Xianhui He

Discovering cognitive strategies with tiny recurrent neural networks - Nature Modelling biological decision-making with tiny recurrent neural networks enables more accurate predictions of animal choices than classical cognitive models and offers insights into the underlying cog...

Thrilled to see our TinyRNN paper in @nature! We show how tiny RNNs predict choices of individual subjects accurately while staying fully interpretable. This approach can transform how we model cognitive processes in both healthy and disordered decisions. doi.org/10.1038/s415...

@oxexppsy.bsky.social @oxcin.bsky.social @erc.europa.eu @medsci.ox.ac.uk @royalsociety.org #neuroscience #memory #sleep #cognitivescience #research #brainscience

Thanks my supervisor @bstaresina.bsky.social and my coauthors @philippbuchel.bsky.social @simonfsoubeyrand.bsky.social Janina Klingspohr @mskehl.bsky.social for their kind help! More details please check our preprint! www.biorxiv.org/content/10.1... 9/9

In sum: Our research shows that sequence learning reshapes our neural representations to be more predictive. And sleep, especially deep sleep, is crucial for transforming these representations to be more abstract. This process helps us update our internal world model by external experiences. 8/9

FINDING 3: So, what's the driving force behind this transformation? **Sleep**! Specifically, deep sleep (slow-wave sleep). We found that participants who got more slow-wave sleep after learning had stronger and more abstract successor representations. 🧠💤 7/9

FINDING 2: Even more interesting, using RSA with a deep neural network, we found the successor representation wasn't just a faint copy of the original image. It became more abstract and "high-level," shifting from simple visual features to the core concept of the image after learning. 6/9

FINDING 1: It worked! We found that even when the sequence was no longer relevant for the task at hand, when participants saw image A, we could decode the information for the successor image B from their brain activity. This confirms the existence of successor representations. 5/9

To find out, we designed an experiment: participants first learned image sequences. We then recorded their brain activity using high-density EEG, including throughout a 2-hr nap, to see if their brain would spontaneously activate a representation of the next image. 4/9

Our study focused on three key questions:

1. After learning a sequence (e.g., A→B→C), does our brain automatically anticipate the successor (B)?

2. Is this prediction based on concrete visual details or more abstract concepts?

3. What role does sleep play in this process? 3/9

There's a cool theory behind this called "Successor Representation." It suggests our brain doesn't just process the "now" but also maintains a "predictive map," anticipating what comes next based on past experiences. We wanted to understand how this map is drawn and updated. 2/9

How does our brain learn that thunder follows lightning? We don't just remember two separate events; we build a predictive model to anticipate the world. My research dives into this very question: how we learn and predict the order of events. 🧵👇 1/9 #neuroscience #memory #sleep

What a great start to #CNS2025!

@manqisha.bsky.social kicked of the first session with her Data Blitz on ‘Coupled sleep rhythms in the human hippocampus support memory consolidation’

The Staresina Lab is ready for CNS 2025 in Boston!🇺🇸

If you are around come and check out our work at the following poster sessions:

Poster Session C
Poster C16: Coupled sleep rhythms in the human hippocampus support memory consolidation - Sha, Manqi