Enjoying a flying visit to Boston. Strange to think that last time I was here was back in 2017, on the way to my first GRC (Catecholamines). I remember my PhD advisor’s sage advice when choosing between conferences: SfN is fun, but go to a GRC to really make connections with people in the field.

Important topics being discussed on the #cosyne2025 Whova app

Looks like Janelia is on blooskee @hhmijanelia.bsky.social! Now that we have an official Janelia account on here, here’s a particularly autumnal photo taken from the roof of the Landscape Building.

Chatting about language in Bob’s last night, I was reminded of how much I enjoyed translating from Latin Pliny the Younger’s account of his uncle’s heroism during the Mt Vesuvius eruption back at school. His vivid description of how people tied pillows to their heads - 2000 years ago - still gets me

This is a photo of my current bedside table…Most from a book club I co-organise here at Janelia.

Who else is going to this #foraging conference in Lyon in December? Any locals have recommendations for whilst we’re there?

And NAcC dopamine changed as it should as a dynamic learning rate applied to all options.

The RL model still did a great job of looking like mouse behaviour in both rate of learning and choices made.

To really put these ideas to the test, we pulled a fast one on the mice and switched up the quality of the options across days...

In fact, if we replaced the value of the learning rate term with NAc DA in simulations, the model did just as well at replicating mouse behaviour.

Whilst mice foraged, we also recorded dopamine in the striatum with photometry. Dopamine in the NAc core - but not the DMS - looked just like that global, dynamic learning rate from our RL model.

This model really looked like mouse behaviour in a lot of ways, so we felt confident that we'd captured the core processes driving their decision-making and learning in FoMO.

To gain insight into how mice were able to learn so rapidly, we designed a new RL model. Two key features made this model successful in replicating mouse behavior: framing the decision as 'stay or leave – and if leave, where next?’, and a dynamic learning rate across all options.

In fact, this is actually the rational solution given the setup of the FoMO environment, meaning mice were near-optimal in their choices.

Within tens of minutes, mice showed rapid learning. And by the end of the first session, they were even matching their choices to the history of reward across all six options.

So we developed a new task and environment where naive, untrained mice were free to Forage across Many (six) Options of differing quality: FoMO.

The first preprint of my postdoc with @dudman.bsky.social is now online! Here we wanted to understand the algorithms mice use when learning to forage in large environments with many potential resources. www.biorxiv.org/content/10.1...

Sunbeans.

Shameless self-promotion but really looking forward to talking about my work on learning, foraging, and dopamine at the SWC in January!