gradys

8 ms·

211.
by gradys 10y ago
Well put. I think it's really valuable to take inventory of human cognitive capabilities that we don't yet know how to implement in neural nets. We seem to have more or less solved perception. Given a high dimensional, "raw&q
212.
by gradys 10y ago
The T-shape metaphor is imperfect. In reality, going deep doesn't mean digging down into an isolated vertical of knowledge. Knowledge is a graph. To gain more knowledge in one topic, you've got to traverse the graph, following edg
213.
by gradys 10y ago
For me, it is almost the opposite. I can't get into games that are highly "stateful" anymore. I find that the only games I can consistently play for relaxation are games like match-based FPSs that I'd call "stateles
214.
by gradys 10y ago
My understanding is that the probability and expected magnitude of an earthquake actually increases over time as tectonic tension builds. So no, not like dice.
215.
by gradys 10y ago
Nautilus has an article[1] that explores exactly this question. I beleive the conclusion was that it is the number of cortical neurons, not the total number of neurons and not brain mass that best predicts intelligence, or at least best cor
216.
by gradys 10y ago
They spend a lot of time talking about how dataset owners don't have to do anything to expose it to GOODS. It sounds like a big part of the work that goes into GOODS is automatic metadata inference.
217.
by gradys 10y ago
There are different notions of "fast" as well. Theano might be faster on a single GPU (not actually sure if this is still true), but TensorFlow can be distributed to multiple GPUs in one machine or even across several machines. Se
218.
by gradys 10y ago
If I were representing a big, well known company with lots of money to give out in settlements, I'd be much more likely not to say anything than to bother lying.
219.
by gradys 10y ago
Sibling comments have mentioned Neural Turing Machines, but even just regular recurrent neural networks are technically Turing complete[1]. [1] - http://lipas.uwasa.fi/stes/step96/step96/hyotyniemi1/
220.
by gradys 10y ago
I'm with you. I liked the steak count as a source of personal motivation, but we still have that don't we? Now you need to look at the squares, but an unbroken sequence of colored squares is just as motivating as a number, isn&#x2
221.
by gradys 10y ago
Generally, they don't. word2vec and GloVe, the two most popular word vector models in the circles I run with, don't have any solution to this at all. As a result, when you do two dimensional visualizations of these vector spaces,
222.
by gradys 10y ago
Or train it to predict code in a code base conditioned on the rest of the code in that same code base. That way, when you condition it on your current code base, it might be able to generate sensible completions that could actually be usefu
223.
by gradys 10y ago
Adding $25 USD bills would break the greedy choice property for making change. Making $40 in change using a greedy strategy would mean selecting a 25, a 10, and a 5, while the optimal choice is to select two 20s.
224.
by gradys 10y ago
No, it's more like a prior.
225.
by gradys 10y ago
They could also tell that you didn't leave the SERP between clicks.
226.
by gradys 10y ago
I doubt that using their customers' audio as training data was a major motivation for offering this service. But , assuming that was their plan, they'd have a couple options: - Like you said, they could turn it into supervised
227.
by gradys 11y ago
I guess I should walk back what I said a little bit. Natural language is the ultimate interface for many domains. For flying a plane or doing anything else that requires continuous high dimensional input, physical interfaces are probably be
228.
by gradys 11y ago
I'm very much with you on most of this, but to reply to this: > but do they need to understand our language? Natural language is the highest-bandwidth means of communication available to us. It's the easiest and most powerful m
229.
by gradys 11y ago
How surprising is this when Janet Napolitano[1], former Secretary of Homeland Security, is the president of the UC system? [1] - https://en.wikipedia.org/wiki/Janet_Napolitano
230.
by gradys 11y ago
Do you know of any data sets of examples of logical fallacies?
231.
by gradys 11y ago
Seems useful for any kind of local web development for one, plus there are lots of consumer applications that run locally as a web app, like Plex, for instance.
232.
by gradys 11y ago
I think the primary difference is efficiency. You won't gain strength as quickly doing construction as you will following a strength training program. If your goal is to gain as much strength as possible with the lowest time commitment
233.
by gradys 11y ago
I get the feeling that the point of this work was more about expanding our understanding of CNNs than about solving the problem of texture synthesis. Interesting references nonetheless.
234.
by gradys 11y ago
I find it interesting that the network seems to have trouble getting the global structure right. This is particularly clear when the source features a regular pattern that carries through the whole image. If you zoom in on a small enough re
235.
by gradys 11y ago
Your understanding is correct. TensorFlow is a library for defining and executing mathematical operations on multidimensional arrays. It just so happens that much of machine learning, and especially deep learning, can be described as mathem
236.
by gradys 11y ago
> They are storing data about you for the people who, like it or not, you shared your data with. Exactly. I'll add that on a technical level, you aren't being tracked like a hunter would track prey; your machine is being period
237.
by gradys 11y ago
> While you can do the math, and get the right results, it does not necessarily make it real outside of the human brain. Following this same logic, nothing is necessarily real outside of the human brain. No experiment you can do, includi
238.
by gradys 11y ago
PageRank and RankBrain both seem to be (complicated) features to the actual core search algorithm, which is some unspecified machine learned model [1]. PageRank, for example, can't be the whole search algorithm since it's not even
239.
by gradys 11y ago
You could simulate it by recording its position on the front page for each impression. Assume that higher positions on the page are seen more. Then your metric is something like upvotes per position-discounted impression. On the other hand,
240.
by gradys 11y ago
Neurons are also relatively simple, at least in comparison to the mind. I don't think the simplicity or complexity of the underlying model has much bearing on the higher-level properties of the network . Now, this isn't to say th

More ›