Audrey Sederberg

Asst Prof of Neuroscience at the University of Minnesota / Theory and computation / MDT on Optical Imaging and Brain Science

[Audrey Sederberg @AJSederberg]

New postdoc position in computational neuro/psych available! Come study neural ensemble dynamics in health and disease with me and @adredish as part of the @neuroprsmh center at UMN. DM with questions, link to apply: https://t.co/EVTuznU7tV

[Audrey Sederberg @AJSederberg]

Perhaps one should re-examine the very notion of criticality: what does it mean to be critical at all? 🤔

[Audrey Sederberg @AJSederberg]

Spearheaded by @MiaMorrell1, an @EmoryUniversity undergrad-now-grad, and with @INemenman, we present a model that reproduces multiple signatures of criticality measured in large-scale neural recordings, within experimental uncertainty, without fine-tuning https://t.co/oQm0A0O9Bq

[Audrey Sederberg @AJSederberg]

The model applies to any system with widespread coupling to hidden variables. So data from many biological (eg, microbes in a gut flora responding to changes in a diet) or social (eg, tweets affected by current events) systems are likely to appear critical via the same mechanism.

[Audrey Sederberg @AJSederberg]

@jmourabarbosa @KameronDHarris @farznaj_PhD @rajankdr We just wanted to ask about selectivity, decoder weights and noise XC, so we kept it simple. (And yes, @farznaj_PhD‘s study was the inspiration for this paper. Much respect for the beautiful data!) 2/2

[Audrey Sederberg @AJSederberg]

All models are wrong, but this one could be useful. Happy to answer questions by tweet.

[Ilya Nemenman @INemenman]

Led by @AJSederberg, we propose a random network model that explains some features of population codes observed by @anne_churchland lab. Thanks to @KavliFoundation for supporting interactions that led to this work!
https://t.co/kU1tPsOQsz

[Audrey Sederberg @AJSederberg]

@jmourabarbosa @KameronDHarris @farznaj_PhD Probably a network can generate persistent, decodable activity without having to put in distinct clustered sub-networks. You may not call that network fully ‘random’ anymore - maybe it looks something like the more realistic PPC models (e.g. @rajankdr’s work). 1/2

[Audrey Sederberg @AJSederberg]

@DrJigsaww Not disturbing at all! The data is what it is - selectivity, correlations, decoding properties are all real. We just showed you can get these (quantitatively !) from a heterogeneous network with nonlinearities. That’s good news, b/c we do not want to model every single detail...

[Audrey Sederberg @AJSederberg]

@KameronDHarris Definitely! It’s a simple task for the network - the fun thing is that it matches features of the data as well. It’s a cool ? tho: how much connectivity has to be specified for the brain to work? A lot, for sure, but it’s nice to get something for ‘free’ from emergent properties

[Audrey Sederberg @AJSederberg]

@dlevenstein It’s a nice null-model, but it’s also raising the question: how much can we get from emergent dynamics, without necessarily putting in specific connectivity? Surprisingly a lot!

[Audrey Sederberg @AJSederberg]

@ksederbe Is this a translation error (“not absolutely necessary” vs “absolutely unnecessary”) or a philosophical difference?

[Audrey Sederberg @AJSederberg]

Boarding flight to Berlin for #BernsteinConference satellite workshop! Looking forward to predictive coding... or coding for prediction?

[Audrey Sederberg @AJSederberg]

⁦@ColinVParker⁩ according to this, you are a theorist now too. cc ⁦@gordonberman⁩ and Jen https://t.co/z9lmQiiric

[Audrey Sederberg @AJSederberg]

@cian_neuro Literally ignored Twitter for six months b/c I was so tired of Python vs Matlab tweets. But I missed reading people’s preprint tweets, so hello. Julia? Sounds promising...

[Audrey Sederberg @AJSederberg]

@Yoav_Livneh Hey, I thought this was really interesting too. Are any of the authors on Twitter that you know of? I have questions...