Reusable modular architecture enables flexible cognitive operations
Osako, Y.; Buschman, T. J.; Sur, M.
Show abstract
Complex behaviors are thought to be built by combining simpler cognitive components. Computational modeling has shown that artificial neural networks can perform a variety of tasks by flexibly combining small functional modules of neurons, each specialized for a specific computation, to construct a complex task. However, empirical evidence for such reusable modular networks in the brain has been lacking. Neural recordings in mice performing a delayed match-to-sample with delayed report (DMS-dr) task identified subspaces of neural activity that were specialized for stimulus processing and memory maintenance. These subspaces were reused during the task to represent new stimulus inputs and different types of memories, respectively. Clustering analyses showed each subspace was supported by a functionally distinct cluster of neurons in medial prefrontal cortex (mPFC) and posterior parietal cortex (PPC). Recurrent neural networks trained to capture the observed neural dynamics demonstrated that silencing specific clusters disrupted specific computations, highlighting the modular and reusable organization of these networks. By bridging theoretical predictions with empirical evidence, our findings suggest that the brain can flexibly reuse computational components to perform a complex cognitive task.
Matching journals
The top 4 journals account for 50% of the predicted probability mass.
Similar papers in this journal
Similar papers in this journal
- A learning-evoked slow-oscillatory architecture paces population activity for offline reactivation across the human medial temporal lobe 95%
- Inhibitory and excitatory populations in parietal cortex are equally selective for decision outcome in both novices and experts 95%
- Neural state space alignment for magnitude generalization in humans and recurrent networks 94%
Similar papers in this journal
"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.