iiinova — where your digital self keeps evolving

Brad Hulse

@bradkhulse.bsky.social

445 followers 476 following 28 posts

Neuroscientist | Navigation | Central complex bumpologist | Senior scientist at Janelia

Posts Replies Media

Brad Hulse @bradkhulse.bsky.social · 21/05/2026

Thanks for reading, and huge shoutout to my wonderful coauthors and collaborators P.B. Aneesh, Sandro Romani, Vivek Jayaraman, and Ann Hermundstad! Link: www.biorxiv.org/content/10.6...

1110

Brad Hulse @bradkhulse.bsky.social · 21/05/2026

Guided by this understanding, we analyzed multiple fly connectomes and found evidence for hidden symmetries in the recurrent connectivity of head direction cells with shared directional tuning, suggesting the fly’s heterogeneous connectivity can support ring attractor dynamics.

150

Brad Hulse @bradkhulse.bsky.social · 21/05/2026

Unlike classic models, which clone entire rings to build in phase shifts, our algorithm uses single-unit phase shifts for velocity integration. By understanding how hidden symmetries are generated, we could reverse the algorithm to recover them from heterogeneous networks.

120

Brad Hulse @bradkhulse.bsky.social · 21/05/2026

To test this idea, we developed an algorithm that clones the dynamics of individual neurons, allowing us to transform networks with symmetric connectivity into functionally equivalent networks with “hidden symmetries” without sacrificing exact attractor dynamics.

140

Brad Hulse @bradkhulse.bsky.social · 21/05/2026

These solutions suggest a large space of ring attractor architectures with varying degrees of modularity and heterogeneity.

171

Brad Hulse @bradkhulse.bsky.social · 21/05/2026

Connectome-constrained RNNs and fully unconstrained RNNs learned solutions with three properties not found in classic models: clustered head direction tuning (where multiple units share a ring location), heterogeneous phase shifts, and variable velocity tuning.

150

Brad Hulse @bradkhulse.bsky.social · 21/05/2026

To start, we trained RNNs to integrate angular velocity, but with recurrent weights that ranged from fully ring-constrained (as in classic models) to fully unconstrained. Comparing these to connectome-constrained RNNs let us place the fly attractor in this broader solution space.

130

Brad Hulse @bradkhulse.bsky.social · 21/05/2026

Story time friends... Ring attractor networks rely on fine-tuned symmetric connectivity. The fly head direction network has ring attractor dynamics but heterogeneous connectivity. How is this possible? 1/🧵 Link: www.biorxiv.org/content/10.6...

58830

Brad Hulse @bradkhulse.bsky.social · 15/01/2025

Jeff Taube joined Jim ~2 yrs later and, together with Bob Muller, the trio quantitatively characterized HD cells’ tuning curves for the first time (using a 2-spot video tracker!). In 1990, they published their classic papers in the journal of neuroscience.. (2/3) www.jneurosci.org/content/10/2...

182

Brad Hulse @bradkhulse.bsky.social · 15/01/2025

Happy head direction cell day y’all! On this day, 41 years ago, Jim Ranck recorded the first “HD cell” in his Brooklyn lab in 1984. Jim shared the news with the world at SFN that year, showing a video of the cell’s firing on a TV he somehow acquired and brought to his poster… (1/3)

47031

Brad Hulse @bradkhulse.bsky.social · 04/10/2024

Individual variability is super interesting! At the same time, I agree with your point above... the fly brain is stereotyped enough that we can generate driver lines targeting specific cell types and in most cases map those to a connectomic cell type. Video for fun (magenta = EM; green = light) :)

120