I am not an expert, but the way that I understand self-driving systems is that there are multiple models running, and then those outputs are fused into yet another model which outputs the raw controls/actuations. In other words, I see this model/trainer as the "conductor", telling the car how it should approach an intersection, enter a highway, deal with merging traffic or construction zones, etc.
There is another model which interprets visual data to assist with lane-keeping, slow down or stop for pedestrians, inform the conductor of road signs... The final model combines all these inputs and incorporates the user preferences and then decides whether to brake or accelerate, how much to rotate the steering wheel.
Idk heh. The point of the high performance training is you can train the "conductor" role faster, and run inference faster. Assuming the car has limited compute/gpu resources, if you have a very high performance conductor function, you can dedicate that much more budget to visual/sensor inference and or any other models like the Trolley Problem decider (jk).
I am "the react guy" -- but before that I built knockout applications with requirejs & angular applications with coffeescript and gulp, and before that with a 2000 line long script.js & jquery ;)
I don't really disagree with anything you said, except to point out that a lot of the current crop of web devs have never experienced what it is to build an app without any abstraction such as React, and quite understandably have no idea what problems it is doing for them.
I do know that since hopping onto react with all of that ~baggage~ context, I have never wanted to program UIs with a different model. It is true that hooks introduce a layer of abstraction that is sometimes difficult to reason about, but IMO they boil down the problems we faced with class components/lifecycle/server rendering gotchas, and put them front and center - forcing you to confront and fix them rather than settling for a solution that works 99% of the time.
I thought the same thing - reminded me of birds. (I was casually aware that we knew they had some sort of magnetism sensing, but hadn't caught up with the fact that we think it is a visual thing)
where & how is the 'configuration' encoded? [we don't know, right] Is this why string theory requires higher dimensions? Also when I see the term 'harmonic oscillations', I think of plucking a string -- it seems so elegant for the physical particles to be manifestations of higher dimensional ~~vibrating~~ oscillating strings :) (thinking of flat-lander analogy)
gp's premise about learning with the multi-particle landscape first sort of blew my mind.
> Later, it optimizes the process to reward you earlier for just planning to do the thing - sketching, architecting, dreaming, writing task lists. Eventually, it gives you the same "I'm really proud of that accomplishment" reward for just watching someone on Youtube do it (you don't even have to look at all the off-camera drudge work).
I'm deep in the depths like OP and I can relate to this, like, a lot. Do you know of any literature or 'what to google for' regarding this sentiment/behavior?
I agree - it's great but... is there a hosted solution? If I just wanna run lint and unit tests on PRs for my shitty open-source project, Jenkins is overkill and not really viable.
Why was I not surprised to see that David Fravor (Navy Fighter Pilot, 'credible' UFO sighting) garnered the most views by a wide margin in the dataset? Sigh...
There is another model which interprets visual data to assist with lane-keeping, slow down or stop for pedestrians, inform the conductor of road signs... The final model combines all these inputs and incorporates the user preferences and then decides whether to brake or accelerate, how much to rotate the steering wheel.
Idk heh. The point of the high performance training is you can train the "conductor" role faster, and run inference faster. Assuming the car has limited compute/gpu resources, if you have a very high performance conductor function, you can dedicate that much more budget to visual/sensor inference and or any other models like the Trolley Problem decider (jk).
edit: grammar/details