moultano

21 ms·

211.
Castle in the Sky: Dynamic Sky Replacement and Harmonization in Videos(jiupinjia.github.io)
1 points by moultano 6y ago|0 comments
212.
by moultano 6y ago
The log function has a vertical asymptote as x approaches 0.
213.
by moultano 6y ago
Yes, that's the right likelihood function. To get probabilities distributions out of the products in the post, all you have to do is normalize them. That would be equivalent to applying bayes rule and combining independent evidence abo
214.
by moultano 6y ago
Yes! Also called the surprisal.
215.
by moultano 6y ago
Sure, but mixture models and mutually exclusive events are less common to work with than just "models." Obviously there are times when you add probabilities together, but usually you don't. Even when you're writing a mix
216.
Logs, Tails, Long Tails (2013)(moultano.wordpress.com)
214 points by moultano 6y ago|59 comments
217.
by moultano 6y ago
That's true. I left out saying that deeper networks represent a wider variety of functions more easily, because that seems generally intuitive to everyone. But the arguments about how easy it is to optimize them should apply equally we
218.
by moultano 6y ago
Let us know when polynomial regression succeeds at any machine learning task. A lot of people publish results that say deep learning is "just" something else, where the something else doesn't work.
219.
by moultano 6y ago
I'm curious which category you put this one in. :D
220.
by moultano 6y ago
People coming from the perspectives of general optimization believed that they were impossible to train, due to being very non-convex. People coming from the perspective of classical statistics believed that they couldn't generalize du
221.
by moultano 6y ago
Yeah I was thinking about that as I was writing and trying to convey why I feel like deep models are different. I think one way of thinking about it is that protein structure, even though it has lots of parameters, it is all happening withi
222.
by moultano 6y ago
With squared loss where it's easy for the loss to be zero, then yes, it will have lots of global minima, all at a loss of zero. For losses that asympotote, like logloss, they may have no minima.
223.
by moultano 6y ago
If that worked, people would use it, but it doesn't, so they don't.
224.
by moultano 6y ago
Thank you. Several people have pointed that out, and I'm probably not reading the right papers. Is it common when people introduce a new flavor of adaptive SGD to address how it handles saddles specifically? It is probably just a a mat
225.
by moultano 6y ago
This isn't true. A loss function that asymptotes can also have no minima, which commonly used loss functions do.
226.
by moultano 6y ago
That isn't what I'm arguing at all.
227.
by moultano 6y ago
Think of the model as a point in high d space that you're trying to trap inside a cage. (Corresponding to it being at a local minima and surrounded by higher points.) So you're trying to build a cube in n dimensions, which has 2n
228.
Why Deep Learning Works Even Though It Shouldn't(moultano.wordpress.com)
2 points by moultano 6y ago|0 comments
229.
by moultano 6y ago
I know a lot of folks that have rented victorian living rooms, sometimes even with a curtain to divide it in two. There just isn't that much else.
230.
by moultano 6y ago
That's actually what the title is, mostly I just wanted to write about clams and crabs. The person who submitted it changed the title in a way that got more attention than the original that I submitted it with. :)
231.
by moultano 6y ago
People still eat them, just illegally. The rangers have to bust people for it all the time. https://amp.sanluisobispo.com/news/local/environment/article...
232.
Abundance(moultano.wordpress.com)
3 points by moultano 6y ago|1 comments
233.
Cosmonaut Crashed into Earth 'Crying in Rage' (2011)(npr.org)
29 points by moultano 6y ago|16 comments
234.
by moultano 6y ago
I'd bet that a shallow transformer that was distilled on gpt2, possibly with quantized weights, would easily win. GPT2 isn't optimized for size at all, but there are tons of ways to do so.
235.
by moultano 6y ago
Given the replication crisis, these are the kind of results that I reflexively don't believe anymore. Has anyone read any of the studies? What are n and p for these results? What's the effect size?
236.
by moultano 6y ago
It was obviously not the status quo, because they were banned from the country. Companies should advocate for the rights of their employees. They should demand their employees treat each other with respect. Both of those stances will often
237.
by moultano 6y ago
Having the time and energy to devote to it is not the same thing as whether it materially changes your life. Often those things are anticorrelated.
238.
by moultano 6y ago
An Iranian immigrant can't help but advocate for their political views simply by existing. Living and working in the US is an assertion that they should have the legal right to do so. Simply by being glad to have them as a coworker (
239.
by moultano 6y ago
In January 2017, several of my coworkers were banned from reentering the country to go home to their families, while they were on business trips to visit other offices. How is anyone supposed to be apolitical in that context? How are the pe
240.
'The Social Dilemma' and the Last Fucking Thing I'll Ever Write About Facebook(pullrequest.substack.com)
2 points by moultano 6y ago|0 comments

More ›