qiemem

I think the author meant multi-label.

GP for NNs has, but this isn't GP; "self-reproduction" (as in the original title) here means that the neural network is learning to output a copy of itself. In other words, the NN is learning to be a quine (as indicated by the current title). This is a very different than just applying genetic programming to NNs. In GP, you're updating the NN (or whatever) by copying it with mutations/recombinations. Here, the NN itself is learning to create a copy of itself.

The fact that they manage to also train to the network to simultaneously perform somewhat complicated tasks is super crazy.

Mind sharing your dx7 port? I'd love to check it out!

> One of the strengths that's getting overlooked here is that crowd-sourced news seems to be more reliable than news from any one source. Everyone is biased, but if enough people are talking about something the bias tends to cancel out.

The problem with this is that people only have time to check a limited number of sources. Furthermore, many orders of magnitude more sources means that there are many orders of sources that are dangerously wrong as well. Thus, if someone sees a number articles/tweets/posts/whatever from a variety of people espousing that the Texas church shooter was an antifa commie starting the revolution, they are much more likely to believe it.

On top this, with so many sources, people need methods of filtering, and, of course, one of our primary methods of filtering is to find communities/people we tend to agree with. Thus, the now cliche echo chambers are born. These communities are furthermore highly susceptible to manipulation (see the russian facebook ads).

Furthermore, this makes the system fairly easy to take advantage of. If a number of people that seem unrelated online engage in a coordinated effort to spread a particular message, they can do it with ease, since when you see the same message from multiple, seemingly unrelated sources, you're much more likely to believe it.

So while you're right that we used to rely on gossip and hearsay, simply amplifying the number of sources in no way implies that we're getting a less biased message.

I don't follow. This would only work if the distribution of numbers was predetermined and influenced the distribution of mines. I was under the impression that the mines were uniformly distributed though. So in a "T" scenario, it really is 50-50. The fact that 4s are more rare doesn't matter at that point. Sort of like in a series of coin flips, of you've flipped 5 heads in a row, tails isn't more likely to come since 6 heads are rare: it's still just 50-60.

k-nearest neighbors (the k is just the number of nearby data points you consider when classifying a point), a clustering algorithm, and artificial neutral network

What about fingerprints? You don't need to buy anything, phones and such are already coming out with fingerprint readers and the technology is improving all the time. Finally, you can't lose your finger print (except in the case of extreme accidents, which exist for any type of security).

The fact that it just outputs to stdout means you can run any command with output from fzf as well. Just

anycommand $(fzf)

I've been looking for something like python's pandas as well, so curious what they say.

Loving Wisp! I've been needing something exactly like it and will definitely be making use of it. So thanks!

The next slide shows that.

I think that particular slide was poorly worded. He's actually introduction a consequence of the rules, but not saying it's an obvious consequence. Slide 19 shows how (1∠90∘)2 = -1.

Honeypot, but that doesn't hold up since torrenting movies is so wide spread already.