mrdrozdov

Thanks for pointing this out. The same acronym is used to refer to both versions. I like the description of Recursive Neural Nets versus Recurrent Neural Nets in this Stack Overflow post [1].

> A recursive network is just a generalization of a recurrent network. In a recurrent network the weights are shared (and dimensionality remains constant) along the length of the sequence because how would you deal with position-dependent weights when you encounter a sequence at test-time of different length to any you saw at train-time. In a recursive network the weights are shared (and dimensionality remains constant) at every node for the same reason.

In the TreeRNN (Tree Recursive Neural Nets) Github project and associated paper [2], there is indeed an implementation of a TreeLSTM which is a recursive LSTM inspired by the recurrent version.

[1]: http://stats.stackexchange.com/a/158995/92040

[2]: http://www.aclweb.org/anthology/P15-1150

Recursive Neural Networks are not the same as Recurrent Neural Networks (RNNs).

The paper presents improved techniques for training Generative Adversarial Networks (GANs). Code is published here: https://github.com/openai/improved-gan (uses TensorFlow, Theano, Lasgne)

Nice work team!

Publishers can blacklist ads, so unless they have tried to do so and failed then this is a way to shift blame from themselves on to a system that has been helping them all along.

What's a server room?

Apps have existed with subscriptions long before now. I suspect you will still be able to buy a flashlight app for a fixed price, but maybe you'll subscribe to Hulu through Apple rather than Hulu's backend.

Passive scans that hit your port 22 can hardly be called a security issue, and changing your port number definitely does not add any sort of security. This is a confusing concept, since changing your port number might decrease the probability your server gets taken advantage of (temporarily). It's a trivial fix for attackers everywhere to scan multiple ports instead of one at marginal cost.

Firing seems like a toxic move, especially without knowing what happened. I'd like to think that Amazon has multiple layers of alerts built into place for situations like this, so this was probably a complex bug (perhaps a perfect storm of faulty configurations like what happened with Google's Cloud offering not too long ago).

I'd say that number is probably a ceiling on their losses, and that the actual number should be much lower. I'm super curious if non-tech news organizations will pick up this story, in which case the biggest hit could be from the PR. That being said, it doesn't seem like such a bad time to go down.

If there is no spike of sales, all hours of sales is worth the same, and assuming Amazon makes 100B per year in online sales, then it seems like a considerable loss, but still a drop in the bucket when compared to their total revenue:

  (1 - (365 * 24 - 4) / (365 * 24)) = 0.000456621...
  0.000456621... * 100,000,000,000 = $45,662,100.4566...

My impression on DL papers is that they are relatively clear to other papers in machine learning/math/stats. Especially for big conferences, many of the accepted papers have code released on github before the conference date either by the original researchers or by developers looking to get stars (usually googling paper and + github will help you here). A lot of these implementations end up turning into blog posts or being contributed to Theano/TensorFlow/Torch/Caffe/etc.

I don't see a huge competitive advantage in not releasing code. It's probably more beneficial to share ideas and collaborate. I imagine the reason it's not done more often than it is is because it takes time and effort to put this stuff out there.

The only way to really stay caught up is to read the papers that are being released (NIPS, ICML, ICLR, CPVR, etc). If you use a course or guide, then you're behind by definition. If you're okay with being a little behind, then the Deep Learning courses for NLP or Computer Vision are both excellent. NYU also has similar courses. You can also try following some major Deep Learning people on Facebook/Twitter as they post about the new stuff daily. People to follow: Lecun, Lawrence, Bengio, Cho, Karpathy, just to name a few. Interested to hear others ideas on how to stay up to date.

My top reasons why everyone getting into maths/stats/cs should go straight for deep learning:

a. recent findings are documented incredibly well in both research and code

b. because of its success, there are many areas for useful contribution at relatively less effort from the researcher

c. because of its success, it'll help you develop marketable skills

d. it's fun

Maybe it won't solve General AI, but it seems like a damn good foundation for the person/people that will eventually come out with ideas that move us closer in that direction.

