AI skills are worth less than you think (2019)

But I wasn't talking about a poor model vs a good model. Like I said, going from 93-96% accuracy is generally not going to have a lot of value add.

>Do you have experience in industry? I would like to hear more from your perspective?

Yes and in my experience if your model has lift over random and has a positive roi vs doing nothing it's usually going to be worth building and implementing it. In the fields I've worked in the models are not in direct competition with models from other companies so if Co. A's model for some task is getting 95% accuracy and Co. B's is getting 91% there's not much cause for concern. If Co. B's model is still generating lift over a random guesser it's worth having in production.

For most consumers I doubt they could even tell the difference between service like that. If Walmart's product recommendation is 2% better than Amazon's I find it extremely unlikely most consumers could even tell and Amazon's primary concern is whether the model is driving increased sales and not whether it's stealing customers from Walmart(the model, not generally).

It's like someone saying "well, electrical engineering is really like playing with building blocks. Take zener diodes for example, the problem was that you can have a lot of power in a circuit, but if there's a power spike it might break. So they just...plugged in a piece that breaks by shunting the power spike into the ground, then reset. And this is now a major piece of electronics." Or "so, everyone is always going off about arabian mathematicians, but one "big development" they did was to invent the zero - just make up a symbol where previously you'd leave a space. It's basically just a change of notation!".

Deep learning theory (statistical, information theoretical and optimisation wise) is our process of understanding how to design systems that adapt themselves to feedback, and how to encode tasks in them. Batchnorm was inspired by one thing (internal covariate shift), and that thing was plausible, but as it turns out, in systems as complex (not complicated, complex as in interactions) as universal function approximators, adding one thing can radically change things. As it turns out, batchnorm smooths the function, it decouples parameter magnitude and directions and it positively improves signal propagation. How else would you have figured this out without having systems like neural networks with batchnorm already in place that you can study? And now there are lines of work emerging that do away with batchnorm, but have distilled the positive properties into smaller techniques (https://arxiv.org/pdf/2102.06171.pdf, Soham De gave a lecture at our lab recently).

Same thing with skip connections: Jürgen Schmidhuber will rightfully point out we've had highway networks since his heyday, but details matter. It is really not intuitive before you do it that in such a complex system, skip connections will be beneficial, because before we had them and started studying them on complex system, the ideas of thinking of them as learning small adjustments to a signal, or as an ensemble of shallow learners or the other perspectives that they have been studied under had not been developed.

And how would you? Without having them working really well, you'd have to start thinking about them from first principles in the giant design space of nonlinear, nonconvex functions, without being able to prove anything because we don't have the mathematical formalism yet.

Deep learning theory and nonconvex optimisation right now is a new physics born out of the marriage of information theory, computer science and computer engineering (and not surprisingly in a menage a trois, a lot of groundwork was laid by the french and other weird europeans /joke). We have a bunch of theory nerds trying to explain what we see in elegant and concise mathematical frameworks and trying to come up with testable predictions, and a bunch of experimentational people actually coming up with ways of testing it, gluing together the bits of understanding we have with soft knowledge to make the learning engine go brr and give feedback to the theorists on what held up, what didn't work predictable and what didn't go according to predictions. And people mouth off about the empirical nature of things.

Well, I ask: How else would you figure this stuff out? I think there is a cult of genius at play here, where if you don't start with category theory and platonic ideal conceptions of reality and derive your model without any experiment, you are somehow lesser.

Well, despite what people like to sell, disruption is a lie, everything is incremental, and without having the hackers make things work in clunky ways, the theorists would circle jerk themselves in creative dead ends because of a lack of stimulus.

And as always, there are a lot of people who make themselves sounds smarter by affecting superiority and disdain on this scientific process, while in the background nerds deepen our understanding of the universe.

It is pretty well understood.

Some of us are actually in it to scratch an intellectual itch, not for the money. I gave up a lucrative career as a software dev to study the subject I'm really interested in and took a huge pay cut for the privilege of working in the only environment that offers anything near the freedom to research whatever satisfies the little gray cells. The OP's turn of phrase about gatekepping sounds like a bitter joke to me. My field gets maybe one or two new entrants a year, folks like me who don't know what's good for them and can't tame their intellectual curiosity enough to get a real job. Most PhD students, most researchers of any level, don't want to gatekeep anything, we want to tell the world about our work. But most people don't give a shit, unless Google is interested. Unless you can grab my research and be an instant millionaire nobody cares.