back to article Forget Finding Nemo: This AI can identify a single zebrafish out of a 100-strong shoal

AI systems excel in pattern recognition, so much so that they can stalk individual zebrafish and fruit flies even when the animals are in groups of up to a hundred. To demonstrate this, a group of researchers from the Champalimaud Foundation, a private biomedical research lab in Portugal, trained two convolutional neural …

  1. Anonymous Coward
    Anonymous Coward

    yet it works for fruit flies, too

    hence it should work on humans, despite one's hopes. I'm sure "the force" would be VERY interested, another "disruptive" technology for crowd "control". Long live democracy, long live control! Long live scientists who play stupid at what their research is about!

    1. Khaptain Silver badge

      Re: yet it works for fruit flies, too

      Exactly, the French could use it to identify and eventually follow the trouble makers within the various "Gilet Jaune" manifestation ( be it the manifestors or the police).

      1. Symon Silver badge
        Coat

        Re: yet it works for fruit flies, too

        Anyone up for some antanaclasis? Time flies like an arrow. Fruit flies like a banana.

        https://en.wikipedia.org/wiki/Time_flies_like_an_arrow;_fruit_flies_like_a_banana

  2. Wellyboot Silver badge

    sometimes humans have to manually label the pictures

    Are humans anywhere near 100% accuracy when trying to identify individual fish & fruit flies? I'm asking because that is the only way to verify these results.

    1. Neil Barnes Silver badge

      Re: sometimes humans have to manually label the pictures

      Maybe they identified the one they were interested in with a spot or something, and the AI is following the spot?

    2. Tom 7 Silver badge

      Re: sometimes humans have to manually label the pictures

      They're already barcoded!

    3. Ragarath

      Re: sometimes humans have to manually label the pictures

      If you were given 30 images of the fish in question and asked to compare I'm sure you would be 100% accurate.

      It may take a while depending on the $animal and you may be bored out of your skull by then end of the process. But that is what this is, it's like the silhouette puzzles or a Where's Wally/Waldo). Your given the relevant info, you just need to study it.

      To verify this I assume they had a human do so or how would you tell the computer got it right?

      1. Peter2 Silver badge

        Re: sometimes humans have to manually label the pictures

        You wouldn't. It's called "finding nemo" because "Nemo" means "nobody" in Latin.

        So it's not following the fish. Says so right there in the title. ;)

    4. Korev Silver badge
      Boffin

      Re: sometimes humans have to manually label the pictures

      I did an internship in a developmental biology lab before I came over to the dark IT side.

      One of my tasks was to select flies by their phenotype (eye colour, wing shape etc) so they could be bred together*. This was very difficult and tedious to do**, what was especially hard was telling the sexes apart on newly-born flies (those randy buggers get at it four hours after birth).

      * We were studying a gene known to be involved in neurodegeneration in humans.

      ** Probably why they gave the job to the undergraduate

  3. Tom 7 Silver badge

    A cool

    While I was in NZ I spotted some flies doing what must have been a mating dance but analysis was near impossible as you couldn't tell which of the fuckers was which. You could if you filmed them a slowed them down but it had to be very slowed down. This would help work out what was going on if its possible to identify them.

    1. Allan George Dyer Silver badge
      Coat

      Re: A cool

      I call Rule 34 on slo-mo fly porn!

      1. Gene Cash Silver badge
        Coat

        Re: A cool

        I'm sure Gav & Dan are already on it... for various values of "on it"

        1. Is It Me

          Re: A cool

          Any reference to the slomo guys is good

  4. Unep Eurobats
    Coat

    AI can identify a single zebrafish out of a 100-strong shoal

    That's easy, it's the stripy one. Duh.

  5. Eddy Ito Silver badge

    From the study

    We randomly selected 3,000 images per animal for training.

    There's a problem. Where are the spy monkeys going to get anywhere near that many facial images of everyone to train their facial recognition? Oh that's right, they aren't so concerned with accuracy when it comes to upending the lives of we plebs.

  6. Anonymous Coward
    Anonymous Coward

    Real test is if they try on Chinese wearing the same outfit.

    1. Anonymous Coward
      Anonymous Coward

      Whilst right from a technical perspective it's wrong in so many others, even as a joke. If only you had chosen Norfolk residents instead of the more genetically diverse group of Chinese.

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Biting the hand that feeds IT © 1998–2019