luchak

Facet | San Francisco, CA | Software Engineer | Full-time | Onsite or Remote

Facet is harnessing machine intelligence to make photo editing quick, collaborative, and scalable. We understand that pictures are more than just pixels, so we're building a platform that lets creative professionals edit collections of photographs in terms of the people, objects, and colors they contain.

We're looking for engineers who can think deeply about complex systems and empathize with artists, and are also curious and creative. We use Typescript, React, WebGL, GraphQL, Postgres, Node, Python, and Tensorflow, but it's not a big deal if you haven't.

Our team includes folks from Google, Adobe, and Cruise, and we're backed by great VCs, former Dropbox execs, and AI researchers from Google Brain and Salesforce.

If you'd like to talk, I'm the CTO and you can drop me an email at matt at facet.ai.

Because, while it is a landmark achievement, it has little relevance to which people have access to health care, which people are confined to cages their entire lives, or which people are allowed to make decisions about their own bodies -- and, in the US, other stories are very relevant to these questions right now.

I would love for Hayabusa to get more attention, but the problem isn't that it's being drowned out by nothing. It's being drowned out by stories that will have real and long-lasting effects on people's everyday lives.

Facet | San Francisco, CA | Software Engineers | Full-time | Onsite

Facet is harnessing machine intelligence to make photo editing quick, collaborative, and scalable. We understand that pictures are more than just pixels, so we're building a platform that lets creative professionals edit whole collections of photographs in terms of the people, objects, and colors they contain.

To build this platform, we need engineers (like you?) who can think deeply about complex systems and empathize with artists, and are also curious, respectful, and creative. We use Typescript, React, WebGL, GraphQL, Postgres, Node, Python, Tensorflow, and Terraform, but it's not a big deal if you haven't.

We're seed funded -- backed by great VCs, former Dropbox execs, and AI researchers from Google Brain and Salesforce -- and building out our founding engineering team in SF.

If you'd like to talk, I'm the CTO and you can drop me an email at matt@facet.ai.

Yup. It also:

- Is a signal that the driver intends to turn. Drivers should also be using their actual turn signals, of course, but as a cyclist I would much rather have a driver who fails to signal merge early and turn, rather than turn with no warning. - Allows the cyclist to more easily get around to the left.

You might check out Mitsuba (http://www.mitsuba-renderer.org/) and LuxRender (http://www.luxrender.net/). Both are more modern open-source renderers with Python APIs, and I know that Mitsuba is quite popular in the graphics research community (in part because that's where it comes from).

Berkeley, CA - Contract - REMOTE - iOS front-end engineer

We’re bringing interactive cinema-quality physics to mobile devices. We need someone to help us build a slick, seamless iOS app to show off our awesome physics. If you want to push the frontiers of computer graphics, work with brilliant researchers, and build a game with the most beautiful interactive 3D liquids the world has ever seen: come talk to us!

Our team is three professors and a Ph.D. student, from CMU and Berkeley. We have a track record of producing great research, with six SIGGRAPH papers among us this year alone. This is a research project, not a company, but we do have a reasonable budget. If all goes well, we'll be presenting this project at SIGGRAPH next year.

If you write fantastic iOS apps, are interested in computer graphics, and (ideally) have some experience with video on iOS, you should get in touch!

Send an email to mlstanto@cs.cmu.edu with your resume and tell us a bit about why you’re interested.

In any case, running global illumination often causes a major increase in rendering time. So it's understandable that Pixar, which has to render a huge number of frames at huge resolutions, did not traditionally use it much.

There's also another factor at play, which is directability. Physical correctness is not usually a priority except as far as it advances the artistic goals of the people making the movie. If the director says, "can you make the right side of that table look less red?", you need to have some way for the artist to achieve that goal, even if that's not how the scene would "really" look. I expect that the development of new tools and processes to allow precise manipulation of the lighting in globally illuminated scenes was just as much, if not more, of a barrier than the additional cost in rendering time.

Secord et al. also did a perceptual study in which they determined which features were most correlated with user viewpoint preference and built models for viewpoint preference based on this information: http://gfx.cs.princeton.edu/pubs/Secord_2011_PMO/index.php

What would worry me is that I don't understand what Facebook really wants out of this acquisition, and why this will motivate it to do right by Parse users. It looks to me like a number of other people in this thread don't understand either and are coming up with their own theories, few of which are good news for the developer.

What the hell? I clicked through and I have no idea why you are so disgusted. As far as I can tell:

* This is on a site for expatriate women in Japan.

* This page belongs to their monthly series of profiles on various women living in Japan.

* A substantial part of her profile discusses how being pregnant and taking maternity leave affected her experiences at work.

* Nobody made you read it.

The pregnancy focus seems entirely relevant to me. And, come on, it's a major life event. Why shouldn't she be allowed to be excited about it?

If you sympathize with the author: do not post things like this. Listen to him. What makes you think you were the intended audience? What makes you think you know why he was writing? Do you see the irony in you trying to "educate" a black person on talking about racism?

If you actually want to help, first listen.

Don't fault people who have to put up with this shit, day in and day out, for not having limitless reserves of patience. His job isn't Racism Educator, so why should he have to act like it?

I mean, yeah, ideally everyone is infinitely calm and can call these things out in a measured way every time they happen. But you're asking a lot from the victims here in order to spare the feelings of the aggressors. (Perhaps unintentional aggressors, but aggressors just the same.) What you're suggesting isn't easy.

And what's to say that the person he calls out will take it well? Look at what happened at his workplace: they told him that the things he was complaining about were only jokes and that he was too sensitive. Look at what is happening in this thread: people are complaining about his tone, complaining about his word choice, and insinuating that he's making things up. That's not even going into the people who start out ostensibly agreeing before segueing into what sounds like their main point: how racist they think the author was being.

So give the tone argument a rest. It's one of the most reliable distractions that people fall back on to avoid discussing racism, sexism, and all other kinds of oppression, and it's already taken over way too much of this thread.

It looks to me like the robot just exhibits less motion in general. This might be because (as far as I can tell) the authors optimize only to match a set of static poses, not for physical or perceptual realism of motion.

Better than Apple can without Google? Maybe. Better than the existing Maps app? I use transit directions in Maps a lot and the transit apps I've downloaded very rarely.

I'm skeptical about this.

No, but your page still makes me play "hunt the noisy tab", and then, if I decide I want to watch the video, requires me to pause my own music to listen to your loop.

I don't object to all audio. Click-to-play and relevant? Great! But auto-playing, un-mutable, and completely unrelated is a terrible combination.

That's not really something I would want to try. Omitting references can lead to accusations that you are trying to claim as your own novel contributions things that others have already done.

I thought it was a little odd that he says the proper way to solve the problem is to have a higher-res screen, just before he says that we can't do super-sampling because it's too expensive.

It's a nice interface, but I'm a little confused about your column adjacency cost function. In particular, why sum only the bottom quartile of pixel distances to get column costs?

advantage leads to more enjoyment, which leads to more practice...etc

I actually agree with you here. This is why I think things like sexist/marginalizing presentations at conferences are a big deal: a hostile environment seems to me a lot more likely to achieve this than possible differences at the bare fringe of the aptitude spectrum.

Do you think there is some huge pool of people good enough for YC who just don't get in?

Yes, I believe that there are probably more than 300 people in the United States who could potentially be accepted to Y Combinator. I'd believe there are a lot more than 3000. We're talking about aptitude here, remember, so you have to consider people who could potentially get in, but don't apply. Do you really believe that the vast majority of the people who could even conceivably get in to YC, given some set of life choices, do in fact apply?

The only point I'm making is that small differences in underlying probability distributions can have large effects in the composition of people accepted into a highly selective program.

Yeah, if we're willing to grant you all kinds of crazy concessions, you can cobble together an explanation. But you're asking for a lot to make YC selective and sensitive enough for your example to be reasonable. The YC selection process has to be consistent and incredibly accurate. Most of the people who are even capable of getting in must apply. Acumen in various areas has to be minimally or negatively correlated, and to even be considered for YC, one person has to have all of the relevant skills.

And this is just to get close to the gender disparity that YC sees.

Sure, I agree that normal distributions with minor differences in variance are very different at the tails. But to get this to explain away the gender disparity in YC selectees? Epicycles on top of epicycles.

(Also: if this is really about the tails of the aptitude distribution, going by the paper you cited and your same arguments, we'd expect to see a lot more Asian women than Asian men in YC. I suspect this is not the case.)

If anything, I suspect your risk aversion point is closer to the mark, but that opens up an entirely new cultural and political can of worms.

