Arseny Khakhalin

@ampanmdagaba

Assistant prof (Neuro) at . Networks, neurons, data, AI, complexity, beauty, platonic forms.

Vrijeme pridruživanja: lipanj 2012.

Tweetovi

Blokirali ste korisnika/cu @ampanmdagaba

Jeste li sigurni da želite vidjeti te tweetove? Time nećete deblokirati korisnika/cu @ampanmdagaba

  1. Prikvačeni tweet
    27. ožu 2019.

    New : Graph analysis of looming-selective networks in the tectum, and its replication in a simple computational model Featuring in-vivo Ca , directed (!) connectivity reconstruction, science, and open code. (thread, 1/n)

    Prikaži ovu nit
    Poništi
  2. proslijedio/la je Tweet
    1. velj

    Types of peer review: 1) 3 anonymous scientists write secret reviews in a process that takes a few months 2) dozens of known experts write comments that anyone in the world can see hours after a paper is posted (e.g this thread 👇) Explain to me again why is 1 >> 2?

    Poništi
  3. I downloaded a pdf of this "Emoji Guide to Human Genetic Diversity", just to give it a curious look, and ended up reading the entire thing! It's so entertaining, concise, and well-written! I might actually assign it as a reading for one of my classes!! :)

    Poništi
  4. 1. velj

    Interesting: building a fist-person bot is inherently easier than a 3d-person bot, as first-person view meets you half-way in terms of translation of inputs to actions. Arguably RNNs/attention is also like "first-person for text". I wonder how it applies to graphs (crawlers?)

    Poništi
  5. 1. velj

    Next iteration of : not just colorize, but also add bicycles!

    Poništi
  6. 1. velj

    I don't know enough, but I suspect it may depend on whether one took humanities classes in college, or read enough books not about science. It feels like one of those words that have slightly different meaning in common language and academic discourse, like myth, or theory.

    Prikaži ovu nit
    Poništi
  7. 1. velj

    It is quite clear that linguistically there's a weird divide, almost two unrecognized dialects of English, with half people using the word "storytelling" to describe a way to organize thoughts and emotions, and another half ascribing a semantics of lying, exaggeration to it.

    Prikaži ovu nit
    Poništi
  8. proslijedio/la je Tweet
    31. sij

    A wonderful little fact about the maximum power density inside the sun (spoiler: about the same as inside a compost pile!) ht

    Poništi
  9. 31. sij

    For Russians, it is a bad omen to talk across the door threshold. Passing things, or shaking hands across the threshold is a strict taboo. It makes talking to canvassers, proselytizers, and vampires awkward, esp in the winter; I always have to either step out, or invite them in.

    Poništi
  10. proslijedio/la je Tweet
    30. sij

    Did you know that Breiman published both the bagging and random forest papers *after* he retired?!? (This is from our forthcoming book: )

    Poništi
  11. 30. sij

    Lovely directed graph! Colorado, New Hampshire, and Florida have high pagerank centrality (love-sinks). There seem to be 3 cycles, but all length 2 only. Also, adding Canada and "No one" to the list was not fair :)

    Poništi
  12. 30. sij

    The semantics of "rebase" verb in GIT is confusing. "Merge branch into master" somehow reads more intuitive than "Rebase branch onto master". Like, who is moving where? Is it me coming there, or them coming to me, you know? I get it now, but it took me a few months :)

    Poništi
  13. 29. sij

    Yet all we have is heuristics. The good news is that computers are more efficient with heuristics, as they can remember more, and without decay. So in practice, a combo of ANN “intuition”, greedy sampling, and perfect Von Neumann memory can totally do the trick.

    Prikaži ovu nit
    Poništi
  14. 29. sij

    ANN applications often have a combinatorial bottleneck, where locality is violated, and you have to compare everything to everything. It feels similar to P-NP: if you solve it once you solve them all. I suspect it is unsolvable tho, or we would have evolved a solver in our heads!

    Prikaži ovu nit
    Poništi
  15. 28. sij

    Welcome to the Uncanny Era of Tech-Support! (It's like Uncanny Valley, but in time. And for tech support). The period of time between early 2020s and early 2030s, when tech support was already automated, but chat bots were still less helpful than an overworked human contractor.

    Poništi
  16. proslijedio/la je Tweet
    28. sij

    I see arstechnica also has an article about it Though the focus was a bit different obviously. There was tons of interesting footage produced and some of it clearly showing AP approach to objects it's unsure about.

    Poništi
  17. proslijedio/la je Tweet
    28. sij

    “Tell me again, what instrument do you play?”

    Poništi
  18. 28. sij

    Or look at this one. A model that can recognize characters (Cyrillic! Must be Soviet), or read Russian (Glory to Soviet Builders), AND access to texts (Soviet banners are always red) would have colored banners red. A picture-only model cannot do it. A test for _Understanding_!

    Prikaži ovu nit
    Poništi
  19. 28. sij

    What would it take for it to connect the dots? (Another photo from my archive that no model have seen before). Recognize tokens (a portrait of a tyrant), and objects (flags) - jump to an easy conclusion. But it's a great test, as color photos with Stalin almost don't exist.

    Prikaži ovu nit
    Poništi
  20. 28. sij

    But the model doesn't know that! And we can see the same limitations here, in spirit, that GPT-2 demonstrated with language. Both models can do correlations, like coloring a single flag, or a well-known situation. It is "aware" of Soviets. But it cannot connect the dots.

    Prikaži ovu nit
    Poništi
  21. 28. sij

    You probably did well (but you also knew my name, unfair :). Still, from the clothes and optical imperfections you can guess the era (1950s). From the faces (a vague mix of European and Asian), and the roofs -> Russia (Tatarstan). Therefore, all flags should be bright red!

    Prikaži ovu nit
    Poništi

Čini se da učitavanje traje već neko vrijeme.

Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.

    Možda bi vam se svidjelo i ovo:

    ·