snrji

With which option! I have used Colab myself and I don't see live editing. I would like to see the cursor of the collaborators in real time while they edit. Also, the option to share the runtime.

Does anyone now if this system supports Google Docs-like live editing? I'm struggling to find such a service.

Not necessarily, it depends on the advisor, university, program and field.

The problem is that once you start you are committed for 4 years. Abandoning a PhD program and starting another one is quite more complicated than changing of job.

SpaceX always beats my expectations, unlike Tesla.

Hey! I didn't see your answer. That's very useful, thanks!

It's not only about the odds, but the expected return. That's the whole point of insurance, for instance. Having an accident under these circumstances is quite improbable, but in case it happened, the return would be massive. Actually, since it might be a matter of life or death, perhaps it doesn't even make sense to think about the expected utility, because it would be infinite.

For studying it would be fine, but for taking exams no way the instructors can let you in with your phone.

That depends on the language.

This is the kind of ad-hoc PEP proposal that could be completely avoidable if there was a generic modifier or initialization/declaration construction for declaring a variable as constant. So incredibly simple, so incredibly useful that it's hard to believe that it hasn't been accepted yet (it was proposed).

I see, thanks for warning me.

I do it all the time... on Twitter and Reddit.

In real life I would never even look at eyes of a stranger.

Probably some people here can relate.

When I was in college I was mystified by genetic algorithms, without knowing much about them. After taking 2 subjects on the matter and reading some books, I came to the conclusion that apart from being inherently inefficient (that's what you apply them when you have no alternative), they are actually outperformed by hill climbing (which can be seen as a particular case of the former if population = 1). Also, the crossover operator seems to make more harm than good, and it's not fully understood it's usefulness in nature, although there are some theories (this last point is taken from Pedro Domingos book).

AFAIK TensorFlow is not Tensorflow but "Matrixflow". This library is truly tensor-based if I understood it correctly?

I found useful the diagrammatic notation, by the way.

This headline is pure propaganda.

Since always. What is it otherwise?

I know, but isn't that what the OP was asking for? Isn't that an inherent tradeoff? Or: isn't that what Neuromorphic hardware and deep learning are about (ie. composing many relatively simple functions without complex branching/logic)?

Contrary to the belief of progressive billionaires, I think the reform of capitalism will come in form of local "soft socialism". Localism just makes sense and can be accepted by both left and right-wing people.

And why not the GPUs themselves? Which inherent limitations do they have? Aren't they essentially general purpose parallel devices?

Tangentially, is there any estimation of the amount of words/sentences that a kid may have heard by the time he or she learns to talk?

Pytorch is amazing and Facebook's open source contributions to AI are great.

However, tooling for deep learning in general is not ready for industry grade technology. Many bugs could be prevented by dependent types, but compilers are not there. Also, debugging models feels like alchemy and random changes until it works. In addition, in production systems, rigorous testing is not a standard. The closest thing I have heard of is Tesla's data engine and AI system, they do have unit tests and a shadow mode. Of course big companies will have similar technologies for critical systems, but it's not as standardized as testing in software engineering.

For linear models it can be shown to be equivalent to weight decay. For nonlinear ones, it empirically behaves as a regularizer.,

Thank you very much for your explanation! That's exactly what I was asking for.

By "eval" you can also mean the training subset. As I understood is at the code to evaluate the network at a given point with a given dataset. For instance, after epoch epoch, the model is evaluated for both training and validation (you see both losses)

As you said, the test subset should only be used at the very last.

People tend to do a hard distinction between symbolic AI and machine learning, but actually some machine learning algorithms are based on symbols (eg. decision trees and association rules build logical rules). I recommend Pedro Domingos book, The master algorithm, in which he describes the "5 machine learning tribes" (one of them is referred as "the symbolists") and advocates for a unification of different machine learning algorithms. He even proposes a particular instance of algorithm that would fulfill these criteria: Markov logic networks. He has developed an implementation, called Alchemy (https://alchemy.cs.washington.edu/).

If by symbolic AI we mean GOFAI, expert systems etc, I don't think that there will be ever a resurgence. But if by symbolic AI we mean machine learning algorithms that are somehow based on symbolic reasoning, I do think that there will be a resurgence. In particular, this resurgence will start when: a) Deep learning arrives to its limit (ie. research gets stuck) and/or b) Someone finds a scalable and SOTA-ish way to integrate symbols into gradient based algorithms.

It was not mainstream back then.

I agree that most popular dynamic are quite terrible. But, honestly, I think the real problem is not the particular implementations, but the whole idea of dynamic typing. At first it did make sense, but now that compiler writers have figured out "cheap" and general type inference, I don't see the point anymore.

However, I use Python on a daily basis because I have no decent alternative for the libraries I use.

The trend of adding type annotations to dynamically typed languages is now unstoppable. I wonder if some more exotic features (eg. side effects handling, monads or dependent types) will ever become mainstream in the feature.

I assume in both cases the dataset was balanced. Otherwise metrics such as accuracy can be meaningless.