This framework may reconcile conflicting findings in the memory/inference literature and raises the possibility for individualized structured training interventions to improve memory-guided inference and planning.

These effects are further modulated by the sparsity of neural codes: for example, interleaving promotes integration when memory capacity is high, but even more so when neural codes are less sparse and more distributed.

We find that, in general, blocked learning tends to promote integration when memory capacity is lower, whereas interleaved learning tends to promote integration when memory capacity is higher.

Using neural network simulations, we show the likelihood of integrating related/overlapping information depends on 1) the memory capacity of the individual or system, and 2) coding schemes/biases of those representations (sparse vs. distributed).

a stuffed white cat is sitting on a box with its hands outstretched . ALT: a stuffed white cat is sitting on a box with its hands outstretched .

Previous work on this question had found seemingly conflicting results! Some found that structured (blocked) learning promotes memory integration (Schlichting et al., 2015), while others found that randomized (interleaved) presentation does (Zhou et al., 2023).

We ask a question that has a very nuanced answer: How does the order in which we experience the world shape how we represent it in our minds?

New paper! Out now in Neuropsychologia: “Sparsity and memory constraints interact with training sequence to bias learning of associative maps” with @dalezhou.bsky.social*, @kwcooper.bsky.social, @lovecrabmeat.bsky.social, and @aaronbornstein.bsky.social. authors.elsevier.com/a/1mipv6TBG9...
🧵: