วีดีโอ

Wonderful!

great video.

fantastic course. thank you.

Uncanny

At minute 16 is priceless. The teacher is not sure if there has been any big advances in AI other then this small thing called GPT-3. Ha ha ha ha. Nothing big about that model other then it would turn out to be the biggest consumer app to 100 million users in history. And usher in the AI age to the general public. Dude, how did you miss that one…ouch.

This is such a helpful video thank you

Very timely tutorial

This is gold!!

Narrator: deep networks were NOT saturated lol

✅

✅

✅

✅

✅

✅

what a legend!

Thanks i was just searching for this idea

Could you elaborate on “… j ranges over neurons at that position …” ?

terrible exposition - doesn't seem to understand it himself either. "we should do the verification" even your notation is not clear. Also: "unary potential" is called a BIAS. Just read a stat. mech. book before making these videos, sigh.

After watched this video, I finally understand

Thanks for ur interpretation. I have a clear understading of how this equation works. (If could, I still need some detailed teaching on each part of equation. all in all, thanks for your help!!!

I understood when Ps = Pop, contrastive divergence goes to zero as distribution of Y_hat and Y are same. It's not clear to me why the gradient also goes to zero. Thank you in advance. PS: I took this course last quarter.

Another question, why is the gate function G(t) not just an independent parameter between 0 and 1? What do we gain from making it a function of h_t-1 and x? At the end, SGD would find good values for G(t) even if it were an independent parameter.

Is there an explanation for why the three gated RNN architectures here differ in performance? Why is the LSTM, the architecture with the most parameters, not the most effective one? In fact, neither is the simplest one the most effective. It's the intermediate one that takes the gold medal. But why?

I don't understand, what is the benefit of using a gated, i.e. residual, architecture? You talk about gates allowing forgetting or remembering, but why would we want to forget anyway? Also, whether G is zero or one, we always remember the previous state h_t-1 in some way! So I don't get it ...

On slide 11, shouldn't it be self.x.addgrad(self.x.grad*...) ? self.grad isn't defined, right?

Illuminating!

So if KL(p|q) = 0, does it mean that p = q upto a constant? Or are there other symmetries to take into account?

Thanks! I don't attend this institution, but this was an extremely clear lecture :)

The intuition around the 13 minute mark was really helpful! I've been trying to understand this paper for a few days now, and this has really made its goal and reasoning more succinct. Thanks!

Thank you for this video, however, please don't call your variable ŝ 😆 (or at least don't say it out loud)

Great lecture! May I ask how to choose the conditional probability of node i given its neighbors in a continuous case? Thanks!

Thanks for explaining that! Great job. Subscribed!

also, what if you skip the quantization during inference? would you still get images that make sense?

do you train the pixel cnn on the same data and just not update the Vae weights while training?

you are not even taking the examle of a graph with loops, which is the whole point of LBP. Moreover, please introduce a little bit the notations, we should be able to understand even if we did not watch your previous videos. Nevertheless, it's great to teach such method.

Now, the cutting edge scientists are working on the future AIs. Creating an AI by a combination of multiple AI's, which reportedly is similar to how our brain functions, which different portions performing specific functions, which can then understand and perform a multiple set of completely different tasks better than humans? What could go wrong?

I think that too many people are focusing on the game, which I also follow, as if this were an ordinary player. Since I have significant knowledge, and since I believe that Hawking and Musk were right, I am really anxious by the self-taught nature of this AI. This particular AI is not the worrisome thing, albeit it has obvious, potential applications in military logistics, military strategy, etc. The really scary part is how fast this was developed after AlphaGO debuted. We are not creeping up on the goal of human-level intelligence. We are likely to shoot past that goal amazingly soon without even realizing it, if things continue progressing as they have. The early, true AIs will also be narrow and not very competent or threatening, even if they become "superhuman" in intelligence. They will also be harmless, idiot savants at first. Upcoming Threat to Humanity. The scary thing is the fact that computer speed (and thereby, probably eventually AI intelligence) doubles about every year, and will likely double faster when super-intelligent AIs start designing chips, working with quantum computers as co-processors, etc. How fast will our AIs progress to such levels that they become indispensable -- while their utility makes hopeless any attempts to regulate them or retroactively impose restrictions on beings that are smarter than their designers? At first, they may have only base functions, like the reptilian portion of our brain. However, when will they act like Nile crocodiles and react to any threat with aggression? Ever gone skinny dipping with Nile crocodiles? I fear that very soon, before we realize it, we will all be doing the equivalent of skinny dipping with Nile crocodiles, because of how fast AIs will develop by the time that the children born today reach their teens or middle age. Like crocodiles that are raised by humans, AIs may like us for a while. I sure hope that lasts. As the announcer in Jeopardy said about a program that was probably not really an advanced AI long ago, I, for one, welcome our future, AI overlords.

5:50 The reference was meant to be Poole et el. rather than Chen et al.? arxiv.org/abs/1905.06922

Could you please provide reference for your statement at 11:18 *"cross-entropy objective is an upperbound on the population entropy"*