known for long, grammatically correct sentences and abundant footnotes. Some of it was fitting in as an academic?
I find some of the writing tough (i.e. impossible to read at night in bed) but the audio book for Infinite Jest narrated by Sean Pratt [1] is really nice. The narrator's speech is somehow easier to follow despite the number of clauses and whatever else in the writing.
The footnote thing tripped me up. Some time after reading Infinite Jest (and Gravity's Rainbow) I decided to take a shot at writing a novel, and I splattered it with footnotes. It was great fun, but it quickly becomes an annoying habit, the kind you know you don't want to do that but you constantly find yourself doing that anyway. (So breaking the habit was exceedingly annoying.)
Please read Freeman Dysons's "Disturbing The Universe".
This book is autobiographical and Dyson explains his arc of passion for nuclear propulsion and Orion.
His strongest statement in this book is some deep respect for a biological scientist who, after seeing declassified army training manuals on chemical and biological warfare, supposedly discouraged the entire western hemisphere from further develomepment.
This kind of nuclear research is, thankfully, over.
>This kind of nuclear research is, thankfully, over.
Nuclear pulse propulsion is the only currently viable technology that could be used to make humans an interstellar species. It would also allow us to practically ship up enough materials to build self-sustaining habitats in near space. It is extremely unfortunate that this kind of research is over.
Fun fact: the background radiation levels introduced by nuclear propulsion would actually have a very slight positive health effect on humans according to more accurate radiation hormesis models, rather than the very small negative effects suggested by more naive no threshold models.
Radiation hormesis models aren't "more accurate." The long-term effects of small amounts of radiation are simply not well understood. It could be good, it could be bad, it could be neutral.
The far bigger problem for launching Orion from Earth is the electromagnetic pulse frying every satellite above the horizon and a bunch of stuff on the ground.
On the other hand if we'd gone all-in on nuclear pulse launchers, there would have been no incentive for companies like SpaceX and Blue Origin to develop reusable conventional rockets which will hopefully achieve the same result at lower environmental and perhaps also financial cost.
Even if you had effective nuclear pulse propulsion, you'd still need reusable chemical launchers to get them off the surface of Earth. Nuclear pulse engines are generally not something that you start up inside an atmosphere, for both safety and efficiency reasons.
How can chemical rockets with ~3km/s exhaust velocity ever achieve the same result? Also if we had gone all in on this concept maybe today SpaceX and BO would be working with a much more promising technology.
It depends how reusable nuclear launchers would be. A less efficient launch system in terms of ISP that uses thousands of times cheaper fuel and e.g. is 10 times more reusable might be very competitive.
I would rather classify that as "unfortunate". Yes there are obviously biological harms in radiation exposure but that is what science & engineering is for. We know the power of nuclear energy so in order to harvest it, the right thinking should have been like "How can we safely extract it so that it does harm people in the mix?" rather than "Oh its harmful to humans so obviously we should give up and not use it".
Current fission startups are working on versions of reactors which eliminate most of the harmful byproducts in the process.
I'm not sure all of the opposition to nuclear research is about dangerous byproducts from the typical situation(s). It is about the improbable problematic situation which lead to exponential disaster that are scary, and, even if improbable, inevitable...
It was a statistics class where I learned a linear time median algorithm. But that was pretty lucky, I admit.
95% of undergraduate statistics education is focused on formal inference. Data science, in my experience, involves a lot more exploratory data analysis [1] than formal inference (frequentist or Bayesian).
The extreme focus on inference and the hypothesis testing step in the scientific method is something people with a formal statistics education have to overcome to be productive data scientists. Or applied statisticians, really! It is more important to understand the data, organize it creatively, and find unexpected structure.
