Alexey Vorobyov @vorobya

Russia. MIPT. Gambit. Rockwell. Conexant. Pictos. ESS. Proxyconn. GTB. Google California Joined April 2009

Tweets

2K
Followers

251
Following

4K
Likes

49K

Alexey Vorobyov @vorobya

2 years ago

@briansowards @fchollet Completely agree with "an artifact of what we thought we were doing". The mental model that François discusses is not fixed. Our mental models change during development because we discover things we did not know when we started. That's why no API reflects the latest mental model.

0 0 1 19 0

View Details

Alexey Vorobyov @vorobya

2 years ago

@fchollet Then improvements become small relative to the volume of knowledge in this "universe". I guess you call it "logarithmic growth". I would bet if the same people started earlier, our early progress would be faster but shorter: until independent factors get exhausted.

0 0 0 12 0

View Details

Alexey Vorobyov @vorobya

2 years ago

@fchollet And when more people started working on this universe, the growth of the volume is nearly linear. For DL you can play with the number of layers, activation functions, normalization, the structure of filters, etc. People can work independently and improve general performance.

1 0 0 11 0

View Details

Alexey Vorobyov @vorobya

2 years ago

@shortstein where atan(x) = x - x^3/3 + x^5/5 - x^7/7 ... I verified in python that it converges in 12 iterations to 13.14159265 that I remember, and then it needs 5-6 more iterations to get 3 more digits.

0 0 0 12 0

View Details

Alexey Vorobyov @vorobya

2 years ago

@shortstein atan(1) = atan(1/2) + atan(1/3)

1 0 1 45 0

View Details

Alexey Vorobyov @vorobya

2 years ago

@shortstein What about pi^2/6 = sum(1/n^2)?

1 0 1 756 1

View Details

Alexey Vorobyov @vorobya

3 years ago

@fchollet As they say in one Russian movie: "The physicist should be lazy enough to not do UNNECESSARY work".

0 0 1 64 0

View Details

Alexey Vorobyov @vorobya

3 years ago

@tdietterich @fchollet There are different ways to address this. Consider microbes. They exist without consciousness. Sometimes there is a power in numbers, when adapting not a single entity but a population.

0 0 0 32 0

View Details

Alexey Vorobyov @vorobya

3 years ago

@fchollet How then can you explain en.wikipedia.org/wiki/Jaime_Esc… when one teacher changed in a couple of years the aptitudes of dozens of students?

0 0 0 34 0

View Details

Alexey Vorobyov @vorobya

3 years ago

@fchollet Tell this to Mauglis of this world. There are several documented cases.

0 0 0 22 0

View Details

Alexey Vorobyov @vorobya

3 years ago

@fchollet On the contrary. Twins growing in the same family sometimes grew up radically different: one into sports, another one in academics.

0 0 0 129 0

View Details

Alexey Vorobyov @vorobya

3 years ago

@vicktop55 Reference?

0 0 0 97 0

View Details

Alexey Vorobyov @vorobya

3 years ago

@HaraldinChina @philiptangatue Compass

0 0 4 181 0

View Details

Alexey Vorobyov @vorobya

3 years ago

@ME_1948_Updates Or Far East en.wikipedia.org/wiki/Jewish_Au…

0 0 5 2K 4

View Details

Alexey Vorobyov @vorobya

3 years ago

@SovinformMedia Don't imagine enemies where there are none. And with this: "Have a nice one (day, night, morning)!"

0 0 1 37 0

View Details

Alexey Vorobyov @vorobya

3 years ago

@SovinformMedia If you claimed that you have robust methodology, then when I presented you with my sources, you should show me that they were manipulated by demonstrating more reliable sources (from 30s like you said). That would be a real solution to the issue.