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

For more information, including additional comparisons to LFADS, check out the pre-print at the link below! Major thanks to all the co-authors who helped make this happen! @arsedle, @chris_versteeg, @DomenickMifsud, Mattia Rigotti-Thompson and @chethan ! https://t.co/IdmzvwzQ8t

On biological neural data, GNOCCHI-inferred codes can predict target position for held-out trials better than LFADS, suggesting that the representations learned by GNOCCHI can generalize to unseen conditions better than previous models!

With this learned structure, and its relationship to the neural data captured by the diffusion model, we can generate neural activity for unseen conditions by simply linearly interpolating and extrapolating the inferred codes! To quantify the relationship of this generated neural

When we use GNOCCHI to model the hidden unit activity of an RNN performing a random target reaching task, we find that GNOCCHI can identify relevant task variables (e.g. intended target location) in an unsupervised manner, and that these representations generalize to unseen

Cosyne Workshop Alert! On Tuesday March 5th, @chethan and I are proud to bring you: Understanding Neural Computation using Task-trained and Data-trained Networks. https://t.co/eVOVeIRMfy

Are you a postbac interested in neural engineering / BCIs? 🧠🤖🗣️ Come join our team!! Work directly with our amazing BCI participants. Great exposure for prospective grad/med school applicants! https://t.co/nBh3af0O4Y

Stop by NeurReps 2023 tomorrow morning to hear about how ODIN can help you accurately infer neural latent dynamics from neural recordings! Hint: combine low-dimensional dynamical models with injective, nonlinear readouts!

If you are interested in estimating neural dynamics from population recordings, come to our poster Tuesday afternoon to hear more about how injectivity can improve the interpretability of your models! #SfN23 PSTR445.20 / XX40

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