weaklearner

long strolls are bliss

well the answer is a bit more complicated than just replacing dermatologists with a CNN. I am pretty convinced the better approach is something like the approach in this paper-use the CNN on easy cases-have the CNN tell a human which instances are hard to classify. Many images are easy to classify-but some are hard [even for the cnn] and humans should give those images more study.

https://arxiv.org/abs/1903.12220

that's a good point about the rigid surface.

eh some adversarial inputs are very robust to perturbations/orientation. see the famous turtle video.

frankly I am quite skeptical of the macro-economic predictions of a 'blockchain mark consultant'

of course not but the united states definitely tries to protect its nationals from what it views as inappropriate harassment and legal enforcement from foreign governments.

well I definitely do want my tax money being spent on this.

precisely.

less id checks-probably easy to bribe past any border guards-easier to get informal cash based transport.

If I was one of these people I'd find the cheapest bus I could find and get into cambodia asap. I would also wear a very large hat and sunglasses. Then I'd go asap to the us-cambodia embassay.

at least one is an american national. I'd support the united states sanctioning the ever living hell out of the Thai government if they executed an american national for building a buoy.

mmm I was pretty sure I read it. I cannot find the reference so I think you are right. I'm going to edit my comment.

I also think instagram ads are actually pretty close to normal posts [embedded effectively] so to me they are more effective. another reason to avoid being used by zuck

I don't have a citation here but I think facebook tried to create a slow vesting schedule for these aquihires so they would have golden shackles

I think you can act in democratic ways. Like 1. donating 20 dollars to susan collin's opponent for instance 2. voting against your senator

I think the bigger thing is that it is a TORRENT that overwhelms.

Yeah it seems that's true/I was wrong. I was reading george conway's twitter and I saw this tweet thread by a washingtonpost journalist `If this is the only subject to which the charges relate (& there are none on election interference or other topics), that’s notable. It also raises the question of what changed from when the Obama administration decided it couldn’t make a case, aside from political/DoJ leadership'

https://twitter.com/mattzap

It is bullshit PACER isn't free or significantly cheaper

well when obama was president assange was hiding in the embassy so I think attributing much significance to fact the obama administration did not issue this indictment is a bit tenuous.

If your researched is focused on Gans/adversarial stuff the specificity of working with goodfellow might out-weigh some other concerns. Also-hey-I'm not trying to make this tradeoff. I am just saying I can imagine someone else choosing to do so.

personally if I had offer from Google/MSR I'd take it over apple but I'd much rather work for apple than facebook even if yann le cunn told me I was cooler than sliced bread.

edit: also re-reading your comment I bet apple is willing to pay that crazy money. Goodfellow probably has compensation in 1-10 million range...

Thanks this book also looks good!

eh honestly a lot AI people might give up publishing for a little while to work with/around goodfellow.

Ian Goodfellow already has come up with the next great DL architecture. So its possible he will do so again.

edit: thinking about your question more-as an AI person I think the business logic behind hiring AI research talent if you are FAANG is not that they are going to deliver much business value in the next five years but they may deliver astronomical returns in the next 25 years. when you have the financial strength of google hiring these people is not a very large risk.

edit: I'm an ML grad student.

or read a random walk down wallstreet

Another data-point is that I literally just went to talk by a researcher from Toyota's self-driving car group and they 1) mentioned some OTA updates they made on their auto-pilot features in the last several months. 2) talked about their data collection efforts from production cars- order of petabytes/day

I don't know much about subaru but other companies are definitely integrating advanced driver assist features/updates. No idea how they compare to Tesla's and in frequency of improvements but this is definitely a thing other companies are working on.

That's true. Good point. Also I really like this book. It's just not really what I think of when I think of theory.

This is not a TCS textbook. This is core algorithms.

Source-TA ing an algorithms course with tardos