Hacker News Reader

Comment by etrautmann | Hacker News Reader

This paper is from a colleague in my lab at Stanford - happy to answer questions.

For intuition, this is like clicking on an on-screen keyboard, so the typing speed is very high given the constraint of focusing on each key as a separate, discrete movement.

View on HN

It may be faster to use something like Dasher instead of an on-screen keyboard: http://www.inference.phy.cam.ac.uk/dasher/DasherSummary2.htm...

We did this 10 years ago at Wisconsin, during my PhD:

http://www.inference.eng.cam.ac.uk/saw27/dasher/bci/

The video I made of one of our undergrad assistants is still up there!

Part of me wonders though - things seem a lot less discreet in dasher... but given that input is really in one dimension, and not two, it might be able to work faster...

> things seem a lot less discreet in dasher

There are discrete versions of dasher that operate using a single digital switch or actuator, rather than an analog input. As long as you can feed in one bit of information, Dasher can work with that.

Yeah, I wanted to ask why they didn't use some kind of adaptive autocomplete. Was it too disorienting to the monkeys?

This is very cool, but we've had spellers based on the P300 signal for some time; here's an article from 5 years ago explaining how to set one up: http://openvibe.inria.fr/openvibe-p300-speller/

So it looks like the big improvement here is speed; P300 spellers are slow, like 4 characters per minute. What techniques does this new research use to improve that detection speed? The article doesn't go into much detail about how the signal detection or text entry actually works.

There is super work going on in your camp, kudos! I really appreciated your review of fluorescent imaging in NHP, which is absolutely going to be, along with advancements in implantables, a major enabling tech for prosthetics. In our lab, we've recently started down the path of light field imaging for whole-brain, single neuron recording in the larval zebrafish - supplementing existing work in developing very large, dense ephys recordings. As crazy as NESD/MiCRONS programs are, it really does feel like the field is transitioning to stitching pieces, many years in the making, together towards bona-fide clinical applications. Exciting times!

Thanks! Exciting times indeed. Which lab is that?

We'd love to get lightfield imaging working too but starting with vanilla 2P and building from there.

http://www.leaflabs.com/neuro-about - if you check out lotus there are some links in there around the lightfield setup, some papers and a publication that cites perhaps everything you'd need to get started.

Would this also allow for chorded input?

This particular approach would require an extension to do that, likely decoding from the hand and finger related regions of motor cortex. It's a great idea but would require additions to the algorithm to decode overlapping discrete actions for chord detection.

Sounds like the ultimate solution to emacs carpal tunnel syndrome.

What do you think of this? - http://www.dailymail.co.uk/sciencetech/article-3386875/Could...

What do they do with the monkeys after they're finished with their experiment?

The monkeys continue on and often live longer than they would in the wild without expert veterinary care, optimized diets, etc.

Do they suffer much pain?

Well if they do I think we can expect them to write home about it!

Reply on news.ycombinator.com