Arif Hamid Profile picture
Asst. Professor of Neuroscience & head of @Hamid_DiRE_lab | @HHMINEWS Gray Fellow | Neurobiology of Reinforcement Learning #BlackinSTEM #NeuroscienceinAfrica

Aug 13, 2019, 10 tweets

Very excited to share our latest on the organization of activity in dopamine axons in dorsal striatum. We think they are important for reward learning by implementing spatiotemporal credit credit assignment.

biorxiv.org/cgi/content/sh…

What we studied:
1) Scalar dopamine signals are not very useful.
2) There are lots of clues that vector-weighting RPEs can facilitate learning in complex environments.
3) So, local dopamine signals must reflect the functional specialization of underlying subregions...right??

What we did:
-> we used an imaging preparation that allows imaging DA axons chronically across multiple anatomical scales.

Notably, we could image the global activity of dopamine across 60-80% of dorsal striatum!

What we found:
1) We discovered a novel activity pattern in the DA axons across the dorsal striatum. They are organized into waves. Check out this video 👇🏽

It turns out that these patterns correlate activity in functionally related striatal subregions.

They also delivery transients that have different lead/lag relationships

These wave-like patterns have different "motifs" that get sequenced to make more elaborate, complicated trajectories....

But each motifs differs in the pattern of DA distributed to str regions

WOW!
So what are waves used for by the brain??

We designed a couple of simple tasks that vary instrumental contingency.

Turns out, reward delivery resynchronizes activity into directional waves depending on task requirements.

Also, the spatiotemporal response itself is learned, naive mice get a burst of DA at reward, but the trajectory of spread is initially irregular, and random. When the same mice get trained in task, and learn reward contingencies, waves are super smooth, and reliable.

In paper, we unpack a normative interpretation of DA waves in "spatiotemporal" credit assignment.

The credit depends on the eligibility, or responsibility of underlying striatal subregion in reward prediction.

So, we think waves are a mechanism to vectorize RPEs -reward credit

Share this Scrolly Tale with your friends.

A Scrolly Tale is a new way to read Twitter threads with a more visually immersive experience.
Discover more beautiful Scrolly Tales like this.

Keep scrolling