Cross-species connectome comparisons reveal the network attributes of memory capacity and time series prediction
Qian, Y.; Yan, Y.; Wu, G.; She, Q.; Yu, H.
Show abstract
The brains connectome provides a powerful blueprint for designing efficient neural networks, yet the impact of incorporating its intricate, non-random architecture into machine learning models remains underexplored. Here, we integrate empirical structural connectomes from four model species--fruit fly, mouse, rat, and macaque--as the recurrent layer in echo state networks (ESNs). We demonstrate that biologically realized networks, particularly the macaque connectome, achieve superior performance in chaotic time series prediction and exhibit higher memory capacity compared to randomly shuffled controls. This computational advantage correlates with small-world topology, which scales with phylogenetic level. Crucially, we identify that weakly connected but highly central nodes are essential for optimal network dynamics; their targeted perturbation significantly degrades performance. Furthermore, functional connectomes from Alzheimers disease patients show computational deficits resembling those induced by weak-tie disruption in healthy networks. Our findings establish that evolved connectome topology is fundamental to efficient information processing, providing key principles for bio-inspired artificial intelligence.
Matching journals
The top 6 journals account for 50% of the predicted probability mass.
Similar papers in this journal
Similar papers in this journal
- Spatially-embedded recurrent neural networks reveal widespread links between structural and functional neuroscience findings 96%
- Accurate and efficient time-domain classification with adaptive spiking recurrent neural networks 95%
- Recurrent neural networks learn robust representations by dynamically balancing compression and expansion 94%
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.