I sense a serious sample size problem here.

This doesn't exactly seem like the healthiest (or most efficient) routine.

> A pattern emerged in my days during this period. Get up, get washed and dressed, wash down a couple of caffeine tablets with some strong coffee, pick up a red bull, a coke and some cereal bars on my way into the office. Once at work I’d be at my desk non-stop; all meals would be eaten at my desk, though sometimes I’d not eat at all. When I did eat, it would usually mean a sandwich from the supermarket for lunch. Work would often provide dinner, invariably takeout. Sugary caffeinated drinks and sweets would sustain me for the rest of the time.

Working 80 hour weeks is not okay, but working 40 hour weeks is also probably not okay when you're relying on this sort of sustenance.

Thanks incredibly for sharing your experience and helping build such a renowned game. Hopefully people can take away from your experiences strategies to avoid a similar fate!

That's an ironic turn of events.

Maybe they've learned enough about science, history, literature, and the modern world to the point that they realize they can make a difference. I'm not sure how you measure impact, but it feels like they're succeeding.

I wish they had spent more time on scalable anonymous election systems, but I guess we'll get there eventually.

Sounds like it's still a good problem for us to keep working on then :)

Software that everyone in the world can use simultaneously. Sounds like a pretty good invention to me.

Thank you for your valuable input.

I'm confident you can get through an Algorithms and Data Structures course (which is the most difficult aka "weeder" course at most institutions) by memorizing a handful of sorting and search methods. You probably won't impress the Professor, but I'm sure you can pass.

OOP and other design patterns are also susceptible to memorization. I suspect that you're bringing this up because we have vastly different views of what difficult abstract thought is.

Here are two problems that I suspect are difficult to solve on programming experience alone.

1. How does Google display instant results (and recommendations) when you type into the search bar?

2. How would you build a Siri competitor, and what would be the biggest bottlenecks?

A problem that someone with minimal programming experience could probably solve would be.

1. How would you build me an inventory system which has an app, automated deploy, and unit tests?

2. Imagine you have only ever built Android apps with a NodeJS and Mongodb backend. How would you build an iOS app with a Python and MySQL backend?

How does Triplebyte make money? A lot of bootcamps place graduates as part of their process (and is how some of them make money), which is another factor to consider.

> I need someone who can look at a problem and realize that we can cut the amount of work we have to do by understanding programming concepts at an abstract level.

I think in many ways this is the exception not the rule. Many students certainly go through their CS degree without thinking this way, and some of these students go on to get phenomenal jobs because they spent a lot of time programming.

There are lots of good jobs for people who can program effectively. There are also lots of good jobs for people that can think abstractly about CS problems. These jobs don't always overlap.

You're right in that bootcamps are not a replacement for the traditional CS degree. There are things you can do in a 4 year research institution that are difficult to do in a 12-week program. It's probably more fair to compare a University program to a bootcamp + 3.9 years of working professionally.

It also almost unfair to call the CS degree traditional. Of the STEM degrees, it is probably the newest and least understood. Bootcamps are wonderful since they help people get awesome jobs, become amazing contributions to society, and in turn will continue to help the CS degree through its identity crisis.

AI and Machine Learning weren't mentioned. Neither were DevOps or systems-y/database-y types.

Periscope was acquired by Twitter.

"Twitter Confirms Periscope Acquisition, And Here’s How The Livestreaming App Works"

http://techcrunch.com/2015/03/13/how-periscope-works/

Driverless cars and state of the art engineering infrastructure.

Thanks for the useful link!

What's more interesting in the video is quantum computing being explained to a general audience. For young people looking for direction, both the video and the paper serve as persuasive pieces.

This video comes up frequently in social circles as for many people it was their first time having quantum computing explained to them (albeit simplistically). This paper seems to similarly explain briefly what quantum computing is and why it can be useful.

How does this compare to Justin Trudeau's explanation? https://m.youtube.com/watch?v=rRmv4uD2RQ4