Andrew Erskine retweeted
Good time to announce that I will be starting my lab at SIDB & University of Edinburgh early next year! 🎉 We will investigate neuromodulation and cortical learning, as well as their dysfunction in mouse models of autism spectrum disorder @sidb_edinburgh
Fun project with @AriaSafar using StyleGAN to generate new Arabic fonts. Shown here exploring the latent space at different training iterations for 106 different contextual letterforms.
A practical introduction to RNNs for neuro with code samples. Written after this year's COSYNE dev.to/robodoig/a-practical-…
This all started for me almost 5 years ago when @AndreasTSchafer challenged me to hack our olfactometer to produce odour fluctuations. What a journey since then! Grateful to have worked with this incredible team.
1/ Mouse olfaction is high bandwidth – our new paper just out in @nature – great lab collaboration by @TobiasAckels, @incisrdrew, @debanjanmona, @MiceBrains, @thomwarn, @stootoon, @FukunagaIzumi, @JuliaJadeHarris
nature.com/articles/s41586-0…
Andrew Erskine retweeted
Saw some of this in @incisrdrew's PhD thesis, and it was brilliant then - fantastic to see it out now. Congrats all!
1/ Mouse olfaction is high bandwidth – our new paper just out in @nature – great lab collaboration by @TobiasAckels, @incisrdrew, @debanjanmona, @MiceBrains, @thomwarn, @stootoon, @FukunagaIzumi, @JuliaJadeHarris
nature.com/articles/s41586-0…
Andrew Erskine retweeted
1/ Mouse olfaction is high bandwidth – our new paper just out in @nature – great lab collaboration by @TobiasAckels, @incisrdrew, @debanjanmona, @MiceBrains, @thomwarn, @stootoon, @FukunagaIzumi, @JuliaJadeHarris
nature.com/articles/s41586-0…
Andrew Erskine retweeted
Our lab’s work on how cortical representations of shape are constructed from tactile input, and how they reorganize with training is out now in @NeuroCellPress. 🙏 to reviewers (who greatly improved the manuscript), editor & @NINDSfunding. 🧵by project lead @jkimkj. – at Los Angeles, CA
Ever wondered what happens in our brain when we touch a 3D object? How about when we recognize an object just by touching it? We looked into mice brain to get some idea. cell.com/neuron/fulltext/S08…… (Feat. @JonCheung6 and @incisrdrew Prod. by @AndrewHires). A lay-word thread (1/12)
2nd #gamejam #ludumdare46 complete! A minimalist word game on the theme of "Keep it alive" - robodoig.itch.io/keep-it-ali…
1
4
Andrew Erskine retweeted
Bonsai 2.5 + real-time DeepLabCut! You can now connect live tracking to your hardware in closed-loop. Install from: bonsai-rx.org/, DLC GPU install instructions at: github.com/bonsai-rx/deeplab…
Calling time on my first #gamejam #LudumDare45. Introducing medium-sized field simulator 2019, made from scratch in <72 hours.
ldjam.com/events/ludum-dare/…
1
Andrew Erskine retweeted
Our deep-dive into mouse behavior, to address how mice locate objects with their whiskers was published OA today in @CurrentBiology.
If you are interested in learning more, please see this Tweetorial by the first author, Jon Cheung.
cell.com/current-biology/ful…
Tweetorial time!
1. Question: How do mice locate objects with their whiskers? There are many models, but nobody knows for sure.
We investigated six models of whisker-guided object localization to determine the simplest model that fit behavior best.
cell.com/current-biology/ful…
Andrew Erskine retweeted
CHIME: CMOS-hosted in-vivo microelectrodes for massively scalable neuronal recordings dlvr.it/R0Tqpl
Crick researchers have engineered a ‘smart house’ for mice, that allows them to study the animals’ behaviour with minimal disturbance for up to 18 months. 🐭
@AndreasTSchafer 👨🔬@PLOSONE 📖
4
Andrew Erskine retweeted
And, lastly... my favorite part: we demo how to use #deeplabcut to train a single camera-invariant network on a very challenging problem-> a cheetah in high-speed pursuit:
Andrew Erskine retweeted
jULIEs: custom neural probes for recording and stimulation at poster LLL48. Get in touch to see how we're supercharging research and get yours today @MCIneuro #neuroscience #neuroengineering #stimulation #neuromodulation #research #neurotech #Neurosci2018 @SfNtweets @elonmusk
Blown away by how well #DeepLabCut works. Completely random frame selection + overnight model training resulted in a tracker capable of following multiple whiskers simultaneously across different imaging planes, even during overlap.
I used Google's cloud ML engine for the model training step, which required some adaptation of the original code. Maybe useful for those without access to local GPUs for continuous training: github.com/RoboDoig/dlc-clou…

























