postdoc at nyu | (episodic) memory and decision making | jonathanicholas.github.io
Jonathan Nicholas
Loading...
We make flexible choices in new situations by knitting together information from separate relevant memories.
But what governs which memories are retrieved and when?
In a new preprint, we captured how people build decision variables from different memories by tracking their gaze on a blank screen.
The network learned to sample only when a memory was worth the cost.
This resource-rational sampling strategy reproduced how people behaved and where they looked across a wide range of measures.
People differed only in one way: they were also influenced by how memories are organized at encoding.
Because retrieval is effortful, we hypothesized that people are resource-rational: they weigh the benefits of sampling a memory with the costs of doing so.
To test this, we trained a recurrent neural network to make the same decisions people did while paying a cost each time it sampled a memory.
So gaze can serve as a readout of moment-to-moment memory recall.
We had people make decisions where each choice required flexibly recombining a different subset of memories.
Because every memory was encoded at its own location, we used their gaze to reveal which memories they recalled, and when.
As people deliberated, they looked toward the locations of decision-relevant memories and these fixations shaped their evolving choice.
But what is the algorithm that people actually use to guide how they sample information from their memories?
Nature research paper: Neural representation of action symbols in primate frontal cortex
go.nature.com/4v1FDED
These results demonstrate that flexible decisions emerge from a resource-rational process in which memories are sampled to construct decision variables on-the-fly.
Thanks to my co-authors @sixingchen.bsky.social and @marcelomattar.bsky.social, who were instrumental in making this project happen!
Retrieval happens covertly so it's been hard to develop a principled account of how memory guides choice.
But the eyes offer a way in: even on a blank screen, our gaze drifts back to where we first saw something.
Below is someone doing this as they recall out loud (items aren't actually visible).
New preprint w/ @fredcallaway.bsky.social! How does the brain decide which computations to run? We combine rational meta-reasoning with a meta-learning algorithm to build a recurrent network that learns to select computations.
www.biorxiv.org/content/10.6...
Pleased to share that this work is now published in TMLR!
openreview.net/forum?id=RuW...
