Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster D in Poster Session D: Wednesday, August 5, 2:00 – 3:45 pm, Kimmel Center, Shorin & Rosenthal Rooms

From Single Neurons to Networks: Fundamental Trade-Offs in Information Processing Systems with Limited Resources

Dan Hilman Amir¹, Yuval Hart¹; ¹Hebrew University of Jerusalem

Presenter: Dan Hilman Amir

The human brain operates under resource constraints, requiring trade-offs between competing goals. Recent work suggests that large-scale cortical organization spans a Pareto front balancing energetic efficiency, external responsiveness, and internally directed processing and memory. However, the origin of these specific trade-offs remains unclear. Here, we investigate their emergence in neural systems across scales. Recurrent neural networks trained under a triadic objective converge to a low-dimensional manifold with interpretable axes. We show that this structure is captured by input-driven Hopfield networks, where recurrent and feedforward coupling define a Pareto-optimal trade-off between memory and responsiveness. Finally, we demonstrate that the same trade-off arises in a single nonlinear unit, reflecting low-dimensional dynamics balancing internal stability and external drive. These results suggest that memory, responsiveness, and efficiency reflect general constraints on neural computation, providing a mechanistic account of Pareto-optimal organization in brain networks.

Topic Area: Methods, Tools, Theory & Neural Coding