The lengths to which this article goes to construct a model sympathetic to the author's views are incredible. It takes data showing a greater variance among males than females:

- in mathematical ability

- among schoolchildren

- on a standardized test

and generalizes them to:

- many different types of abilities

- among the Y Combinator applicant pool

- on the Y Combinator application process

Not only that, but by its end, the article is postulating a model in which literally one in a million people have sufficient aptitude to be accepted by Y Combinator - whether they're interested in it or not!

Even if you grant the author all of those assumptions, plus the risk aversion thing, you're only down to 13% women, three times what TechCrunch says Y Combinator actually accepts. I guess you could add another independent 99.9th percentile ability requirement, but then you're talking about one in a billion people being Y Combinator worthy. Or you could try to find a different, more tilted risk aversion statistic -- but at that point I think we'd be cherry-picking citations to fit a conclusion.

So, sure, if you accept a whole raft of dubious assumptions, you can explain the 4% acceptance rate by aptitude alone. But enough dubious assumptions can explain almost anything.

Have you seen this paper: http://grail.cs.washington.edu/projects/crowd-flows/ ? It touches on many of the same issues.

What problem are you trying to fix? If I'm offline, what good is being able to see my article list if I can't read the articles? If I'm online, why do I feel any less irritated when I try to go back to an article I was just reading only to find that I need to wait for it to download again?

As best I can tell, cyclist's opinions of me as a driver tend to be directly correlated to how far into my lane they ride.

If there are parked cars, you have to move into the lane or risk getting doored. Near intersections, you have to move into the lane or people will cut you off and turn right in front of you when you're going straight. If the road has no shoulder, you have to move into the lane to dissuade people from running you off the road by trying to pass with insufficient space (and to give yourself some margin in case someone tries). The same holds if the shoulder is full of debris. If you're descending a hill, you have to move into the lane so you can maneuver around turns.

Any safe cyclist will spend some time taking a lane of traffic.

I think you're confusing the extent of what is possible in your current type systems with what is theoretically possible with types.

Why can't a type system keep track of units? Why can't it be aware of precision? If it's possible for you to find a proof that a particular implementation of your algorithm won't fall into an infinite loop, then there is a type system that can verify that your algorithm will terminate. Of course, actually designing a sound type system with this property could be quite tricky, and I have no idea what the resulting language would look like -- but it would be possible.

I like the sound of that policy -- for some courses, it would work fine. For other, there might be problem:

* Some courses are hard enough, and many undergraduates are undisciplined enough, that a lot of students would end up failing. Of course, when this happens, the professor gets a whole bunch of reviews back telling him that he taught badly.

* Sometimes exams aren't the best setting for evaluation.

Perhaps other people are different, but I rarely learn important things or acquire deep understanding without doing a certain amount of "busy work".

It took me a semester of physics homework to actually understand vector calculus. I had to write philosophy papers in order to see the beautiful arguments in my head fall apart on the page. I tried to learn a language by immersion alone -- but that left gaps it took hours and hours with a textbook to fix.

Even points aren't a terrible idea: I'm very good at overestimating the depth of my understanding, and I often have to do a certain amount of grinding to uncover the deficiencies in my knowledge. During my undergraduate career, points (when properly applied) were very good at getting me to realize how important some of the pointless-seeming work was.

Good professors will probably give you some "busy work". Whether they're assigning the right work, and providing the right incentives, is another question.

Come on. You can take issue with the methodology of the study, but at least have a little basic respect.

Essentially the conclusion was that not everything can be explained through simply dividing by hours worked,

Which is exactly what you'd expect if there were a pay gap.

but there are a variety of possible explanations including both discrimination, career choice and work/family balance.

Fair enough, there are multiple causes. But if you look at the factors they account for (and which therefore fail to explain the missing 20 cents on the dollar, see pages 2 and 9), many of them are closely tied to work/family balance and career choices: experience, occupation, marital status, hours worked per year, etc. Few, if any, are proxies for discrimination or other social phenomena out of any particular woman's immediate control.

Okay, let's go with the GAO: http://www.gao.gov/products/GAO-04-35

Unless I'm horribly misreading the study, the claim a figure around 80 cents on the dollar after controlling for external factors, including hours worked, experience, and occupation. It's a lot worse before that.

But, somehow, as a consumer, I don't get any protection against Apple.

Hmm.