argonaut

24 ms·

211.
by argonaut 8y ago
Almost nobody reads the entire owner's manual for their car, and even those who do read them skim over warnings like that.
212.
by argonaut 8y ago
Ignoring any one specific company for a moment, an investor consistently able to invest early and time their exits near the peak of each company's valuation would be a roaring success.
213.
by argonaut 8y ago
Other brokers don't want to hurt their margins. Etrade made 614 million in profit in 2017. 19% of revenue was from order commissions, or roughly 449 million. So Etrade could eliminate commissions and still remain profitable. https:&#x
214.
Tesla’s VP of Autopilot and chip guru Jim Keller is leaving(electrek.co)
20 points by argonaut 8y ago|7 comments
215.
by argonaut 8y ago
Recessions generally cause oil prices to fall.
216.
by argonaut 9y ago
When it comes to using super new research models or methods in deep learning, implementing it yourself instead of looking for other implementations is almost a requirement. In my experience the majority of online "implementations"
217.
by argonaut 9y ago
What do you mean by badly ? I'd guess that pretty much every batch has startups that implode very quickly due to disagreements within the team. The company is dissolved and the founders end up disliking each other. This is pretty ba
218.
by argonaut 9y ago
This "fatal flaw" works strongly in their favor. Good luck building a version of Facebook that billions of people would pay for.
219.
by argonaut 9y ago
Silicon Valley is not even close to that expensive. A 2 bedroom apartment in Pacific Heights, San Francisco's most expensive neighborhood, can be had for 6k in rent per month, which is $72k per year, which is still less than 80k.
220.
Replication study: deep learning for detection of diabetic retinopathy(arxiv.org)
2 points by argonaut 9y ago|0 comments
221.
by argonaut 9y ago
Well currently the attacks work 100% of the time.
222.
by argonaut 9y ago
Did you read your own article? It says: "Voter fraud, by any method, is still rare. A study by News21 — a consortium of journalism schools — found 867 cases since 2000 in which someone had admitted guilt or been convicted of a voter-fr
223.
by argonaut 9y ago
They are still very correlated with the other top crypto coins. In the most recent crash Stellar and Ripple plunged just the same.
224.
by argonaut 9y ago
This is a terrible deal. A 2% annual fee is a hedge-fund-esque fee. For 2% I'd expect an actively managed fund that gets in on ICO presales, carefully vets teams/technologies behind new coins and gets in early, has a strategy for
225.
by argonaut 9y ago
It's not an interesting question whether networks learn abstractions. It's almost tautological - an image classification network will by definition (attempt to) distill an image into a distribution over categories. So when people
226.
by argonaut 9y ago
Your own report contradicts your argument all over the place. "There is little doubt that the Durbin Amendment had a major effect on consumers — and would-be bank customers — as issuers have, in various ways, passed on some
227.
Reinforcement Learning never worked, and 'deep' only helped a bit(himanshusahni.github.io)
3 points by argonaut 9y ago|0 comments
228.
by argonaut 9y ago
It also doesn't work for adversarial attacks. "Adversarial training" helps somewhat but is still very susceptible to adversarial attacks.
229.
by argonaut 9y ago
Isn't the point of the article that you need less tooling with monorepos, not more (examples given: cross-repo changes, code searching, git bisect, dependency management). Now, you might be referring to the fact that Facebook and Goo
230.
by argonaut 9y ago
None of those things tell me Binance is safe. Binance is not even a year old! Low fees and the fact that their servers haven't crashed yet are hardly reassuring. Being based in China doesn't help me trust them either.
231.
by argonaut 9y ago
Also agree with ponyfleisch. Taxis as a foreigner in Thailand were always a gamble. 50/50 chance of a negative experience.
232.
by argonaut 9y ago
Except by your standards, Grab is a vast improvement over Google, an advertising company.
233.
by argonaut 9y ago
This seems pretty misleading, since honestly 99% of the machine learning that goes on here happens when running the questions/answers through Google Search. There are probably millions of man-years of machine learning / informatio
234.
by argonaut 9y ago
I am well aware of Keras' checkpoint code. It only supports checkpointing every epoch. So no, it does not work out of the box with Keras.
235.
by argonaut 9y ago
My point is, you would have to write a nontrivial amount of code to get this working. It would not work out of the box with Keras right now, for example.
236.
by argonaut 9y ago
Great! Depending on the framework you use this will not be trivial to implement correctly.
237.
by argonaut 9y ago
You're talking about support for checkpointing every epoch, but a 30 second shutoff would require checkpointing every N mini-batches, which is not a typical use for checkpointing.
238.
by argonaut 9y ago
The 30 second shut-off notice is pretty bad for deep learning training, IMO. You often will not be able to finish a mini-batch of training and save a model to disk in 30 seconds. There are probably ways to interrupt processing of a mini-bat
239.
by argonaut 9y ago
Although I would agree that GANs are part of machine learning, some people definitely do disagree, and their concerns are valid. It's definitely an area of open research. Your third point is actually the point of those who disagree. It
240.
by argonaut 9y ago
WaveNet2 is not a GAN, and describing it as a "variant of GANs" is like calling Python a variant of Java: it's highly misleading (WaveNet2 and GANs are both generative models, Python and Java are both programming languages).

More ›