Andrew Sedler

Postdoc at @EmoryUniversity with @chethan
ML PhD '23 from @GeorgiaTech

[Andrew Sedler @arsedle]

TLDR: We brought (Auto)LFADS to PyTorch, now available on GitHub and NeuroCAAS! 🔥

@chethan and I are excited to announce lfads-torch, a new impl. of (Auto)LFADS, designed to be easier to understand, configure, and extend! ⚡️
https://t.co/XSOY06uk85
https://t.co/KLzygarpk2

[Jonathan McCart @JonathanDMcCart]

Modern neuroscientists routinely record the complex, goal-oriented, and time-varying activity of thousands of neurons. Can we find representations of neural activity that 1) are human-interpretable and 2) enable the generation of neural activity for unrecorded behavioral conditions? We present our recent work on Generating Neural Observations Conditioned on Codes with High Information (GNOCCHI) 🥔🍝 !!

By leveraging unsupervised, information-based diffusion models, GNOCCHI can discover interpretable latent spaces from neural data and generate high-quality neural activity for specific conditions outside of the set of available neural recordings!

[Judd Rosenblatt @juddrosenblatt]

I started my company originally to build maximally agency-increasing BCI without profit motive, and that's worked decently well so far

BUT AGI timelines are shortening, so we are pivoting to work on neglected approaches to alignment

[Chethan Pandarinath @chethan]

So fortunate to lead an amazingly talented lab. These folks are 𝒄𝒓𝒂𝒏𝒌𝒊𝒏𝒈 𝒐𝒖𝒕 great science.

🧵 of must-see posters, which span NeuroML / AI, clinical BCI (+@BrainGateTeam), population dynamics, 2P Ca imaging, spinal pattern generation, EMG… 🧠🔬👩‍🔬👨‍🔬📈 #SfN23
⬇️

[Mehdi Azabou @ NeurIPS @mehdiazabou]

Is a universal brain decoder possible? Can we train a decoding system that easily transfers to new individuals/tasks?

Check out our #NeurIPS2023 paper where we show that it’s possible to transfer from a large pretrained model to achieve SOTA 🧠!

Link: https://t.co/0Iebjpt4TM 🧵

[Joel Ye @_JoelYe]

Neural Data Transformer 2 (NDT2), preprint + accepted to NeurIPS 23! A study on Transformer pretraining neuronal spiking activity across multiple sessions, subjects, and experimental tasks! With @jenpgh, Leila Wehbe, and @robert_gaunt1! 1/7

[Chethan Pandarinath @chethan]

Excited about making models of neural dynamics more interpretable. An ongoing, multi-year project w/efforts led by @arsedle and then @chris_versteeg.

Grateful for convos w/@SussilloDavid @MattGolub_Neuro @brody_lab Tim Kim, and our friend @shenoystanford

To be continued...

[Chris Versteeg @chris_versteeg]

Ever wondered whether the dynamics learned by LFADS-like models could help us understand neural computation? @chethan,@arsedle, @JonathanDMcCart, and I developed ODIN to robustly recover latent dynamical features through the power of injectivity! 📜 1/ https://t.co/dovP0LxVNt

[Andrew Sedler @arsedle]

@ConsciousMeat @chethan Yes! We’d recommend chopping the continuous session up into fixed-length, overlapping segments and then linearly merging the segments back together after modeling. We used this technique for the MC RTT dataset in the NM paper so there’s more detail there!

[Andrew Sedler @arsedle]

TLDR: We brought (Auto)LFADS to PyTorch, now available on GitHub and NeuroCAAS! 🔥

@chethan and I are excited to announce lfads-torch, a new impl. of (Auto)LFADS, designed to be easier to understand, configure, and extend! ⚡️
https://t.co/XSOY06uk85
https://t.co/KLzygarpk2

[Andrew Sedler @arsedle]

@NinelK1 @chethan Thanks, we hope so! Great question-- we haven't done a rigorous comparison but anecdotally training steps seem comparable to if not faster than TF2. May have something to do with async GPU operations?
https://t.co/Uuz37omGhH

[Andrew Sedler @arsedle]

This project is the result of a significant engineering effort to make (Auto)LFADS more accessible to the community and we are excited to see how it’s used. Please reach out and let us know what you think!

[Andrew Sedler @arsedle]

lfads-torch also includes recent extensions of LFADS to EMG and 2-photon calcium imaging, and provides a modular interface for experimenting with new reconstruction costs, input priors, and data augmentations.
https://t.co/S1khDzgGp4
https://t.co/QPnQ8q2DG6

[Chethan Pandarinath @chethan]

A wise mentor once helped me understand that this job “is all about the people.”

I am so proud that these two amazing people are the first PhD graduates from my lab.

Congratulations, Feng Zhu and Andrew Sedler!

[Chethan Pandarinath @chethan]

Excited about this new line of work in my lab, led by @arsedle w/ @chris_versteeg, to probe the relationship between expressivity and interpretability in models of neural population dynamics.
https://t.co/UrhnrEBjSF