What specifically am I working towards?

Questions
#1

Hi!

I discovered Eyewire a few days ago and I’m loving it so far. However, I only have a vague sense of what I’m actually working towards.
I think my work gets reviewed by more experienced people who might not have to spend as much time on it as I have, because I did a rough chunk of the leg work?

I also know that to manually continue this work is not exactly feasible to get a map of the retina. So we’re ultimately training the algorithm, right?

I’m also interested in finding out more about what has been achieved or improved since this project started. So if anyone could list or link to more information about that, that would be really cool.

Thanks!

#2

Hello, and welcome! Always good to see new people!

I think my work gets reviewed by more experienced people who might not have to spend as much time on it as I have, because I did a rough chunk of the leg work?

You are correct, in some ways. Eyewire primarily works on a consensus system. Each cube will typically be played by at least three players. The parts that the majority agree should be included in the cell are assumed to be correct, and the cell continues to grow if it doesnt end right there.

At least two higher level players (scythes or mystics) will then go through each cube in the cell to double check the consensus, before handing it over to the admins who will also give it a quick look before marking it as “complete”. We also have scouts (usually scythes in training ;P) who keep an eye on the cells too, and report any issues they find for us to fix.

I also know that to manually continue this work is not exactly feasible to get a map of the retina. So we’re ultimately training the algorithm, right?

A map of a full retina is definitely a big job. Eyewire is built around the E2198 dataset, which is a small section (just 350×300×60 µm^3) of Harold’s (the mouse) retina. You can read a little about the dataset itself on the wiki, and there was a decent write-up for one of our more recent competitions that covers a bit about the data, and some of the research that has come from us playing in this blog post.

As far as I’m aware, our tracing isnt being used to retrain the AI for eyewire (I could be wrong on that), but I think Pyr will be different in that respect, once we are allowed our hands on it.

I’m also interested in finding out more about what has been achieved or improved since this project started.

I think this blog post from october 2020 might be a good overview of how much of the dataset we have reconstructed so far - we are just about to wrap up sector 2, which is great progress. There is also the museum which i like to look at now and again, which has a nice collection of a lot of the cells we have traced.

However, I only have a vague sense of what I’m actually working towards.

I have… purposefully(?) not answered this, as I’m definitely under-qualified to give you a good answer. I’m sure somebody who knows much more than me will be along shortly. I believe that the research done with our reconstructions is focused on finding the synapses, and how the cells all interact with each other, how signals propagate, etc. I’m mainly here for the tracing and the community, though, so who knows ¯\_(ツ)_/¯

I’ve rambled a bit more than i usually do, lol. Hopefully some of that gives you what you were looking for. Everyone here is a friendly bunch, and there will usually be somebody online who can answer any other questions you might have.

3 Likes
#3

Thank you so much for the write up!

This definitely helps me convert my ambiguous sense of doing something productive into a more concrete motivation. Tracing is just so chill it almost felt too good to be true. :grin:

I’ve read up a little bit on Pyr, it looks very interesting too. I’ve got plenty of other questions about it (born from pure enthusiasm), but I think a lot of them are probably covered by the blog so I’ll keep an eye out!

Thanks again :smile:

1 Like
#4

There have also been done some actual science with Eyewire. Both E2198 and Zfish datasets got us interesting scientific papers.

Here is a blog post about the first discovery done with E2198:

And the paper itself:
https://www.researchgate.net/publication/262146309_Space-time_wiring_specificity_supports_direction_selectivity_in_the_retina


Another blog post and paper for **E2198**:




Here’s a Zfish discovery:


And here's the latest paper based on the **Zfish** dataset:




And finally, here are two interesting papers about the MICrONS dataset, and how it was processed to prepare the data, on which will be working in Pyr:




Also, if you seach for “Eyewire” in bioRxiv, you can find dozens of papers based directly or indirectly on Eyewire:

https://www.biorxiv.org/search/eyewire

3 Likes
#5

I agree - why work on eyewire when algorithms already do so much? There are multiple reasons, but it’s up to you in the end. The short answer’s human power to help out algorithms - they can’t do it alone.

TL:DR (as in, do read):
In a sense, working on eyewire is training ourselves to help out algorithms with their work, as they don’t do a perfect job and it’s up to us to fix all the details. That’s how it works whenever you work on a cube - the AI did some part and you do the rest. You can see how far algorithms have come in every cube.

Imagine if the algorithms do 99% of the work and humans do 1% - that’s still a lot of work!

I mean if you feel a little disenfranchised from it all, why not use your time on eyewire to develop your own algorithms to do your work for you, publish them on the forum, and maybe others can learn from that. That’s what @KrzysztofKruk does. A lot of people do that.

For me, it’s the repetition that brings me to new heights - with the citizen science project on exoplanets, the humdrum inspired someone to look for exomoons! So working on eyewire might inspire you to offshoot a new project from it! Beat the algorithm - we have brains, we might as well use them lol jk :slight_smile: Necessity is the mother of invention.

All of everything on eyewire’s learning and teamwork - for science!

2 Likes
#6

As for your question, you are correct that the work you do on Eyewire is reviewed by more experienced players, and ultimately helps train the algorithm to map the retina. This process is known as “citizen science”, where regular people like you and me can contribute to scientific research by performing tasks that help train machine learning algorithms.